feat(ui): 优化密钥添加和仪表盘空状态体验

- KeyFormDialog: 添加模式下保存后不关闭对话框，清除表单以便连续添加 - KeyFormDialog: 按钮文案根据编辑/添加模式动态显示 - Dashboard: 优化统计卡片加载状态和空数据占位显示
fix: 修复普通用户无法访问仪表盘接口的权限问题
2026-01-11 12:08:30 +08:00 · 2026-01-10 19:32:36 +08:00 · 2026-01-10 19:31:19 +08:00 · 2026-01-10 18:48:35 +08:00 · 2026-01-08 13:37:41 +08:00 · 2026-01-08 13:37:13 +08:00
102 changed files with 7077 additions and 4255 deletions
--- a/alembic/versions/20260110_2000_consolidated_schema_updates.py
+++ b/alembic/versions/20260110_2000_consolidated_schema_updates.py
@@ -0,0 +1,530 @@
+"""consolidated schema updates
+
+Revision ID: m4n5o6p7q8r9
+Revises: 02a45b66b7c4
+Create Date: 2026-01-10 20:00:00.000000
+
+This migration consolidates all schema changes from 2026-01-08 to 2026-01-10:
+
+1. provider_api_keys: Key 直接关联 Provider (provider_id, api_formats)
+2. provider_api_keys: 添加 rate_multipliers JSON 字段（按格式费率）
+3. models: global_model_id 改为可空（支持独立 ProviderModel）
+4. providers: 添加 timeout, max_retries, proxy（从 endpoint 迁移）
+5. providers: display_name 重命名为 name，删除原 name
+6. provider_api_keys: max_concurrent -> rpm_limit（并发改 RPM）
+7. provider_api_keys: 健康度改为按格式存储（health_by_format, circuit_breaker_by_format）
+8. provider_endpoints: 删除废弃的 rate_limit 列
+9. usage: 添加 client_response_headers 字段
+10. provider_api_keys: 删除 endpoint_id（Key 不再与 Endpoint 绑定）
+11. provider_endpoints: 删除废弃的 max_concurrent 列
+12. providers: 删除废弃的 rpm_limit, rpm_used, rpm_reset_at 列
+"""
+
+import logging
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+from sqlalchemy import inspect
+
+
+# 配置日志
+alembic_logger = logging.getLogger("alembic.runtime.migration")
+
+revision = "m4n5o6p7q8r9"
+down_revision = "02a45b66b7c4"
+branch_labels = None
+depends_on = None
+
+
+def _column_exists(table_name: str, column_name: str) -> bool:
+    """Check if a column exists in the table"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    columns = [col["name"] for col in inspector.get_columns(table_name)]
+    return column_name in columns
+
+
+def _constraint_exists(table_name: str, constraint_name: str) -> bool:
+    """Check if a constraint exists"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    fks = inspector.get_foreign_keys(table_name)
+    return any(fk.get("name") == constraint_name for fk in fks)
+
+
+def _index_exists(table_name: str, index_name: str) -> bool:
+    """Check if an index exists"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    indexes = inspector.get_indexes(table_name)
+    return any(idx.get("name") == index_name for idx in indexes)
+
+
+def upgrade() -> None:
+    """Apply all consolidated schema changes"""
+    bind = op.get_bind()
+
+    # ========== 1. provider_api_keys: 添加 provider_id 和 api_formats ==========
+    if not _column_exists("provider_api_keys", "provider_id"):
+        op.add_column("provider_api_keys", sa.Column("provider_id", sa.String(36), nullable=True))
+
+        # 数据迁移：从 endpoint 获取 provider_id
+        op.execute("""
+            UPDATE provider_api_keys k
+            SET provider_id = e.provider_id
+            FROM provider_endpoints e
+            WHERE k.endpoint_id = e.id AND k.provider_id IS NULL
+        """)
+
+        # 检查无法关联的孤儿 Key
+        result = bind.execute(sa.text(
+            "SELECT COUNT(*) FROM provider_api_keys WHERE provider_id IS NULL"
+        ))
+        orphan_count = result.scalar() or 0
+        if orphan_count > 0:
+            # 使用 logger 记录更明显的告警
+            alembic_logger.warning("=" * 60)
+            alembic_logger.warning(f"[MIGRATION WARNING] 发现 {orphan_count} 个无法关联 Provider 的孤儿 Key")
+            alembic_logger.warning("=" * 60)
+            alembic_logger.info("正在备份孤儿 Key 到 _orphan_api_keys_backup 表...")
+
+            # 先备份孤儿数据到临时表，避免数据丢失
+            op.execute("""
+                CREATE TABLE IF NOT EXISTS _orphan_api_keys_backup AS
+                SELECT *, NOW() as backup_at
+                FROM provider_api_keys
+                WHERE provider_id IS NULL
+            """)
+
+            # 记录备份的 Key ID
+            orphan_ids = bind.execute(sa.text(
+                "SELECT id, name FROM provider_api_keys WHERE provider_id IS NULL"
+            )).fetchall()
+            alembic_logger.info("备份的孤儿 Key 列表：")
+            for key_id, key_name in orphan_ids:
+                alembic_logger.info(f"  - Key: {key_name} (ID: {key_id})")
+
+            # 删除孤儿数据
+            op.execute("DELETE FROM provider_api_keys WHERE provider_id IS NULL")
+            alembic_logger.info(f"已备份并删除 {orphan_count} 个孤儿 Key")
+
+            # 提供恢复指南
+            alembic_logger.warning("-" * 60)
+            alembic_logger.warning("[恢复指南] 如需恢复孤儿 Key：")
+            alembic_logger.warning("  1. 查询备份表: SELECT * FROM _orphan_api_keys_backup;")
+            alembic_logger.warning("  2. 确定正确的 provider_id")
+            alembic_logger.warning("  3. 执行恢复:")
+            alembic_logger.warning("     INSERT INTO provider_api_keys (...)")
+            alembic_logger.warning("     SELECT ... FROM _orphan_api_keys_backup WHERE ...;")
+            alembic_logger.warning("-" * 60)
+
+        # 设置 NOT NULL 并创建外键
+        op.alter_column("provider_api_keys", "provider_id", nullable=False)
+
+        if not _constraint_exists("provider_api_keys", "fk_provider_api_keys_provider"):
+            op.create_foreign_key(
+                "fk_provider_api_keys_provider",
+                "provider_api_keys",
+                "providers",
+                ["provider_id"],
+                ["id"],
+                ondelete="CASCADE",
+            )
+
+        if not _index_exists("provider_api_keys", "idx_provider_api_keys_provider_id"):
+            op.create_index("idx_provider_api_keys_provider_id", "provider_api_keys", ["provider_id"])
+
+    if not _column_exists("provider_api_keys", "api_formats"):
+        op.add_column("provider_api_keys", sa.Column("api_formats", sa.JSON(), nullable=True))
+
+        # 数据迁移：从 endpoint 获取 api_format
+        op.execute("""
+            UPDATE provider_api_keys k
+            SET api_formats = json_build_array(e.api_format)
+            FROM provider_endpoints e
+            WHERE k.endpoint_id = e.id AND k.api_formats IS NULL
+        """)
+
+        op.alter_column("provider_api_keys", "api_formats", nullable=False, server_default="[]")
+
+    # 修改 endpoint_id 为可空，外键改为 SET NULL
+    if _constraint_exists("provider_api_keys", "provider_api_keys_endpoint_id_fkey"):
+        op.drop_constraint("provider_api_keys_endpoint_id_fkey", "provider_api_keys", type_="foreignkey")
+        op.alter_column("provider_api_keys", "endpoint_id", nullable=True)
+        # 不再重建外键，因为后面会删除这个字段
+
+    # ========== 2. provider_api_keys: 添加 rate_multipliers ==========
+    if not _column_exists("provider_api_keys", "rate_multipliers"):
+        op.add_column(
+            "provider_api_keys",
+            sa.Column("rate_multipliers", postgresql.JSON(astext_type=sa.Text()), nullable=True),
+        )
+
+        # 数据迁移：将 rate_multiplier 按 api_formats 转换
+        op.execute("""
+            UPDATE provider_api_keys
+            SET rate_multipliers = (
+                SELECT jsonb_object_agg(elem, rate_multiplier)
+                FROM jsonb_array_elements_text(api_formats::jsonb) AS elem
+            )
+            WHERE api_formats IS NOT NULL
+              AND api_formats::text != '[]'
+              AND api_formats::text != 'null'
+              AND rate_multipliers IS NULL
+        """)
+
+    # ========== 3. models: global_model_id 改为可空 ==========
+    op.alter_column("models", "global_model_id", existing_type=sa.String(36), nullable=True)
+
+    # ========== 4. providers: 添加 timeout, max_retries, proxy ==========
+    if not _column_exists("providers", "timeout"):
+        op.add_column(
+            "providers",
+            sa.Column("timeout", sa.Integer(), nullable=True, comment="请求超时（秒）"),
+        )
+
+    if not _column_exists("providers", "max_retries"):
+        op.add_column(
+            "providers",
+            sa.Column("max_retries", sa.Integer(), nullable=True, comment="最大重试次数"),
+        )
+
+    if not _column_exists("providers", "proxy"):
+        op.add_column(
+            "providers",
+            sa.Column("proxy", postgresql.JSONB(), nullable=True, comment="代理配置"),
+        )
+
+    # 从端点迁移数据到 provider
+    op.execute("""
+        UPDATE providers p
+        SET
+            timeout = COALESCE(
+                p.timeout,
+                (SELECT MAX(e.timeout) FROM provider_endpoints e WHERE e.provider_id = p.id AND e.timeout IS NOT NULL),
+                300
+            ),
+            max_retries = COALESCE(
+                p.max_retries,
+                (SELECT MAX(e.max_retries) FROM provider_endpoints e WHERE e.provider_id = p.id AND e.max_retries IS NOT NULL),
+                2
+            ),
+            proxy = COALESCE(
+                p.proxy,
+                (SELECT e.proxy FROM provider_endpoints e WHERE e.provider_id = p.id AND e.proxy IS NOT NULL ORDER BY e.created_at LIMIT 1)
+            )
+        WHERE p.timeout IS NULL OR p.max_retries IS NULL
+    """)
+
+    # ========== 5. providers: display_name -> name ==========
+    # 注意：这里假设 display_name 已经被重命名为 name
+    # 如果 display_name 仍然存在，则需要执行重命名
+    if _column_exists("providers", "display_name"):
+        # 删除旧的 name 索引
+        if _index_exists("providers", "ix_providers_name"):
+            op.drop_index("ix_providers_name", table_name="providers")
+
+        # 如果存在旧的 name 列，先删除
+        if _column_exists("providers", "name"):
+            op.drop_column("providers", "name")
+
+        # 重命名 display_name 为 name
+        op.alter_column("providers", "display_name", new_column_name="name")
+
+        # 创建新索引
+        op.create_index("ix_providers_name", "providers", ["name"], unique=True)
+
+    # ========== 6. provider_api_keys: max_concurrent -> rpm_limit ==========
+    if _column_exists("provider_api_keys", "max_concurrent"):
+        op.alter_column("provider_api_keys", "max_concurrent", new_column_name="rpm_limit")
+
+    if _column_exists("provider_api_keys", "learned_max_concurrent"):
+        op.alter_column("provider_api_keys", "learned_max_concurrent", new_column_name="learned_rpm_limit")
+
+    if _column_exists("provider_api_keys", "last_concurrent_peak"):
+        op.alter_column("provider_api_keys", "last_concurrent_peak", new_column_name="last_rpm_peak")
+
+    # 删除废弃字段
+    for col in ["rate_limit", "daily_limit", "monthly_limit"]:
+        if _column_exists("provider_api_keys", col):
+            op.drop_column("provider_api_keys", col)
+
+    # ========== 7. provider_api_keys: 健康度改为按格式存储 ==========
+    if not _column_exists("provider_api_keys", "health_by_format"):
+        op.add_column(
+            "provider_api_keys",
+            sa.Column(
+                "health_by_format",
+                postgresql.JSONB(astext_type=sa.Text()),
+                nullable=True,
+                comment="按API格式存储的健康度数据",
+            ),
+        )
+
+    if not _column_exists("provider_api_keys", "circuit_breaker_by_format"):
+        op.add_column(
+            "provider_api_keys",
+            sa.Column(
+                "circuit_breaker_by_format",
+                postgresql.JSONB(astext_type=sa.Text()),
+                nullable=True,
+                comment="按API格式存储的熔断器状态",
+            ),
+        )
+
+    # 数据迁移：如果存在旧字段，迁移数据到新结构
+    if _column_exists("provider_api_keys", "health_score"):
+        op.execute("""
+            UPDATE provider_api_keys
+            SET health_by_format = (
+                SELECT jsonb_object_agg(
+                    elem,
+                    jsonb_build_object(
+                        'health_score', COALESCE(health_score, 1.0),
+                        'consecutive_failures', COALESCE(consecutive_failures, 0),
+                        'last_failure_at', last_failure_at,
+                        'request_results_window', COALESCE(request_results_window::jsonb, '[]'::jsonb)
+                    )
+                )
+                FROM jsonb_array_elements_text(api_formats::jsonb) AS elem
+            )
+            WHERE api_formats IS NOT NULL
+              AND api_formats::text != '[]'
+              AND health_by_format IS NULL
+        """)
+
+    # Circuit Breaker 迁移策略：
+    # 不复制旧的 circuit_breaker_open 状态到所有 format，而是全部重置为 closed
+    # 原因：旧的单一 circuit breaker 状态可能因某一个 format 失败而打开，
+    #       如果复制到所有 format，会导致其他正常工作的 format 被错误标记为不可用
+    if _column_exists("provider_api_keys", "circuit_breaker_open"):
+        op.execute("""
+            UPDATE provider_api_keys
+            SET circuit_breaker_by_format = (
+                SELECT jsonb_object_agg(
+                    elem,
+                    jsonb_build_object(
+                        'open', false,
+                        'open_at', NULL,
+                        'next_probe_at', NULL,
+                        'half_open_until', NULL,
+                        'half_open_successes', 0,
+                        'half_open_failures', 0
+                    )
+                )
+                FROM jsonb_array_elements_text(api_formats::jsonb) AS elem
+            )
+            WHERE api_formats IS NOT NULL
+              AND api_formats::text != '[]'
+              AND circuit_breaker_by_format IS NULL
+        """)
+
+    # 设置默认空对象
+    op.execute("""
+        UPDATE provider_api_keys
+        SET health_by_format = '{}'::jsonb
+        WHERE health_by_format IS NULL
+    """)
+    op.execute("""
+        UPDATE provider_api_keys
+        SET circuit_breaker_by_format = '{}'::jsonb
+        WHERE circuit_breaker_by_format IS NULL
+    """)
+
+    # 创建 GIN 索引
+    if not _index_exists("provider_api_keys", "ix_provider_api_keys_health_by_format"):
+        op.create_index(
+            "ix_provider_api_keys_health_by_format",
+            "provider_api_keys",
+            ["health_by_format"],
+            postgresql_using="gin",
+        )
+    if not _index_exists("provider_api_keys", "ix_provider_api_keys_circuit_breaker_by_format"):
+        op.create_index(
+            "ix_provider_api_keys_circuit_breaker_by_format",
+            "provider_api_keys",
+            ["circuit_breaker_by_format"],
+            postgresql_using="gin",
+        )
+
+    # 删除旧字段
+    old_health_columns = [
+        "health_score",
+        "consecutive_failures",
+        "last_failure_at",
+        "request_results_window",
+        "circuit_breaker_open",
+        "circuit_breaker_open_at",
+        "next_probe_at",
+        "half_open_until",
+        "half_open_successes",
+        "half_open_failures",
+    ]
+    for col in old_health_columns:
+        if _column_exists("provider_api_keys", col):
+            op.drop_column("provider_api_keys", col)
+
+    # ========== 8. provider_endpoints: 删除废弃的 rate_limit 列 ==========
+    if _column_exists("provider_endpoints", "rate_limit"):
+        op.drop_column("provider_endpoints", "rate_limit")
+
+    # ========== 9. usage: 添加 client_response_headers ==========
+    if not _column_exists("usage", "client_response_headers"):
+        op.add_column(
+            "usage",
+            sa.Column("client_response_headers", sa.JSON(), nullable=True),
+        )
+
+    # ========== 10. provider_api_keys: 删除 endpoint_id ==========
+    # Key 不再与 Endpoint 绑定，通过 provider_id + api_formats 关联
+    if _column_exists("provider_api_keys", "endpoint_id"):
+        # 确保外键已删除（前面可能已经删除）
+        try:
+            bind = op.get_bind()
+            inspector = inspect(bind)
+            for fk in inspector.get_foreign_keys("provider_api_keys"):
+                constrained = fk.get("constrained_columns") or []
+                if "endpoint_id" in constrained:
+                    name = fk.get("name")
+                    if name:
+                        op.drop_constraint(name, "provider_api_keys", type_="foreignkey")
+        except Exception:
+            pass  # 外键可能已经不存在
+        op.drop_column("provider_api_keys", "endpoint_id")
+
+    # ========== 11. provider_endpoints: 删除废弃的 max_concurrent 列 ==========
+    if _column_exists("provider_endpoints", "max_concurrent"):
+        op.drop_column("provider_endpoints", "max_concurrent")
+
+    # ========== 12. providers: 删除废弃的 RPM 相关字段 ==========
+    if _column_exists("providers", "rpm_limit"):
+        op.drop_column("providers", "rpm_limit")
+    if _column_exists("providers", "rpm_used"):
+        op.drop_column("providers", "rpm_used")
+    if _column_exists("providers", "rpm_reset_at"):
+        op.drop_column("providers", "rpm_reset_at")
+
+    alembic_logger.info("[OK] Consolidated migration completed successfully")
+
+
+def downgrade() -> None:
+    """
+    Downgrade is complex due to data migrations.
+    For safety, this only removes new columns without restoring old structure.
+    Manual intervention may be required for full rollback.
+    """
+    bind = op.get_bind()
+
+    # 12. 恢复 providers RPM 相关字段
+    if not _column_exists("providers", "rpm_limit"):
+        op.add_column("providers", sa.Column("rpm_limit", sa.Integer(), nullable=True))
+    if not _column_exists("providers", "rpm_used"):
+        op.add_column(
+            "providers",
+            sa.Column("rpm_used", sa.Integer(), server_default="0", nullable=True),
+        )
+    if not _column_exists("providers", "rpm_reset_at"):
+        op.add_column(
+            "providers",
+            sa.Column("rpm_reset_at", sa.DateTime(timezone=True), nullable=True),
+        )
+
+    # 11. 恢复 provider_endpoints.max_concurrent
+    if not _column_exists("provider_endpoints", "max_concurrent"):
+        op.add_column("provider_endpoints", sa.Column("max_concurrent", sa.Integer(), nullable=True))
+
+    # 10. 恢复 endpoint_id
+    if not _column_exists("provider_api_keys", "endpoint_id"):
+        op.add_column("provider_api_keys", sa.Column("endpoint_id", sa.String(36), nullable=True))
+
+    # 9. 删除 client_response_headers
+    if _column_exists("usage", "client_response_headers"):
+        op.drop_column("usage", "client_response_headers")
+
+    # 8. 恢复 provider_endpoints.rate_limit（如果需要）
+    if not _column_exists("provider_endpoints", "rate_limit"):
+        op.add_column("provider_endpoints", sa.Column("rate_limit", sa.Integer(), nullable=True))
+
+    # 7. 删除健康度 JSON 字段
+    bind.execute(sa.text("DROP INDEX IF EXISTS ix_provider_api_keys_health_by_format"))
+    bind.execute(sa.text("DROP INDEX IF EXISTS ix_provider_api_keys_circuit_breaker_by_format"))
+    if _column_exists("provider_api_keys", "health_by_format"):
+        op.drop_column("provider_api_keys", "health_by_format")
+    if _column_exists("provider_api_keys", "circuit_breaker_by_format"):
+        op.drop_column("provider_api_keys", "circuit_breaker_by_format")
+
+    # 6. rpm_limit -> max_concurrent（简化版：仅重命名）
+    if _column_exists("provider_api_keys", "rpm_limit"):
+        op.alter_column("provider_api_keys", "rpm_limit", new_column_name="max_concurrent")
+    if _column_exists("provider_api_keys", "learned_rpm_limit"):
+        op.alter_column("provider_api_keys", "learned_rpm_limit", new_column_name="learned_max_concurrent")
+    if _column_exists("provider_api_keys", "last_rpm_peak"):
+        op.alter_column("provider_api_keys", "last_rpm_peak", new_column_name="last_concurrent_peak")
+
+    # 恢复已删除的字段
+    if not _column_exists("provider_api_keys", "rate_limit"):
+        op.add_column("provider_api_keys", sa.Column("rate_limit", sa.Integer(), nullable=True))
+    if not _column_exists("provider_api_keys", "daily_limit"):
+        op.add_column("provider_api_keys", sa.Column("daily_limit", sa.Integer(), nullable=True))
+    if not _column_exists("provider_api_keys", "monthly_limit"):
+        op.add_column("provider_api_keys", sa.Column("monthly_limit", sa.Integer(), nullable=True))
+
+    # 5. name -> display_name (需要先删除索引)
+    if _index_exists("providers", "ix_providers_name"):
+        op.drop_index("ix_providers_name", table_name="providers")
+    op.alter_column("providers", "name", new_column_name="display_name")
+    # 重新添加原 name 字段
+    op.add_column("providers", sa.Column("name", sa.String(100), nullable=True))
+    op.execute("""
+        UPDATE providers
+        SET name = LOWER(REPLACE(REPLACE(display_name, ' ', '_'), '-', '_'))
+    """)
+    op.alter_column("providers", "name", nullable=False)
+    op.create_index("ix_providers_name", "providers", ["name"], unique=True)
+
+    # 4. 删除 providers 的 timeout, max_retries, proxy
+    if _column_exists("providers", "proxy"):
+        op.drop_column("providers", "proxy")
+    if _column_exists("providers", "max_retries"):
+        op.drop_column("providers", "max_retries")
+    if _column_exists("providers", "timeout"):
+        op.drop_column("providers", "timeout")
+
+    # 3. models: global_model_id 改回 NOT NULL
+    result = bind.execute(sa.text(
+        "SELECT COUNT(*) FROM models WHERE global_model_id IS NULL"
+    ))
+    orphan_model_count = result.scalar() or 0
+    if orphan_model_count > 0:
+        alembic_logger.warning(f"[WARN] 发现 {orphan_model_count} 个无 global_model_id 的独立模型，将被删除")
+        op.execute("DELETE FROM models WHERE global_model_id IS NULL")
+        alembic_logger.info(f"已删除 {orphan_model_count} 个独立模型")
+    op.alter_column("models", "global_model_id", nullable=False)
+
+    # 2. 删除 rate_multipliers
+    if _column_exists("provider_api_keys", "rate_multipliers"):
+        op.drop_column("provider_api_keys", "rate_multipliers")
+
+    # 1. 删除 provider_id 和 api_formats
+    if _index_exists("provider_api_keys", "idx_provider_api_keys_provider_id"):
+        op.drop_index("idx_provider_api_keys_provider_id", table_name="provider_api_keys")
+    if _constraint_exists("provider_api_keys", "fk_provider_api_keys_provider"):
+        op.drop_constraint("fk_provider_api_keys_provider", "provider_api_keys", type_="foreignkey")
+    if _column_exists("provider_api_keys", "api_formats"):
+        op.drop_column("provider_api_keys", "api_formats")
+    if _column_exists("provider_api_keys", "provider_id"):
+        op.drop_column("provider_api_keys", "provider_id")
+
+    # 恢复 endpoint_id 外键（简化版：仅创建外键，不强制 NOT NULL）
+    if _column_exists("provider_api_keys", "endpoint_id"):
+        if not _constraint_exists("provider_api_keys", "provider_api_keys_endpoint_id_fkey"):
+            op.create_foreign_key(
+                "provider_api_keys_endpoint_id_fkey",
+                "provider_api_keys",
+                "provider_endpoints",
+                ["endpoint_id"],
+                ["id"],
+                ondelete="SET NULL",
+            )
+
+    alembic_logger.info("[OK] Downgrade completed (simplified version)")
--- a/frontend/src/api/admin.ts
+++ b/frontend/src/api/admin.ts
@@ -67,7 +67,6 @@ export interface GlobalModelExport {

 export interface ProviderExport {
  name: string
-  display_name: string
  description?: string | null
  website?: string | null
  billing_type?: string | null
@@ -76,10 +75,13 @@ export interface ProviderExport {
  rpm_limit?: number | null
  provider_priority?: number
  is_active: boolean
-  rate_limit?: number | null
  concurrent_limit?: number | null
+  timeout?: number | null
+  max_retries?: number | null
+  proxy?: any
  config?: any
  endpoints: EndpointExport[]
+  api_keys: ProviderKeyExport[]
  models: ModelExport[]
 }

@@ -89,27 +91,26 @@ export interface EndpointExport {
  headers?: any
  timeout?: number
  max_retries?: number
-  max_concurrent?: number | null
-  rate_limit?: number | null
  is_active: boolean
  custom_path?: string | null
  config?: any
-  keys: KeyExport[]
+  proxy?: any
 }

-export interface KeyExport {
+export interface ProviderKeyExport {
  api_key: string
  name?: string | null
  note?: string | null
+  api_formats: string[]
  rate_multiplier?: number
+  rate_multipliers?: Record<string, number> | null
  internal_priority?: number
  global_priority?: number | null
-  max_concurrent?: number | null
-  rate_limit?: number | null
-  daily_limit?: number | null
-  monthly_limit?: number | null
-  allowed_models?: string[] | null
+  rpm_limit?: number | null
+  allowed_models?: any
  capabilities?: any
+  cache_ttl_minutes?: number
+  max_probe_interval_minutes?: number
  is_active: boolean
 }

--- a/frontend/src/api/dashboard.ts
+++ b/frontend/src/api/dashboard.ts
@@ -155,6 +155,7 @@ export interface RequestDetail {
  request_body?: Record<string, any>
  provider_request_headers?: Record<string, any>
  response_headers?: Record<string, any>
+  client_response_headers?: Record<string, any>
  response_body?: Record<string, any>
  metadata?: Record<string, any>
  // 阶梯计费信息
--- a/frontend/src/api/endpoints/adaptive.ts
+++ b/frontend/src/api/endpoints/adaptive.ts
@@ -14,7 +14,7 @@ export async function toggleAdaptiveMode(
  message: string
  key_id: string
  is_adaptive: boolean
-  max_concurrent: number | null
+  rpm_limit: number | null
  effective_limit: number | null
 }> {
  const response = await client.patch(`/api/admin/adaptive/keys/${keyId}/mode`, data)
@@ -22,16 +22,16 @@ export async function toggleAdaptiveMode(
 }

 /**
- * 设置 Key 的固定并发限制
+ * 设置 Key 的固定 RPM 限制
 */
-export async function setConcurrentLimit(
+export async function setRpmLimit(
  keyId: string,
  limit: number
 ): Promise<{
  message: string
  key_id: string
  is_adaptive: boolean
-  max_concurrent: number
+  rpm_limit: number
  previous_mode: string
 }> {
  const response = await client.patch(`/api/admin/adaptive/keys/${keyId}/limit`, null, {
--- a/frontend/src/api/endpoints/endpoints.ts
+++ b/frontend/src/api/endpoints/endpoints.ts
@@ -27,15 +27,9 @@ export async function createEndpoint(
    api_format: string
    base_url: string
    custom_path?: string
-    auth_type?: string
-    auth_header?: string
    headers?: Record<string, string>
    timeout?: number
    max_retries?: number
-    priority?: number
-    weight?: number
-    max_concurrent?: number
-    rate_limit?: number
    is_active?: boolean
    config?: Record<string, any>
    proxy?: ProxyConfig | null
@@ -52,16 +46,10 @@ export async function updateEndpoint(
  endpointId: string,
  data: Partial<{
    base_url: string
-    custom_path: string
-    auth_type: string
-    auth_header: string
+    custom_path: string | null
    headers: Record<string, string>
    timeout: number
    max_retries: number
-    priority: number
-    weight: number
-    max_concurrent: number
-    rate_limit: number
    is_active: boolean
    config: Record<string, any>
    proxy: ProxyConfig | null
@@ -74,7 +62,7 @@ export async function updateEndpoint(
 /**
 * 删除 Endpoint
 */
-export async function deleteEndpoint(endpointId: string): Promise<{ message: string; deleted_keys_count: number }> {
+export async function deleteEndpoint(endpointId: string): Promise<{ message: string; affected_keys_count: number }> {
  const response = await client.delete(`/api/admin/endpoints/${endpointId}`)
  return response.data
 }
--- a/frontend/src/api/endpoints/health.ts
+++ b/frontend/src/api/endpoints/health.ts
@@ -32,16 +32,21 @@ export async function getKeyHealth(keyId: string): Promise<HealthStatus> {

 /**
 * 恢复Key健康状态（一键恢复：重置健康度 + 关闭熔断器 + 取消自动禁用）
+ * @param keyId Key ID
+ * @param apiFormat 可选，指定 API 格式（如 CLAUDE、OPENAI），不指定则恢复所有格式
 */
-export async function recoverKeyHealth(keyId: string): Promise<{
+export async function recoverKeyHealth(keyId: string, apiFormat?: string): Promise<{
  message: string
  details: {
+    api_format?: string
    health_score: number
    circuit_breaker_open: boolean
    is_active: boolean
  }
 }> {
-  const response = await client.patch(`/api/admin/endpoints/health/keys/${keyId}`)
+  const response = await client.patch(`/api/admin/endpoints/health/keys/${keyId}`, null, {
+    params: apiFormat ? { api_format: apiFormat } : undefined
+  })
  return response.data
 }

--- a/frontend/src/api/endpoints/keys.ts
+++ b/frontend/src/api/endpoints/keys.ts
@@ -1,5 +1,5 @@
 import client from '../client'
-import type { EndpointAPIKey } from './types'
+import type { EndpointAPIKey, AllowedModels } from './types'

 /**
 * 能力定义类型
@@ -49,67 +49,6 @@ export async function getModelCapabilities(modelName: string): Promise<ModelCapa
  return response.data
 }

-/**
- * 获取 Endpoint 的所有 Keys
- */
-export async function getEndpointKeys(endpointId: string): Promise<EndpointAPIKey[]> {
-  const response = await client.get(`/api/admin/endpoints/${endpointId}/keys`)
-  return response.data
-}
-
-/**
- * 为 Endpoint 添加 Key
- */
-export async function addEndpointKey(
-  endpointId: string,
-  data: {
-    endpoint_id: string
-    api_key: string
-    name: string  // 密钥名称（必填）
-    rate_multiplier?: number  // 成本倍率（默认 1.0）
-    internal_priority?: number  // Endpoint 内部优先级（数字越小越优先）
-    max_concurrent?: number  // 最大并发数（留空=自适应模式）
-    rate_limit?: number
-    daily_limit?: number
-    monthly_limit?: number
-    cache_ttl_minutes?: number  // 缓存 TTL（分钟），0=禁用
-    max_probe_interval_minutes?: number  // 熔断探测间隔（分钟）
-    allowed_models?: string[]  // 允许使用的模型列表
-    capabilities?: Record<string, boolean>  // 能力标签配置
-    note?: string  // 备注说明（可选）
-  }
-): Promise<EndpointAPIKey> {
-  const response = await client.post(`/api/admin/endpoints/${endpointId}/keys`, data)
-  return response.data
-}
-
-/**
- * 更新 Endpoint Key
- */
-export async function updateEndpointKey(
-  keyId: string,
-  data: Partial<{
-    api_key: string
-    name: string  // 密钥名称
-    rate_multiplier: number  // 成本倍率
-    internal_priority: number  // Endpoint 内部优先级（提供商优先模式，数字越小越优先）
-    global_priority: number  // 全局 Key 优先级（全局 Key 优先模式，数字越小越优先）
-    max_concurrent: number  // 最大并发数（留空=自适应模式）
-    rate_limit: number
-    daily_limit: number
-    monthly_limit: number
-    cache_ttl_minutes: number  // 缓存 TTL（分钟），0=禁用
-    max_probe_interval_minutes: number  // 熔断探测间隔（分钟）
-    allowed_models: string[] | null  // 允许使用的模型列表，null 表示允许所有
-    capabilities: Record<string, boolean> | null  // 能力标签配置
-    is_active: boolean
-    note: string  // 备注说明
-  }>
-): Promise<EndpointAPIKey> {
-  const response = await client.put(`/api/admin/endpoints/keys/${keyId}`, data)
-  return response.data
-}
-
 /**
 * 获取完整的 API Key（用于查看和复制）
 */
@@ -119,22 +58,71 @@ export async function revealEndpointKey(keyId: string): Promise<{ api_key: strin
 }

 /**
- * 删除 Endpoint Key
+ * 删除 Key
 */
 export async function deleteEndpointKey(keyId: string): Promise<{ message: string }> {
  const response = await client.delete(`/api/admin/endpoints/keys/${keyId}`)
  return response.data
 }

+
+// ========== Provider 级别的 Keys API ==========
+
+
 /**
- * 批量更新 Endpoint Keys 的优先级（用于拖动排序）
+ * 获取 Provider 的所有 Keys
 */
-export async function batchUpdateKeyPriority(
-  endpointId: string,
-  priorities: Array<{ key_id: string; internal_priority: number }>
-): Promise<{ message: string; updated_count: number }> {
-  const response = await client.put(`/api/admin/endpoints/${endpointId}/keys/batch-priority`, {
-    priorities
-  })
+export async function getProviderKeys(providerId: string): Promise<EndpointAPIKey[]> {
+  const response = await client.get(`/api/admin/endpoints/providers/${providerId}/keys`)
+  return response.data
+}
+
+/**
+ * 为 Provider 添加 Key
+ */
+export async function addProviderKey(
+  providerId: string,
+  data: {
+    api_formats: string[]  // 支持的 API 格式列表（必填）
+    api_key: string
+    name: string
+    rate_multiplier?: number  // 默认成本倍率
+    rate_multipliers?: Record<string, number> | null  // 按 API 格式的成本倍率
+    internal_priority?: number
+    rpm_limit?: number | null  // RPM 限制（留空=自适应模式）
+    cache_ttl_minutes?: number
+    max_probe_interval_minutes?: number
+    allowed_models?: AllowedModels
+    capabilities?: Record<string, boolean>
+    note?: string
+  }
+): Promise<EndpointAPIKey> {
+  const response = await client.post(`/api/admin/endpoints/providers/${providerId}/keys`, data)
+  return response.data
+}
+
+/**
+ * 更新 Key
+ */
+export async function updateProviderKey(
+  keyId: string,
+  data: Partial<{
+    api_formats: string[]  // 支持的 API 格式列表
+    api_key: string
+    name: string
+    rate_multiplier: number  // 默认成本倍率
+    rate_multipliers: Record<string, number> | null  // 按 API 格式的成本倍率
+    internal_priority: number
+    global_priority: number | null
+    rpm_limit: number | null  // RPM 限制（留空=自适应模式）
+    cache_ttl_minutes: number
+    max_probe_interval_minutes: number
+    allowed_models: AllowedModels
+    capabilities: Record<string, boolean> | null
+    is_active: boolean
+    note: string
+  }>
+): Promise<EndpointAPIKey> {
+  const response = await client.put(`/api/admin/endpoints/keys/${keyId}`, data)
  return response.data
 }
--- a/frontend/src/api/endpoints/models.ts
+++ b/frontend/src/api/endpoints/models.ts
@@ -147,14 +147,26 @@ export async function queryProviderUpstreamModels(

 /**
 * 从上游提供商导入模型
+ * @param providerId 提供商 ID
+ * @param modelIds 模型 ID 列表
+ * @param options 可选配置
+ * @param options.tiered_pricing 阶梯计费配置
+ * @param options.price_per_request 按次计费价格
 */
 export async function importModelsFromUpstream(
  providerId: string,
-  modelIds: string[]
+  modelIds: string[],
+  options?: {
+    tiered_pricing?: object
+    price_per_request?: number
+  }
 ): Promise<ImportFromUpstreamResponse> {
  const response = await client.post(
    `/api/admin/providers/${providerId}/import-from-upstream`,
-    { model_ids: modelIds }
+    {
+      model_ids: modelIds,
+      ...options
+    }
  )
  return response.data
 }
--- a/frontend/src/api/endpoints/providers.ts
+++ b/frontend/src/api/endpoints/providers.ts
@@ -1,5 +1,5 @@
 import client from '../client'
-import type { ProviderWithEndpointsSummary } from './types'
+import type { ProviderWithEndpointsSummary, ProxyConfig } from './types'

 /**
 * 获取 Providers 摘要（包含 Endpoints 统计）
@@ -23,7 +23,7 @@ export async function getProvider(providerId: string): Promise<ProviderWithEndpo
 export async function updateProvider(
  providerId: string,
  data: Partial<{
-    display_name: string
+    name: string
    description: string
    website: string
    provider_priority: number
@@ -33,6 +33,10 @@ export async function updateProvider(
    quota_last_reset_at: string  // 周期开始时间
    quota_expires_at: string
    rpm_limit: number | null
+    // 请求配置（从 Endpoint 迁移）
+    timeout: number
+    max_retries: number
+    proxy: ProxyConfig | null
    cache_ttl_minutes: number  // 0表示不支持缓存，>0表示支持缓存并设置TTL(分钟)
    max_probe_interval_minutes: number
    is_active: boolean
@@ -83,7 +87,6 @@ export interface TestModelResponse {
  provider?: {
    id: string
    name: string
-    display_name: string
  }
  model?: string
 }
@@ -92,4 +95,3 @@ export async function testModel(data: TestModelRequest): Promise<TestModelRespon
  const response = await client.post('/api/admin/provider-query/test-model', data)
  return response.data
 }
-
--- a/frontend/src/api/endpoints/types.ts
+++ b/frontend/src/api/endpoints/types.ts
@@ -20,6 +20,38 @@ export const API_FORMAT_LABELS: Record<string, string> = {
  [API_FORMATS.GEMINI_CLI]: 'Gemini CLI',
 }

+// API 格式缩写映射（用于空间紧凑的显示场景）
+export const API_FORMAT_SHORT: Record<string, string> = {
+  [API_FORMATS.OPENAI]: 'O',
+  [API_FORMATS.OPENAI_CLI]: 'OC',
+  [API_FORMATS.CLAUDE]: 'C',
+  [API_FORMATS.CLAUDE_CLI]: 'CC',
+  [API_FORMATS.GEMINI]: 'G',
+  [API_FORMATS.GEMINI_CLI]: 'GC',
+}
+
+// API 格式排序顺序（统一的显示顺序）
+export const API_FORMAT_ORDER: string[] = [
+  API_FORMATS.OPENAI,
+  API_FORMATS.OPENAI_CLI,
+  API_FORMATS.CLAUDE,
+  API_FORMATS.CLAUDE_CLI,
+  API_FORMATS.GEMINI,
+  API_FORMATS.GEMINI_CLI,
+]
+
+// 工具函数：按标准顺序排序 API 格式数组
+export function sortApiFormats(formats: string[]): string[] {
+  return [...formats].sort((a, b) => {
+    const aIdx = API_FORMAT_ORDER.indexOf(a)
+    const bIdx = API_FORMAT_ORDER.indexOf(b)
+    if (aIdx === -1 && bIdx === -1) return 0
+    if (aIdx === -1) return 1
+    if (bIdx === -1) return -1
+    return aIdx - bIdx
+  })
+}
+
 /**
 * 代理配置类型
 */
@@ -37,18 +69,9 @@ export interface ProviderEndpoint {
  api_format: string
  base_url: string
  custom_path?: string  // 自定义请求路径（可选，为空则使用 API 格式默认路径）
-  auth_type: string
-  auth_header?: string
  headers?: Record<string, string>
  timeout: number
  max_retries: number
-  priority: number
-  weight: number
-  max_concurrent?: number
-  rate_limit?: number
-  health_score: number
-  consecutive_failures: number
-  last_failure_at?: string
  is_active: boolean
  config?: Record<string, any>
  proxy?: ProxyConfig | null
@@ -58,25 +81,55 @@ export interface ProviderEndpoint {
  updated_at: string
 }

+/**
+ * 模型权限配置类型（支持简单列表和按格式字典两种模式）
+ *
+ * 使用示例：
+ * 1. 不限制（允许所有模型）: null
+ * 2. 简单列表模式（所有 API 格式共享同一个白名单）: ["gpt-4", "claude-3-opus"]
+ * 3. 按格式字典模式（不同 API 格式使用不同的白名单）:
+ *    { "OPENAI": ["gpt-4"], "CLAUDE": ["claude-3-opus"] }
+ */
+export type AllowedModels = string[] | Record<string, string[]> | null
+
+// AllowedModels 类型守卫函数
+export function isAllowedModelsList(value: AllowedModels): value is string[] {
+  return Array.isArray(value)
+}
+
+export function isAllowedModelsDict(value: AllowedModels): value is Record<string, string[]> {
+  if (value === null || typeof value !== 'object' || Array.isArray(value)) {
+    return false
+  }
+  // 验证所有值都是字符串数组
+  return Object.values(value).every(
+    (v) => Array.isArray(v) && v.every((item) => typeof item === 'string')
+  )
+}
+
 export interface EndpointAPIKey {
  id: string
-  endpoint_id: string
+  provider_id: string
+  api_formats: string[]  // 支持的 API 格式列表
  api_key_masked: string
  api_key_plain?: string | null
  name: string  // 密钥名称（必填，用于识别）
-  rate_multiplier: number  // 成本倍率（真实成本 = 表面成本 × 倍率）
-  internal_priority: number  // Endpoint 内部优先级
+  rate_multiplier: number  // 默认成本倍率（真实成本 = 表面成本 × 倍率）
+  rate_multipliers?: Record<string, number> | null  // 按 API 格式的成本倍率，如 {"CLAUDE": 1.0, "OPENAI": 0.8}
+  internal_priority: number  // Key 内部优先级
  global_priority?: number | null  // 全局 Key 优先级
-  max_concurrent?: number
-  rate_limit?: number
-  daily_limit?: number
-  monthly_limit?: number
-  allowed_models?: string[] | null  // 允许使用的模型列表（null = 支持所有模型）
+  rpm_limit?: number | null  // RPM 速率限制 (1-10000)，null 表示自适应模式
+  allowed_models?: AllowedModels  // 允许使用的模型列表（null=不限制，列表=简单白名单，字典=按格式区分）
  capabilities?: Record<string, boolean> | null  // 能力标签配置（如 cache_1h, context_1m）
  // 缓存与熔断配置
  cache_ttl_minutes: number  // 缓存 TTL（分钟），0=禁用
  max_probe_interval_minutes: number  // 熔断探测间隔（分钟）
+  // 按格式的健康度数据
+  health_by_format?: Record<string, FormatHealthData>
+  circuit_breaker_by_format?: Record<string, FormatCircuitBreakerData>
+  // 聚合字段（从 health_by_format 计算，用于列表显示）
  health_score: number
+  circuit_breaker_open?: boolean
  consecutive_failures: number
  last_failure_at?: string
  request_count: number
@@ -89,10 +142,10 @@ export interface EndpointAPIKey {
  last_used_at?: string
  created_at: string
  updated_at: string
-  // 自适应并发字段
-  is_adaptive?: boolean  // 是否为自适应模式（max_concurrent=NULL）
-  effective_limit?: number  // 当前有效限制（自适应使用学习值，固定使用配置值）
-  learned_max_concurrent?: number
+  // 自适应 RPM 字段
+  is_adaptive?: boolean  // 是否为自适应模式（rpm_limit=NULL）
+  effective_limit?: number  // 当前有效 RPM 限制（自适应使用学习值，固定使用配置值）
+  learned_rpm_limit?: number  // 学习到的 RPM 限制
  // 滑动窗口利用率采样
  utilization_samples?: Array<{ ts: number; util: number }>  // 利用率采样窗口
  last_probe_increase_at?: string  // 上次探测性扩容时间
@@ -100,8 +153,7 @@ export interface EndpointAPIKey {
  rpm_429_count?: number
  last_429_at?: string
  last_429_type?: string
-  // 熔断器字段（滑动窗口 + 半开模式）
-  circuit_breaker_open?: boolean
+  // 单格式场景的熔断器字段
  circuit_breaker_open_at?: string
  next_probe_at?: string
  half_open_until?: string
@@ -110,17 +162,36 @@ export interface EndpointAPIKey {
  request_results_window?: Array<{ ts: number; ok: boolean }>  // 请求结果滑动窗口
 }

+// 按格式的健康度数据
+export interface FormatHealthData {
+  health_score: number
+  error_rate: number
+  window_size: number
+  consecutive_failures: number
+  last_failure_at?: string | null
+  circuit_breaker: FormatCircuitBreakerData
+}
+
+// 按格式的熔断器数据
+export interface FormatCircuitBreakerData {
+  open: boolean
+  open_at?: string | null
+  next_probe_at?: string | null
+  half_open_until?: string | null
+  half_open_successes: number
+  half_open_failures: number
+}
+
 export interface EndpointAPIKeyUpdate {
+  api_formats?: string[]  // 支持的 API 格式列表
  name?: string
  api_key?: string  // 仅在需要更新时提供
-  rate_multiplier?: number
+  rate_multiplier?: number  // 默认成本倍率
+  rate_multipliers?: Record<string, number> | null  // 按 API 格式的成本倍率
  internal_priority?: number
  global_priority?: number | null
-  max_concurrent?: number | null  // null 表示切换为自适应模式
-  rate_limit?: number
-  daily_limit?: number
-  monthly_limit?: number
-  allowed_models?: string[] | null
+  rpm_limit?: number | null  // RPM 速率限制 (1-10000)，null 表示切换为自适应模式
+  allowed_models?: AllowedModels
  capabilities?: Record<string, boolean> | null
  cache_ttl_minutes?: number
  max_probe_interval_minutes?: number
@@ -198,7 +269,6 @@ export interface PublicEndpointStatusMonitorResponse {
 export interface ProviderWithEndpointsSummary {
  id: string
  name: string
-  display_name: string
  description?: string
  website?: string
  provider_priority: number
@@ -208,9 +278,10 @@ export interface ProviderWithEndpointsSummary {
  quota_reset_day?: number
  quota_last_reset_at?: string  // 当前周期开始时间
  quota_expires_at?: string
-  rpm_limit?: number | null
-  rpm_used?: number
-  rpm_reset_at?: string
+  // 请求配置（从 Endpoint 迁移）
+  timeout?: number  // 请求超时（秒）
+  max_retries?: number  // 最大重试次数
+  proxy?: ProxyConfig | null  // 代理配置
  is_active: boolean
  total_endpoints: number
  active_endpoints: number
@@ -253,13 +324,10 @@ export interface HealthSummary {
  }
 }

-export interface ConcurrencyStatus {
-  endpoint_id?: string
-  endpoint_current_concurrency: number
-  endpoint_max_concurrent?: number
-  key_id?: string
-  key_current_concurrency: number
-  key_max_concurrent?: number
+export interface KeyRpmStatus {
+  key_id: string
+  current_rpm: number
+  rpm_limit?: number
 }

 export interface ProviderModelMapping {
@@ -361,7 +429,6 @@ export interface ModelPriceRange {
 export interface ModelCatalogProviderDetail {
  provider_id: string
  provider_name: string
-  provider_display_name?: string | null
  model_id?: string | null
  target_model: string
  input_price_per_1m?: number | null
@@ -534,10 +601,10 @@ export interface UpstreamModel {
 */
 export interface ImportFromUpstreamSuccessItem {
  model_id: string
-  global_model_id: string
-  global_model_name: string
  provider_model_id: string
-  created_global_model: boolean
+  global_model_id?: string  // 可选，未关联时为空字符串
+  global_model_name?: string  // 可选，未关联时为空字符串
+  created_global_model: boolean  // 始终为 false（不再自动创建 GlobalModel）
 }

 /**
--- a/frontend/src/components/ui/collapsible-content.vue
+++ b/frontend/src/components/ui/collapsible-content.vue
@@ -0,0 +1,15 @@
+<script setup lang="ts">
+import { CollapsibleContent, type CollapsibleContentProps } from 'radix-vue'
+import { cn } from '@/lib/utils'
+
+const props = defineProps<CollapsibleContentProps & { class?: string }>()
+</script>
+
+<template>
+  <CollapsibleContent
+    v-bind="props"
+    :class="cn('overflow-hidden data-[state=closed]:animate-collapsible-up data-[state=open]:animate-collapsible-down', props.class)"
+  >
+    <slot />
+  </CollapsibleContent>
+</template>
--- a/frontend/src/components/ui/collapsible-trigger.vue
+++ b/frontend/src/components/ui/collapsible-trigger.vue
@@ -0,0 +1,11 @@
+<script setup lang="ts">
+import { CollapsibleTrigger, type CollapsibleTriggerProps } from 'radix-vue'
+
+const props = defineProps<CollapsibleTriggerProps>()
+</script>
+
+<template>
+  <CollapsibleTrigger v-bind="props" as-child>
+    <slot />
+  </CollapsibleTrigger>
+</template>
--- a/frontend/src/components/ui/collapsible.vue
+++ b/frontend/src/components/ui/collapsible.vue
@@ -0,0 +1,15 @@
+<script setup lang="ts">
+import { CollapsibleRoot, type CollapsibleRootEmits, type CollapsibleRootProps } from 'radix-vue'
+import { useForwardPropsEmits } from 'radix-vue'
+
+const props = defineProps<CollapsibleRootProps>()
+const emits = defineEmits<CollapsibleRootEmits>()
+
+const forwarded = useForwardPropsEmits(props, emits)
+</script>
+
+<template>
+  <CollapsibleRoot v-bind="forwarded">
+    <slot />
+  </CollapsibleRoot>
+</template>
--- a/frontend/src/components/ui/index.ts
+++ b/frontend/src/components/ui/index.ts
@@ -65,3 +65,8 @@ export { default as RefreshButton } from './refresh-button.vue'

 // Tooltip 提示系列
 export { Tooltip, TooltipContent, TooltipProvider, TooltipTrigger } from './tooltip'
+
+// Collapsible 折叠系列
+export { default as Collapsible } from './collapsible.vue'
+export { default as CollapsibleTrigger } from './collapsible-trigger.vue'
+export { default as CollapsibleContent } from './collapsible-content.vue'
--- a/frontend/src/features/api-keys/components/StandaloneKeyFormDialog.vue
+++ b/frontend/src/features/api-keys/components/StandaloneKeyFormDialog.vue
@@ -186,7 +186,7 @@
                    @click.stop
                    @change="toggleSelection('allowed_providers', provider.id)"
                  >
-                  <span class="text-sm">{{ provider.display_name || provider.name }}</span>
+                  <span class="text-sm">{{ provider.name }}</span>
                </div>
                <div
                  v-if="providers.length === 0"
--- a/frontend/src/features/models/components/ModelDetailDrawer.vue
+++ b/frontend/src/features/models/components/ModelDetailDrawer.vue
@@ -460,13 +460,13 @@
                        <TableHead class="h-10 font-semibold">
                          Provider
                        </TableHead>
-                        <TableHead class="w-[120px] h-10 font-semibold">
+                        <TableHead class="w-[100px] h-10 font-semibold">
                          能力
                        </TableHead>
-                        <TableHead class="w-[180px] h-10 font-semibold">
+                        <TableHead class="w-[200px] h-10 font-semibold">
                          价格 ($/M)
                        </TableHead>
-                        <TableHead class="w-[80px] h-10 font-semibold text-center">
+                        <TableHead class="w-[100px] h-10 font-semibold text-center">
                          操作
                        </TableHead>
                      </TableRow>
@@ -484,7 +484,7 @@
                              :class="provider.is_active ? 'bg-green-500' : 'bg-gray-300'"
                              :title="provider.is_active ? '活跃' : '停用'"
                            />
-                            <span class="font-medium truncate">{{ provider.display_name }}</span>
+                            <span class="font-medium truncate">{{ provider.name }}</span>
                          </div>
                        </TableCell>
                        <TableCell class="py-3">
@@ -595,7 +595,7 @@
                            class="w-2 h-2 rounded-full shrink-0"
                            :class="provider.is_active ? 'bg-green-500' : 'bg-gray-300'"
                          />
-                          <span class="font-medium truncate">{{ provider.display_name }}</span>
+                          <span class="font-medium truncate">{{ provider.name }}</span>
                        </div>
                        <div class="flex items-center gap-1 shrink-0">
                          <Button
--- a/frontend/src/features/providers/components/BatchAssignModelsDialog.vue
+++ b/frontend/src/features/providers/components/BatchAssignModelsDialog.vue
@@ -531,20 +531,23 @@ watch(() => props.open, async (isOpen) => {
 // 加载数据
 async function loadData() {
  await Promise.all([loadGlobalModels(), loadExistingModels()])
-  // 默认折叠全局模型组
-  collapsedGroups.value = new Set(['global'])

  // 检查缓存，如果有缓存数据则直接使用
  const cachedModels = getCachedModels(props.providerId)
-  if (cachedModels) {
+  if (cachedModels && cachedModels.length > 0) {
    upstreamModels.value = cachedModels
    upstreamModelsLoaded.value = true
-    // 折叠所有上游模型组
+    // 有多个分组时全部折叠
+    const allGroups = new Set(['global'])
    for (const model of cachedModels) {
      if (model.api_format) {
-        collapsedGroups.value.add(model.api_format)
+        allGroups.add(model.api_format)
      }
    }
+    collapsedGroups.value = allGroups
+  } else {
+    // 只有全局模型时展开
+    collapsedGroups.value = new Set()
  }
 }

@@ -585,8 +588,8 @@ async function fetchUpstreamModels(forceRefresh = false) {
      } else {
        upstreamModels.value = result.models
        upstreamModelsLoaded.value = true
-        // 折叠所有上游模型组
-        const allGroups = new Set(collapsedGroups.value)
+        // 有多个分组时全部折叠
+        const allGroups = new Set(['global'])
        for (const model of result.models) {
          if (model.api_format) {
            allGroups.add(model.api_format)
--- a/frontend/src/features/providers/components/EndpointFormDialog.vue
+++ b/frontend/src/features/providers/components/EndpointFormDialog.vue
@@ -1,245 +1,200 @@
 <template>
  <Dialog
    :model-value="internalOpen"
-    :title="isEditMode ? '编辑 API 端点' : '添加 API 端点'"
-    :description="isEditMode ? `修改 ${provider?.display_name} 的端点配置` : '为提供商添加新的 API 端点'"
-    :icon="isEditMode ? SquarePen : Link"
-    size="xl"
+    title="端点管理"
+    :description="`管理 ${provider?.name} 的 API 端点`"
+    :icon="Settings"
+    size="2xl"
    @update:model-value="handleDialogUpdate"
  >
-    <form
-      class="space-y-6"
-      @submit.prevent="handleSubmit()"
-    >
-      <!-- API 配置 -->
-      <div class="space-y-4">
-        <h3
-          v-if="isEditMode"
-          class="text-sm font-medium"
-        >
-          API 配置
-        </h3>
-
-        <div class="grid grid-cols-2 gap-4">
-          <!-- API 格式 -->
-          <div class="space-y-2">
-            <Label for="api_format">API 格式 *</Label>
-            <template v-if="isEditMode">
-              <Input
-                id="api_format"
-                v-model="form.api_format"
-                disabled
-                class="bg-muted"
-              />
-              <p class="text-xs text-muted-foreground">
-                API 格式创建后不可修改
-              </p>
-            </template>
-            <template v-else>
-              <Select
-                v-model="form.api_format"
-                v-model:open="selectOpen"
-                required
-              >
-                <SelectTrigger>
-                  <SelectValue placeholder="请选择 API 格式" />
-                </SelectTrigger>
-                <SelectContent>
-                  <SelectItem
-                    v-for="format in apiFormats"
-                    :key="format.value"
-                    :value="format.value"
-                  >
-                    {{ format.label }}
-                  </SelectItem>
-                </SelectContent>
-              </Select>
-            </template>
-          </div>
-
-          <!-- API URL -->
-          <div class="space-y-2">
-            <Label for="base_url">API URL *</Label>
-            <Input
-              id="base_url"
-              v-model="form.base_url"
-              placeholder="https://api.example.com"
-              required
-            />
-          </div>
-        </div>
-
-        <!-- 自定义路径 -->
+    <div class="space-y-4">
+      <!-- 已有端点列表 -->
+      <div
+        v-if="localEndpoints.length > 0"
+        class="space-y-2"
+      >
+        <Label class="text-muted-foreground">已配置的端点</Label>
        <div class="space-y-2">
-          <Label for="custom_path">自定义请求路径（可选）</Label>
-          <Input
-            id="custom_path"
-            v-model="form.custom_path"
-            :placeholder="defaultPathPlaceholder"
-          />
-        </div>
-      </div>
-
-      <!-- 请求配置 -->
-      <div class="space-y-4">
-        <h3 class="text-sm font-medium">
-          请求配置
-        </h3>
-
-        <div class="grid grid-cols-3 gap-4">
-          <div class="space-y-2">
-            <Label for="timeout">超时（秒）</Label>
-            <Input
-              id="timeout"
-              v-model.number="form.timeout"
-              type="number"
-              placeholder="300"
-            />
-          </div>
-
-          <div class="space-y-2">
-            <Label for="max_retries">最大重试</Label>
-            <Input
-              id="max_retries"
-              v-model.number="form.max_retries"
-              type="number"
-              placeholder="3"
-            />
-          </div>
-
-          <div class="space-y-2">
-            <Label for="max_concurrent">最大并发</Label>
-            <Input
-              id="max_concurrent"
-              :model-value="form.max_concurrent ?? ''"
-              type="number"
-              placeholder="无限制"
-              @update:model-value="(v) => form.max_concurrent = parseNumberInput(v)"
-            />
-          </div>
-        </div>
-
-        <div class="grid grid-cols-2 gap-4">
-          <div class="space-y-2">
-            <Label for="rate_limit">速率限制（请求/分钟）</Label>
-            <Input
-              id="rate_limit"
-              :model-value="form.rate_limit ?? ''"
-              type="number"
-              placeholder="无限制"
-              @update:model-value="(v) => form.rate_limit = parseNumberInput(v)"
-            />
+          <div
+            v-for="endpoint in localEndpoints"
+            :key="endpoint.id"
+            class="rounded-md border px-3 py-2"
+            :class="{ 'opacity-50': !endpoint.is_active }"
+          >
+            <!-- 编辑模式 -->
+            <template v-if="editingEndpointId === endpoint.id">
+              <div class="space-y-2">
+                <div class="flex items-center gap-2">
+                  <span class="text-sm font-medium w-24 shrink-0">{{ API_FORMAT_LABELS[endpoint.api_format] || endpoint.api_format }}</span>
+                  <div class="flex items-center gap-1 ml-auto">
+                    <Button
+                      variant="ghost"
+                      size="icon"
+                      class="h-7 w-7"
+                      title="保存"
+                      :disabled="savingEndpointId === endpoint.id"
+                      @click="saveEndpointUrl(endpoint)"
+                    >
+                      <Check class="w-3.5 h-3.5" />
+                    </Button>
+                    <Button
+                      variant="ghost"
+                      size="icon"
+                      class="h-7 w-7"
+                      title="取消"
+                      @click="cancelEdit"
+                    >
+                      <X class="w-3.5 h-3.5" />
+                    </Button>
+                  </div>
+                </div>
+                <div class="grid grid-cols-2 gap-2">
+                  <div class="space-y-1">
+                    <Label class="text-xs text-muted-foreground">Base URL</Label>
+                    <Input
+                      v-model="editingUrl"
+                      class="h-8 text-sm"
+                      placeholder="https://api.example.com"
+                      @keyup.escape="cancelEdit"
+                    />
+                  </div>
+                  <div class="space-y-1">
+                    <Label class="text-xs text-muted-foreground">自定义路径 (可选)</Label>
+                    <Input
+                      v-model="editingPath"
+                      class="h-8 text-sm"
+                      :placeholder="editingDefaultPath || '留空使用默认路径'"
+                      @keyup.escape="cancelEdit"
+                    />
+                  </div>
+                </div>
+              </div>
+            </template>
+            <!-- 查看模式 -->
+            <template v-else>
+              <div class="flex items-center gap-3">
+                <div class="w-24 shrink-0">
+                  <span class="text-sm font-medium">{{ API_FORMAT_LABELS[endpoint.api_format] || endpoint.api_format }}</span>
+                </div>
+                <div class="flex-1 min-w-0">
+                  <span class="text-sm text-muted-foreground truncate block">
+                    {{ endpoint.base_url }}{{ endpoint.custom_path ? endpoint.custom_path : '' }}
+                  </span>
+                </div>
+                <div class="flex items-center gap-1 shrink-0">
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-7 w-7"
+                    title="编辑"
+                    @click="startEdit(endpoint)"
+                  >
+                    <Edit class="w-3.5 h-3.5" />
+                  </Button>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-7 w-7"
+                    :title="endpoint.is_active ? '停用' : '启用'"
+                    :disabled="togglingEndpointId === endpoint.id"
+                    @click="handleToggleEndpoint(endpoint)"
+                  >
+                    <Power class="w-3.5 h-3.5" />
+                  </Button>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-7 w-7 text-destructive hover:text-destructive"
+                    title="删除"
+                    :disabled="deletingEndpointId === endpoint.id"
+                    @click="handleDeleteEndpoint(endpoint)"
+                  >
+                    <Trash2 class="w-3.5 h-3.5" />
+                  </Button>
+                </div>
+              </div>
+            </template>
          </div>
        </div>
      </div>

-      <!-- 代理配置 -->
-      <div class="space-y-4">
-        <div class="flex items-center justify-between">
-          <h3 class="text-sm font-medium">
-            代理配置
-          </h3>
-          <div class="flex items-center gap-2">
-            <Switch v-model="proxyEnabled" />
-            <span class="text-sm text-muted-foreground">启用代理</span>
-          </div>
-        </div>
-
-        <div
-          v-if="proxyEnabled"
-          class="space-y-4 rounded-lg border p-4"
-        >
-          <div class="space-y-2">
-            <Label for="proxy_url">代理 URL *</Label>
-            <Input
-              id="proxy_url"
-              v-model="form.proxy_url"
-              placeholder="http://host:port 或 socks5://host:port"
-              required
-              :class="proxyUrlError ? 'border-red-500' : ''"
-            />
-            <p
-              v-if="proxyUrlError"
-              class="text-xs text-red-500"
+      <!-- 添加新端点 -->
+      <div
+        v-if="availableFormats.length > 0"
+        class="space-y-3 pt-3 border-t"
+      >
+        <Label class="text-muted-foreground">添加新端点</Label>
+        <div class="flex items-end gap-3">
+          <div class="w-32 shrink-0 space-y-1.5">
+            <Label class="text-xs">API 格式</Label>
+            <Select
+              v-model="newEndpoint.api_format"
+              v-model:open="formatSelectOpen"
            >
-              {{ proxyUrlError }}
-            </p>
-            <p
-              v-else
-              class="text-xs text-muted-foreground"
-            >
-              支持 HTTP、HTTPS、SOCKS5 代理
-            </p>
+              <SelectTrigger class="h-9">
+                <SelectValue placeholder="选择格式" />
+              </SelectTrigger>
+              <SelectContent>
+                <SelectItem
+                  v-for="format in availableFormats"
+                  :key="format.value"
+                  :value="format.value"
+                >
+                  {{ format.label }}
+                </SelectItem>
+              </SelectContent>
+            </Select>
          </div>
-
-          <div class="grid grid-cols-2 gap-4">
-            <div class="space-y-2">
-              <Label for="proxy_user">用户名（可选）</Label>
-              <Input
-                :id="`proxy_user_${formId}`"
-                v-model="form.proxy_username"
-                :name="`proxy_user_${formId}`"
-                placeholder="代理认证用户名"
-                autocomplete="off"
-                data-form-type="other"
-                data-lpignore="true"
-                data-1p-ignore="true"
-              />
-            </div>
-
-            <div class="space-y-2">
-              <Label :for="`proxy_pass_${formId}`">密码（可选）</Label>
-              <Input
-                :id="`proxy_pass_${formId}`"
-                v-model="form.proxy_password"
-                :name="`proxy_pass_${formId}`"
-                type="text"
-                :placeholder="passwordPlaceholder"
-                autocomplete="off"
-                data-form-type="other"
-                data-lpignore="true"
-                data-1p-ignore="true"
-                :style="{ '-webkit-text-security': 'disc', 'text-security': 'disc' }"
-              />
-            </div>
+          <div class="flex-1 space-y-1.5">
+            <Label class="text-xs">Base URL</Label>
+            <Input
+              v-model="newEndpoint.base_url"
+              placeholder="https://api.example.com"
+              class="h-9"
+            />
          </div>
+          <div class="w-40 shrink-0 space-y-1.5">
+            <Label class="text-xs">自定义路径</Label>
+            <Input
+              v-model="newEndpoint.custom_path"
+              :placeholder="newEndpointDefaultPath || '可选'"
+              class="h-9"
+            />
+          </div>
+          <Button
+            variant="outline"
+            size="sm"
+            class="h-9 shrink-0"
+            :disabled="!newEndpoint.api_format || !newEndpoint.base_url || addingEndpoint"
+            @click="handleAddEndpoint"
+          >
+            {{ addingEndpoint ? '添加中...' : '添加' }}
+          </Button>
        </div>
      </div>
-    </form>
+
+      <!-- 空状态 -->
+      <div
+        v-if="localEndpoints.length === 0 && availableFormats.length === 0"
+        class="text-center py-8 text-muted-foreground"
+      >
+        <p>所有 API 格式都已配置</p>
+      </div>
+    </div>

    <template #footer>
      <Button
-        type="button"
        variant="outline"
-        :disabled="loading"
-        @click="handleCancel"
+        @click="handleClose"
      >
-        取消
-      </Button>
-      <Button
-        :disabled="loading || !form.base_url || (!isEditMode && !form.api_format)"
-        @click="handleSubmit()"
-      >
-        {{ loading ? (isEditMode ? '保存中...' : '创建中...') : (isEditMode ? '保存修改' : '创建') }}
+        关闭
      </Button>
    </template>
  </Dialog>
-
-  <!-- 确认清空凭据对话框 -->
-  <AlertDialog
-    v-model="showClearCredentialsDialog"
-    title="清空代理凭据"
-    description="代理 URL 为空，但用户名和密码仍有值。是否清空这些凭据并继续保存？"
-    type="warning"
-    confirm-text="清空并保存"
-    cancel-text="返回编辑"
-    @confirm="confirmClearCredentials"
-    @cancel="showClearCredentialsDialog = false"
-  />
 </template>

 <script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
+import { ref, computed, onMounted, watch } from 'vue'
 import {
  Dialog,
  Button,
@@ -250,17 +205,15 @@ import {
  SelectValue,
  SelectContent,
  SelectItem,
-  Switch,
 } from '@/components/ui'
-import AlertDialog from '@/components/common/AlertDialog.vue'
-import { Link, SquarePen } from 'lucide-vue-next'
+import { Settings, Edit, Trash2, Check, X, Power } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
-import { useFormDialog } from '@/composables/useFormDialog'
-import { parseNumberInput } from '@/utils/form'
 import { log } from '@/utils/logger'
 import {
  createEndpoint,
  updateEndpoint,
+  deleteEndpoint,
+  API_FORMAT_LABELS,
  type ProviderEndpoint,
  type ProviderWithEndpointsSummary
 } from '@/api/endpoints'
@@ -269,7 +222,7 @@ import { adminApi } from '@/api/admin'
 const props = defineProps<{
  modelValue: boolean
  provider: ProviderWithEndpointsSummary | null
-  endpoint?: ProviderEndpoint | null  // 编辑模式时传入
+  endpoints?: ProviderEndpoint[]
 }>()

 const emit = defineEmits<{
@@ -279,258 +232,184 @@ const emit = defineEmits<{
 }>()

 const { success, error: showError } = useToast()
-const loading = ref(false)
-const selectOpen = ref(false)
-const proxyEnabled = ref(false)
-const showClearCredentialsDialog = ref(false)  // 确认清空凭据对话框

-// 生成随机 ID 防止浏览器自动填充
-const formId = Math.random().toString(36).substring(2, 10)
+// 状态
+const addingEndpoint = ref(false)
+const editingEndpointId = ref<string | null>(null)
+const editingUrl = ref('')
+const editingPath = ref('')
+const savingEndpointId = ref<string | null>(null)
+const deletingEndpointId = ref<string | null>(null)
+const togglingEndpointId = ref<string | null>(null)
+const formatSelectOpen = ref(false)

 // 内部状态
 const internalOpen = computed(() => props.modelValue)

-// 表单数据
-const form = ref({
+// 新端点表单
+const newEndpoint = ref({
  api_format: '',
  base_url: '',
  custom_path: '',
-  timeout: 300,
-  max_retries: 2,
-  max_concurrent: undefined as number | undefined,
-  rate_limit: undefined as number | undefined,
-  is_active: true,
-  // 代理配置
-  proxy_url: '',
-  proxy_username: '',
-  proxy_password: '',
 })

 // API 格式列表
-const apiFormats = ref<Array<{ value: string; label: string; default_path: string; aliases: string[] }>>([])
+const apiFormats = ref<Array<{ value: string; label: string; default_path: string }>>([])

-// 加载API格式列表
+// 本地端点列表
+const localEndpoints = ref<ProviderEndpoint[]>([])
+
+// 可用的格式（未添加的）
+const availableFormats = computed(() => {
+  const existingFormats = localEndpoints.value.map(e => e.api_format)
+  return apiFormats.value.filter(f => !existingFormats.includes(f.value))
+})
+
+// 获取指定 API 格式的默认路径
+function getDefaultPath(apiFormat: string): string {
+  const format = apiFormats.value.find(f => f.value === apiFormat)
+  return format?.default_path || ''
+}
+
+// 当前编辑端点的默认路径
+const editingDefaultPath = computed(() => {
+  const endpoint = localEndpoints.value.find(e => e.id === editingEndpointId.value)
+  return endpoint ? getDefaultPath(endpoint.api_format) : ''
+})
+
+// 新端点选择的格式的默认路径
+const newEndpointDefaultPath = computed(() => {
+  return getDefaultPath(newEndpoint.value.api_format)
+})
+
+// 加载 API 格式列表
 const loadApiFormats = async () => {
  try {
    const response = await adminApi.getApiFormats()
    apiFormats.value = response.formats
  } catch (error) {
    log.error('加载API格式失败:', error)
-    if (!isEditMode.value) {
-      showError('加载API格式失败', '错误')
-    }
  }
 }

-// 根据选择的 API 格式计算默认路径
-const defaultPath = computed(() => {
-  const format = apiFormats.value.find(f => f.value === form.value.api_format)
-  return format?.default_path || '/'
-})
-
-// 动态 placeholder
-const defaultPathPlaceholder = computed(() => {
-  return `留空使用默认路径：${defaultPath.value}`
-})
-
-// 检查是否有已保存的密码（后端返回 *** 表示有密码）
-const hasExistingPassword = computed(() => {
-  if (!props.endpoint?.proxy) return false
-  const proxy = props.endpoint.proxy as { password?: string }
-  return proxy?.password === MASKED_PASSWORD
-})
-
-// 密码输入框的 placeholder
-const passwordPlaceholder = computed(() => {
-  if (hasExistingPassword.value) {
-    return '已保存密码，留空保持不变'
-  }
-  return '代理认证密码'
-})
-
-// 代理 URL 验证
-const proxyUrlError = computed(() => {
-  // 只有启用代理且填写了 URL 时才验证
-  if (!proxyEnabled.value || !form.value.proxy_url) {
-    return ''
-  }
-  const url = form.value.proxy_url.trim()
-
-  // 检查禁止的特殊字符
-  if (/[\n\r]/.test(url)) {
-    return '代理 URL 包含非法字符'
-  }
-
-  // 验证协议（不支持 SOCKS4）
-  if (!/^(http|https|socks5):\/\//i.test(url)) {
-    return '代理 URL 必须以 http://, https:// 或 socks5:// 开头'
-  }
-  try {
-    const parsed = new URL(url)
-    if (!parsed.host) {
-      return '代理 URL 必须包含有效的 host'
-    }
-    // 禁止 URL 中内嵌认证信息
-    if (parsed.username || parsed.password) {
-      return '请勿在 URL 中包含用户名和密码，请使用独立的认证字段'
-    }
-  } catch {
-    return '代理 URL 格式无效'
-  }
-  return ''
-})
-
-// 组件挂载时加载API格式
 onMounted(() => {
  loadApiFormats()
 })

-// 重置表单
-function resetForm() {
-  form.value = {
-    api_format: '',
-    base_url: '',
-    custom_path: '',
-    timeout: 300,
-    max_retries: 2,
-    max_concurrent: undefined,
-    rate_limit: undefined,
-    is_active: true,
-    proxy_url: '',
-    proxy_username: '',
-    proxy_password: '',
+// 监听 props 变化
+watch(() => props.modelValue, (open) => {
+  if (open) {
+    localEndpoints.value = [...(props.endpoints || [])]
+    // 重置编辑状态
+    editingEndpointId.value = null
+    editingUrl.value = ''
+    editingPath.value = ''
+  } else {
+    // 关闭对话框时完全清空新端点表单
+    newEndpoint.value = { api_format: '', base_url: '', custom_path: '' }
  }
-  proxyEnabled.value = false
+}, { immediate: true })
+
+watch(() => props.endpoints, (endpoints) => {
+  if (props.modelValue) {
+    localEndpoints.value = [...(endpoints || [])]
+  }
+}, { deep: true })
+
+// 开始编辑
+function startEdit(endpoint: ProviderEndpoint) {
+  editingEndpointId.value = endpoint.id
+  editingUrl.value = endpoint.base_url
+  editingPath.value = endpoint.custom_path || ''
 }

-// 原始密码占位符（后端返回的脱敏标记）
-const MASKED_PASSWORD = '***'
-
-// 加载端点数据（编辑模式）
-function loadEndpointData() {
-  if (!props.endpoint) return
-
-  const proxy = props.endpoint.proxy as { url?: string; username?: string; password?: string; enabled?: boolean } | null
-
-  form.value = {
-    api_format: props.endpoint.api_format,
-    base_url: props.endpoint.base_url,
-    custom_path: props.endpoint.custom_path || '',
-    timeout: props.endpoint.timeout,
-    max_retries: props.endpoint.max_retries,
-    max_concurrent: props.endpoint.max_concurrent || undefined,
-    rate_limit: props.endpoint.rate_limit || undefined,
-    is_active: props.endpoint.is_active,
-    proxy_url: proxy?.url || '',
-    proxy_username: proxy?.username || '',
-    // 如果密码是脱敏标记，显示为空（让用户知道有密码但看不到）
-    proxy_password: proxy?.password === MASKED_PASSWORD ? '' : (proxy?.password || ''),
-  }
-
-  // 根据 enabled 字段或 url 存在判断是否启用代理
-  proxyEnabled.value = proxy?.enabled ?? !!proxy?.url
+// 取消编辑
+function cancelEdit() {
+  editingEndpointId.value = null
+  editingUrl.value = ''
+  editingPath.value = ''
 }

-// 使用 useFormDialog 统一处理对话框逻辑
-const { isEditMode, handleDialogUpdate, handleCancel } = useFormDialog({
-  isOpen: () => props.modelValue,
-  entity: () => props.endpoint,
-  isLoading: loading,
-  onClose: () => emit('update:modelValue', false),
-  loadData: loadEndpointData,
-  resetForm,
-})
+// 保存端点
+async function saveEndpointUrl(endpoint: ProviderEndpoint) {
+  if (!editingUrl.value) return

-// 构建代理配置
-// - 有 URL 时始终保存配置，通过 enabled 字段控制是否启用
-// - 无 URL 时返回 null
-function buildProxyConfig(): { url: string; username?: string; password?: string; enabled: boolean } | null {
-  if (!form.value.proxy_url) {
-    // 没填 URL，无代理配置
-    return null
-  }
-  return {
-    url: form.value.proxy_url,
-    username: form.value.proxy_username || undefined,
-    password: form.value.proxy_password || undefined,
-    enabled: proxyEnabled.value,  // 开关状态决定是否启用
-  }
-}
-
-// 提交表单
-const handleSubmit = async (skipCredentialCheck = false) => {
-  if (!props.provider && !props.endpoint) return
-
-  // 只在开关开启且填写了 URL 时验证
-  if (proxyEnabled.value && form.value.proxy_url && proxyUrlError.value) {
-    showError(proxyUrlError.value, '代理配置错误')
-    return
-  }
-
-  // 检查：开关开启但没有 URL，却有用户名或密码
-  const hasOrphanedCredentials = proxyEnabled.value
-    && !form.value.proxy_url
-    && (form.value.proxy_username || form.value.proxy_password)
-
-  if (hasOrphanedCredentials && !skipCredentialCheck) {
-    // 弹出确认对话框
-    showClearCredentialsDialog.value = true
-    return
-  }
-
-  loading.value = true
+  savingEndpointId.value = endpoint.id
  try {
-    const proxyConfig = buildProxyConfig()
-
-    if (isEditMode.value && props.endpoint) {
-      // 更新端点
-      await updateEndpoint(props.endpoint.id, {
-        base_url: form.value.base_url,
-        custom_path: form.value.custom_path || undefined,
-        timeout: form.value.timeout,
-        max_retries: form.value.max_retries,
-        max_concurrent: form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        is_active: form.value.is_active,
-        proxy: proxyConfig,
-      })
-
-      success('端点已更新', '保存成功')
-      emit('endpointUpdated')
-    } else if (props.provider) {
-      // 创建端点
-      await createEndpoint(props.provider.id, {
-        provider_id: props.provider.id,
-        api_format: form.value.api_format,
-        base_url: form.value.base_url,
-        custom_path: form.value.custom_path || undefined,
-        timeout: form.value.timeout,
-        max_retries: form.value.max_retries,
-        max_concurrent: form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        is_active: form.value.is_active,
-        proxy: proxyConfig,
-      })
-
-      success('端点创建成功', '成功')
-      emit('endpointCreated')
-      resetForm()
-    }
-
-    emit('update:modelValue', false)
+    await updateEndpoint(endpoint.id, {
+      base_url: editingUrl.value,
+      custom_path: editingPath.value || null,  // 空字符串时传 null 清空
+    })
+    success('端点已更新')
+    emit('endpointUpdated')
+    cancelEdit()
  } catch (error: any) {
-    const action = isEditMode.value ? '更新' : '创建'
-    showError(error.response?.data?.detail || `${action}端点失败`, '错误')
+    showError(error.response?.data?.detail || '更新失败', '错误')
  } finally {
-    loading.value = false
+    savingEndpointId.value = null
  }
 }

-// 确认清空凭据并继续保存
-const confirmClearCredentials = () => {
-  form.value.proxy_username = ''
-  form.value.proxy_password = ''
-  showClearCredentialsDialog.value = false
-  handleSubmit(true)  // 跳过凭据检查，直接提交
+// 添加端点
+async function handleAddEndpoint() {
+  if (!props.provider || !newEndpoint.value.api_format || !newEndpoint.value.base_url) return
+
+  addingEndpoint.value = true
+  try {
+    await createEndpoint(props.provider.id, {
+      provider_id: props.provider.id,
+      api_format: newEndpoint.value.api_format,
+      base_url: newEndpoint.value.base_url,
+      custom_path: newEndpoint.value.custom_path || undefined,
+      is_active: true,
+    })
+    success(`已添加 ${API_FORMAT_LABELS[newEndpoint.value.api_format] || newEndpoint.value.api_format} 端点`)
+    // 重置表单，保留 URL
+    const url = newEndpoint.value.base_url
+    newEndpoint.value = { api_format: '', base_url: url, custom_path: '' }
+    emit('endpointCreated')
+  } catch (error: any) {
+    showError(error.response?.data?.detail || '添加失败', '错误')
+  } finally {
+    addingEndpoint.value = false
+  }
+}
+
+// 切换端点启用状态
+async function handleToggleEndpoint(endpoint: ProviderEndpoint) {
+  togglingEndpointId.value = endpoint.id
+  try {
+    const newStatus = !endpoint.is_active
+    await updateEndpoint(endpoint.id, { is_active: newStatus })
+    success(newStatus ? '端点已启用' : '端点已停用')
+    emit('endpointUpdated')
+  } catch (error: any) {
+    showError(error.response?.data?.detail || '操作失败', '错误')
+  } finally {
+    togglingEndpointId.value = null
+  }
+}
+
+// 删除端点
+async function handleDeleteEndpoint(endpoint: ProviderEndpoint) {
+  deletingEndpointId.value = endpoint.id
+  try {
+    await deleteEndpoint(endpoint.id)
+    success(`已删除 ${API_FORMAT_LABELS[endpoint.api_format] || endpoint.api_format} 端点`)
+    emit('endpointUpdated')
+  } catch (error: any) {
+    showError(error.response?.data?.detail || '删除失败', '错误')
+  } finally {
+    deletingEndpointId.value = null
+  }
+}
+
+// 关闭对话框
+function handleDialogUpdate(value: boolean) {
+  emit('update:modelValue', value)
+}
+
+function handleClose() {
+  emit('update:modelValue', false)
 }
 </script>
--- a/frontend/src/features/providers/components/KeyAllowedModelsDialog.vue
+++ b/frontend/src/features/providers/components/KeyAllowedModelsDialog.vue
@@ -1,165 +1,179 @@
 <template>
  <Dialog
    :model-value="isOpen"
-    title="配置允许的模型"
-    description="选择该 API Key 允许访问的模型，留空则允许访问所有模型"
-    :icon="Settings2"
+    title="获取上游模型"
+    :description="`使用密钥 ${props.apiKey?.name || props.apiKey?.api_key_masked || ''} 从上游获取模型列表。导入的模型需要关联全局模型后才能参与路由。`"
+    :icon="Layers"
    size="2xl"
    @update:model-value="handleDialogUpdate"
  >
    <div class="space-y-4 py-2">
-      <!-- 已选模型展示 -->
-      <div
-        v-if="selectedModels.length > 0"
-        class="space-y-2"
-      >
-        <div class="flex items-center justify-between px-1">
-          <div class="text-xs font-medium text-muted-foreground">
-            已选模型 ({{ selectedModels.length }})
-          </div>
-          <Button
-            type="button"
-            variant="ghost"
-            size="sm"
-            class="h-6 text-xs hover:text-destructive"
-            @click="clearModels"
-          >
-            清空
-          </Button>
-        </div>
-        <div class="flex flex-wrap gap-1.5 p-2 bg-muted/20 rounded-lg border border-border/40 min-h-[40px]">
-          <Badge
-            v-for="modelName in selectedModels"
-            :key="modelName"
-            variant="secondary"
-            class="text-[11px] px-2 py-0.5 bg-background border-border/60 shadow-sm"
-          >
-            {{ getModelLabel(modelName) }}
-            <button
-              class="ml-0.5 hover:text-destructive focus:outline-none"
-              @click.stop="toggleModel(modelName, false)"
-            >
-              &times;
-            </button>
-          </Badge>
+      <!-- 操作区域 -->
+      <div class="flex items-center justify-between">
+        <div class="text-sm text-muted-foreground">
+          <span v-if="!hasQueried">点击获取按钮查询上游可用模型</span>
+          <span v-else-if="upstreamModels.length > 0">
+            共 {{ upstreamModels.length }} 个模型，已选 {{ selectedModels.length }} 个
+          </span>
+          <span v-else>未找到可用模型</span>
        </div>
+        <Button
+          variant="outline"
+          size="sm"
+          :disabled="loading"
+          @click="fetchUpstreamModels"
+        >
+          <RefreshCw
+            class="w-3.5 h-3.5 mr-1.5"
+            :class="{ 'animate-spin': loading }"
+          />
+          {{ hasQueried ? '刷新' : '获取模型' }}
+        </Button>
      </div>

-      <!-- 模型列表区域 -->
-      <div class="space-y-2">
+      <!-- 加载状态 -->
+      <div
+        v-if="loading"
+        class="flex flex-col items-center justify-center py-12 space-y-3"
+      >
+        <div class="animate-spin rounded-full h-8 w-8 border-2 border-primary/20 border-t-primary" />
+        <span class="text-xs text-muted-foreground">正在从上游获取模型列表...</span>
+      </div>
+
+      <!-- 错误状态 -->
+      <div
+        v-else-if="errorMessage"
+        class="flex flex-col items-center justify-center py-12 text-destructive border border-dashed border-destructive/30 rounded-lg bg-destructive/5"
+      >
+        <AlertCircle class="w-10 h-10 mb-2 opacity-50" />
+        <span class="text-sm text-center px-4">{{ errorMessage }}</span>
+        <Button
+          variant="outline"
+          size="sm"
+          class="mt-3"
+          @click="fetchUpstreamModels"
+        >
+          重试
+        </Button>
+      </div>
+
+      <!-- 未查询状态 -->
+      <div
+        v-else-if="!hasQueried"
+        class="flex flex-col items-center justify-center py-12 text-muted-foreground border border-dashed rounded-lg bg-muted/10"
+      >
+        <Layers class="w-10 h-10 mb-2 opacity-20" />
+        <span class="text-sm">点击上方按钮获取模型列表</span>
+      </div>
+
+      <!-- 无模型 -->
+      <div
+        v-else-if="upstreamModels.length === 0"
+        class="flex flex-col items-center justify-center py-12 text-muted-foreground border border-dashed rounded-lg bg-muted/10"
+      >
+        <Box class="w-10 h-10 mb-2 opacity-20" />
+        <span class="text-sm">上游 API 未返回可用模型</span>
+      </div>
+
+      <!-- 模型列表 -->
+      <div v-else class="space-y-2">
+        <!-- 全选/取消 -->
        <div class="flex items-center justify-between px-1">
-          <div class="text-xs font-medium text-muted-foreground">
-            可选模型列表
+          <div class="flex items-center gap-2">
+            <Checkbox
+              :checked="isAllSelected"
+              :indeterminate="isPartiallySelected"
+              @update:checked="toggleSelectAll"
+            />
+            <span class="text-xs text-muted-foreground">
+              {{ isAllSelected ? '取消全选' : '全选' }}
+            </span>
          </div>
-          <div
-            v-if="!loadingModels && availableModels.length > 0"
-            class="text-[10px] text-muted-foreground/60"
-          >
-            共 {{ availableModels.length }} 个模型
+          <div class="text-xs text-muted-foreground">
+            {{ newModelsCount }} 个新模型（不在本地）
          </div>
        </div>

-        <!-- 加载状态 -->
-        <div
-          v-if="loadingModels"
-          class="flex flex-col items-center justify-center py-12 space-y-3"
-        >
-          <div class="animate-spin rounded-full h-8 w-8 border-2 border-primary/20 border-t-primary" />
-          <span class="text-xs text-muted-foreground">正在加载模型列表...</span>
-        </div>
-
-        <!-- 无模型 -->
-        <div
-          v-else-if="availableModels.length === 0"
-          class="flex flex-col items-center justify-center py-12 text-muted-foreground border border-dashed rounded-lg bg-muted/10"
-        >
-          <Box class="w-10 h-10 mb-2 opacity-20" />
-          <span class="text-sm">暂无可选模型</span>
-        </div>
-
-        <!-- 模型列表 -->
-        <div
-          v-else
-          class="max-h-[320px] overflow-y-auto pr-1 space-y-1.5 custom-scrollbar"
-        >
+        <div class="max-h-[320px] overflow-y-auto pr-1 space-y-1 custom-scrollbar">
          <div
-            v-for="model in availableModels"
-            :key="model.global_model_name"
+            v-for="model in upstreamModels"
+            :key="`${model.id}:${model.api_format || ''}`"
            class="group flex items-center gap-3 px-3 py-2.5 rounded-lg border transition-all duration-200 cursor-pointer select-none"
            :class="[
-              selectedModels.includes(model.global_model_name)
+              selectedModels.includes(model.id)
                ? 'border-primary/40 bg-primary/5 shadow-sm'
                : 'border-border/40 bg-background hover:border-primary/20 hover:bg-muted/30'
            ]"
-            @click="toggleModel(model.global_model_name, !selectedModels.includes(model.global_model_name))"
+            @click="toggleModel(model.id)"
          >
-            <!-- Checkbox -->
            <Checkbox
-              :checked="selectedModels.includes(model.global_model_name)"
+              :checked="selectedModels.includes(model.id)"
              class="data-[state=checked]:bg-primary data-[state=checked]:border-primary"
              @click.stop
-              @update:checked="checked => toggleModel(model.global_model_name, checked)"
+              @update:checked="checked => toggleModel(model.id, checked)"
            />
-
-            <!-- Info -->
            <div class="flex-1 min-w-0">
-              <div class="flex items-center justify-between gap-2">
-                <span class="text-sm font-medium truncate text-foreground/90">{{ model.display_name }}</span>
-                <span
-                  v-if="hasPricing(model)"
-                  class="text-[10px] font-mono text-muted-foreground/80 bg-muted/30 px-1.5 py-0.5 rounded border border-border/30 shrink-0"
-                >
-                  {{ formatPricingShort(model) }}
+              <div class="flex items-center gap-2">
+                <span class="text-sm font-medium truncate text-foreground/90">
+                  {{ model.display_name || model.id }}
                </span>
+                <Badge
+                  v-if="model.api_format"
+                  variant="outline"
+                  class="text-[10px] px-1.5 py-0 shrink-0"
+                >
+                  {{ API_FORMAT_LABELS[model.api_format] || model.api_format }}
+                </Badge>
+                <Badge
+                  v-if="isModelExisting(model.id)"
+                  variant="secondary"
+                  class="text-[10px] px-1.5 py-0 shrink-0"
+                >
+                  已存在
+                </Badge>
              </div>
              <div class="text-[11px] text-muted-foreground/60 font-mono truncate mt-0.5">
-                {{ model.global_model_name }}
+                {{ model.id }}
              </div>
            </div>
-
-            <!-- 测试按钮 -->
-            <Button
-              variant="ghost"
-              size="icon"
-              class="h-7 w-7 shrink-0"
-              title="测试模型连接"
-              :disabled="testingModelName === model.global_model_name"
-              @click.stop="testModelConnection(model)"
+            <div
+              v-if="model.owned_by"
+              class="text-[10px] text-muted-foreground/50 shrink-0"
            >
-              <Loader2
-                v-if="testingModelName === model.global_model_name"
-                class="w-3.5 h-3.5 animate-spin"
-              />
-              <Play
-                v-else
-                class="w-3.5 h-3.5"
-              />
-            </Button>
+              {{ model.owned_by }}
+            </div>
          </div>
        </div>
      </div>
    </div>

    <template #footer>
-      <div class="flex items-center justify-end gap-2 w-full pt-2">
-        <Button
-          variant="outline"
-          class="h-9"
-          @click="handleCancel"
-        >
-          取消
-        </Button>
-        <Button
-          :disabled="saving"
-          class="h-9 min-w-[80px]"
-          @click="handleSave"
-        >
-          <Loader2
-            v-if="saving"
-            class="w-3.5 h-3.5 mr-1.5 animate-spin"
-          />
-          {{ saving ? '保存中' : '保存配置' }}
-        </Button>
+      <div class="flex items-center justify-between w-full pt-2">
+        <div class="text-xs text-muted-foreground">
+          <span v-if="selectedModels.length > 0 && newSelectedCount > 0">
+            将导入 {{ newSelectedCount }} 个新模型
+          </span>
+        </div>
+        <div class="flex items-center gap-2">
+          <Button
+            variant="outline"
+            class="h-9"
+            @click="handleCancel"
+          >
+            取消
+          </Button>
+          <Button
+            :disabled="importing || selectedModels.length === 0 || newSelectedCount === 0"
+            class="h-9 min-w-[100px]"
+            @click="handleImport"
+          >
+            <Loader2
+              v-if="importing"
+              class="w-3.5 h-3.5 mr-1.5 animate-spin"
+            />
+            {{ importing ? '导入中' : `导入 ${newSelectedCount} 个模型` }}
+          </Button>
+        </div>
      </div>
    </template>
  </Dialog>
@@ -167,19 +181,19 @@

 <script setup lang="ts">
 import { ref, computed, watch } from 'vue'
-import { Box, Loader2, Settings2, Play } from 'lucide-vue-next'
+import { Box, Layers, Loader2, RefreshCw, AlertCircle } from 'lucide-vue-next'
 import { Dialog } from '@/components/ui'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import Checkbox from '@/components/ui/checkbox.vue'
 import { useToast } from '@/composables/useToast'
-import { parseApiError, parseTestModelError } from '@/utils/errorParser'
+import { adminApi } from '@/api/admin'
 import {
-  updateEndpointKey,
-  getProviderAvailableSourceModels,
-  testModel,
+  importModelsFromUpstream,
+  getProviderModels,
  type EndpointAPIKey,
-  type ProviderAvailableSourceModel
+  type UpstreamModel,
+  API_FORMAT_LABELS,
 } from '@/api/endpoints'

 const props = defineProps<{
@@ -196,130 +210,116 @@ const emit = defineEmits<{
 const { success, error: showError } = useToast()

 const isOpen = computed(() => props.open)
-const saving = ref(false)
-const loadingModels = ref(false)
-const availableModels = ref<ProviderAvailableSourceModel[]>([])
+const loading = ref(false)
+const importing = ref(false)
+const hasQueried = ref(false)
+const errorMessage = ref('')
+const upstreamModels = ref<UpstreamModel[]>([])
 const selectedModels = ref<string[]>([])
-const initialModels = ref<string[]>([])
-const testingModelName = ref<string | null>(null)
+const existingModelIds = ref<Set<string>>(new Set())
+
+// 计算属性
+const isAllSelected = computed(() =>
+  upstreamModels.value.length > 0 &&
+  selectedModels.value.length === upstreamModels.value.length
+)
+
+const isPartiallySelected = computed(() =>
+  selectedModels.value.length > 0 &&
+  selectedModels.value.length < upstreamModels.value.length
+)
+
+const newModelsCount = computed(() =>
+  upstreamModels.value.filter(m => !existingModelIds.value.has(m.id)).length
+)
+
+const newSelectedCount = computed(() =>
+  selectedModels.value.filter(id => !existingModelIds.value.has(id)).length
+)
+
+// 检查模型是否已存在
+function isModelExisting(modelId: string): boolean {
+  return existingModelIds.value.has(modelId)
+}

 // 监听对话框打开
 watch(() => props.open, (open) => {
  if (open) {
-    loadData()
+    resetState()
+    loadExistingModels()
  }
 })

-async function loadData() {
-  // 初始化已选模型
-  if (props.apiKey?.allowed_models) {
-    selectedModels.value = [...props.apiKey.allowed_models]
-    initialModels.value = [...props.apiKey.allowed_models]
-  } else {
-    selectedModels.value = []
-    initialModels.value = []
-  }
-
-  // 加载可选模型
-  if (props.providerId) {
-    await loadAvailableModels()
-  }
-}
-
-async function loadAvailableModels() {
-  if (!props.providerId) return
-  try {
-    loadingModels.value = true
-    const response = await getProviderAvailableSourceModels(props.providerId)
-    availableModels.value = response.models
-  } catch (err: any) {
-    const errorMessage = parseApiError(err, '加载模型列表失败')
-    showError(errorMessage, '错误')
-  } finally {
-    loadingModels.value = false
-  }
-}
-
-const modelLabelMap = computed(() => {
-  const map = new Map<string, string>()
-  availableModels.value.forEach(model => {
-    map.set(model.global_model_name, model.display_name || model.global_model_name)
-  })
-  return map
-})
-
-function getModelLabel(modelName: string): string {
-  return modelLabelMap.value.get(modelName) ?? modelName
-}
-
-function hasPricing(model: ProviderAvailableSourceModel): boolean {
-  const input = model.price.input_price_per_1m ?? 0
-  const output = model.price.output_price_per_1m ?? 0
-  return input > 0 || output > 0
-}
-
-function formatPricingShort(model: ProviderAvailableSourceModel): string {
-  const input = model.price.input_price_per_1m ?? 0
-  const output = model.price.output_price_per_1m ?? 0
-  if (input > 0 || output > 0) {
-    return `$${formatPrice(input)}/$${formatPrice(output)}`
-  }
-  return ''
-}
-
-function formatPrice(value?: number | null): string {
-  if (value === undefined || value === null || value === 0) return '0'
-  if (value >= 1) {
-    return value.toFixed(2)
-  }
-  return value.toFixed(2)
-}
-
-function toggleModel(modelName: string, checked: boolean) {
-  if (checked) {
-    if (!selectedModels.value.includes(modelName)) {
-      selectedModels.value = [...selectedModels.value, modelName]
-    }
-  } else {
-    selectedModels.value = selectedModels.value.filter(name => name !== modelName)
-  }
-}
-
-function clearModels() {
+function resetState() {
+  hasQueried.value = false
+  errorMessage.value = ''
+  upstreamModels.value = []
  selectedModels.value = []
 }

-// 测试模型连接
-async function testModelConnection(model: ProviderAvailableSourceModel) {
-  if (!props.providerId || !props.apiKey || testingModelName.value) return
-
-  testingModelName.value = model.global_model_name
+// 加载已存在的模型列表
+async function loadExistingModels() {
+  if (!props.providerId) return
  try {
-    const result = await testModel({
-      provider_id: props.providerId,
-      model_name: model.provider_model_name,
-      api_key_id: props.apiKey.id,
-      message: "hello"
-    })
-
-    if (result.success) {
-      success(`模型 "${model.display_name}" 测试成功`)
-    } else {
-      showError(`模型测试失败: ${parseTestModelError(result)}`)
-    }
-  } catch (err: any) {
-    const errorMsg = err.response?.data?.detail || err.message || '测试请求失败'
-    showError(`模型测试失败: ${errorMsg}`)
-  } finally {
-    testingModelName.value = null
+    const models = await getProviderModels(props.providerId)
+    existingModelIds.value = new Set(
+      models.map((m: { provider_model_name: string }) => m.provider_model_name)
+    )
+  } catch {
+    existingModelIds.value = new Set()
  }
 }

-function areArraysEqual(a: string[], b: string[]): boolean {
-  if (a.length !== b.length) return false
-  const sortedA = [...a].sort()
-  const sortedB = [...b].sort()
-  return sortedA.every((value, index) => value === sortedB[index])
+// 获取上游模型
+async function fetchUpstreamModels() {
+  if (!props.providerId || !props.apiKey) return
+
+  loading.value = true
+  errorMessage.value = ''
+
+  try {
+    const response = await adminApi.queryProviderModels(props.providerId, props.apiKey.id)
+
+    if (response.success && response.data?.models) {
+      upstreamModels.value = response.data.models
+      // 默认选中所有新模型
+      selectedModels.value = response.data.models
+        .filter((m: UpstreamModel) => !existingModelIds.value.has(m.id))
+        .map((m: UpstreamModel) => m.id)
+      hasQueried.value = true
+      // 如果有部分失败，显示警告提示
+      if (response.data.error) {
+        showError(`部分格式获取失败: ${response.data.error}`, '警告')
+      }
+    } else {
+      errorMessage.value = response.data?.error || '获取上游模型失败'
+    }
+  } catch (err: any) {
+    errorMessage.value = err.response?.data?.detail || '获取上游模型失败'
+  } finally {
+    loading.value = false
+  }
+}
+
+// 切换模型选择
+function toggleModel(modelId: string, checked?: boolean) {
+  const shouldSelect = checked !== undefined ? checked : !selectedModels.value.includes(modelId)
+  if (shouldSelect) {
+    if (!selectedModels.value.includes(modelId)) {
+      selectedModels.value = [...selectedModels.value, modelId]
+    }
+  } else {
+    selectedModels.value = selectedModels.value.filter(id => id !== modelId)
+  }
+}
+
+// 全选/取消全选
+function toggleSelectAll(checked: boolean) {
+  if (checked) {
+    selectedModels.value = upstreamModels.value.map(m => m.id)
+  } else {
+    selectedModels.value = []
+  }
 }

 function handleDialogUpdate(value: boolean) {
@@ -332,30 +332,44 @@ function handleCancel() {
  emit('close')
 }

-async function handleSave() {
-  if (!props.apiKey) return
+// 导入选中的模型
+async function handleImport() {
+  if (!props.providerId || selectedModels.value.length === 0) return

-  // 检查是否有变化
-  const hasChanged = !areArraysEqual(selectedModels.value, initialModels.value)
-  if (!hasChanged) {
-    emit('close')
+  // 过滤出新模型（不在已存在列表中的）
+  const modelsToImport = selectedModels.value.filter(id => !existingModelIds.value.has(id))
+  if (modelsToImport.length === 0) {
+    showError('所选模型都已存在', '提示')
    return
  }

-  saving.value = true
+  importing.value = true
  try {
-    await updateEndpointKey(props.apiKey.id, {
-      // 空数组时发送 null，表示允许所有模型
-      allowed_models: selectedModels.value.length > 0 ? [...selectedModels.value] : null
-    })
-    success('允许的模型已更新', '成功')
-    emit('saved')
-    emit('close')
+    const response = await importModelsFromUpstream(props.providerId, modelsToImport)
+
+    const successCount = response.success?.length || 0
+    const errorCount = response.errors?.length || 0
+
+    if (successCount > 0 && errorCount === 0) {
+      success(`成功导入 ${successCount} 个模型`, '导入成功')
+      emit('saved')
+      emit('close')
+    } else if (successCount > 0 && errorCount > 0) {
+      success(`成功导入 ${successCount} 个模型，${errorCount} 个失败`, '部分成功')
+      emit('saved')
+      // 刷新列表以更新已存在状态
+      await loadExistingModels()
+      // 更新选中列表，移除已成功导入的
+      const successIds = new Set(response.success?.map((s: { model_id: string }) => s.model_id) || [])
+      selectedModels.value = selectedModels.value.filter(id => !successIds.has(id))
+    } else {
+      const errorMsg = response.errors?.[0]?.error || '导入失败'
+      showError(errorMsg, '导入失败')
+    }
  } catch (err: any) {
-    const errorMessage = parseApiError(err, '保存失败')
-    showError(errorMessage, '错误')
+    showError(err.response?.data?.detail || '导入失败', '错误')
  } finally {
-    saving.value = false
+    importing.value = false
  }
 }
 </script>
--- a/frontend/src/features/providers/components/KeyAllowedModelsEditDialog.vue
+++ b/frontend/src/features/providers/components/KeyAllowedModelsEditDialog.vue
@@ -0,0 +1,696 @@
+<template>
+  <Dialog
+    :model-value="isOpen"
+    title="模型权限"
+    :description="`管理密钥 ${props.apiKey?.name || ''} 可访问的模型，清空右侧列表表示允许全部`"
+    :icon="Shield"
+    size="4xl"
+    @update:model-value="handleDialogUpdate"
+  >
+    <template #default>
+      <div class="space-y-4">
+        <!-- 字典模式警告 -->
+        <div
+          v-if="isDictMode"
+          class="rounded-lg border border-amber-500/50 bg-amber-50 dark:bg-amber-950/30 p-3"
+        >
+          <p class="text-sm text-amber-700 dark:text-amber-400">
+            <strong>注意：</strong>此密钥使用按 API 格式区分的模型权限配置。
+            编辑后将转换为统一列表模式，原有的格式区分信息将丢失。
+          </p>
+        </div>
+
+        <!-- 密钥信息头部 -->
+        <div class="rounded-lg border bg-muted/30 p-4">
+          <div class="flex items-start justify-between">
+            <div>
+              <p class="font-semibold text-lg">{{ apiKey?.name }}</p>
+              <p class="text-sm text-muted-foreground font-mono">
+                {{ apiKey?.api_key_masked }}
+              </p>
+            </div>
+            <Badge
+              :variant="allowedModels.length === 0 ? 'default' : 'outline'"
+              class="text-xs"
+            >
+              {{ allowedModels.length === 0 ? '允许全部' : `限制 ${allowedModels.length} 个模型` }}
+            </Badge>
+          </div>
+        </div>
+
+        <!-- 左右对比布局 -->
+        <div class="flex gap-2 items-stretch">
+          <!-- 左侧：可添加的模型 -->
+          <div class="flex-1 space-y-2">
+            <div class="flex items-center justify-between gap-2">
+              <p class="text-sm font-medium shrink-0">可添加</p>
+              <div class="flex-1 relative">
+                <Search class="absolute left-2 top-1/2 -translate-y-1/2 w-3.5 h-3.5 text-muted-foreground" />
+                <Input
+                  v-model="searchQuery"
+                  placeholder="搜索模型..."
+                  class="pl-7 h-7 text-xs"
+                />
+              </div>
+              <button
+                v-if="upstreamModelsLoaded"
+                type="button"
+                class="p-1.5 hover:bg-muted rounded-md transition-colors shrink-0"
+                title="刷新上游模型"
+                :disabled="fetchingUpstreamModels"
+                @click="fetchUpstreamModels()"
+              >
+                <RefreshCw
+                  class="w-3.5 h-3.5"
+                  :class="{ 'animate-spin': fetchingUpstreamModels }"
+                />
+              </button>
+              <button
+                v-else-if="!fetchingUpstreamModels"
+                type="button"
+                class="p-1.5 hover:bg-muted rounded-md transition-colors shrink-0"
+                title="从提供商获取模型"
+                @click="fetchUpstreamModels()"
+              >
+                <Zap class="w-3.5 h-3.5" />
+              </button>
+              <Loader2
+                v-else
+                class="w-3.5 h-3.5 animate-spin text-muted-foreground shrink-0"
+              />
+            </div>
+            <div class="border rounded-lg h-80 overflow-y-auto">
+              <div
+                v-if="loadingGlobalModels"
+                class="flex items-center justify-center h-full"
+              >
+                <Loader2 class="w-6 h-6 animate-spin text-primary" />
+              </div>
+              <div
+                v-else-if="totalAvailableCount === 0 && !upstreamModelsLoaded"
+                class="flex flex-col items-center justify-center h-full text-muted-foreground"
+              >
+                <Shield class="w-10 h-10 mb-2 opacity-30" />
+                <p class="text-sm">{{ searchQuery ? '无匹配结果' : '暂无可添加模型' }}</p>
+              </div>
+              <div v-else class="p-2 space-y-2">
+                <!-- 全局模型折叠组 -->
+                <div
+                  v-if="availableGlobalModels.length > 0 || !upstreamModelsLoaded"
+                  class="border rounded-lg overflow-hidden"
+                >
+                  <div class="flex items-center gap-2 px-3 py-2 bg-muted/30">
+                    <button
+                      type="button"
+                      class="flex items-center gap-2 flex-1 hover:bg-muted/50 -mx-1 px-1 rounded transition-colors"
+                      @click="toggleGroupCollapse('global')"
+                    >
+                      <ChevronDown
+                        class="w-4 h-4 transition-transform shrink-0"
+                        :class="collapsedGroups.has('global') ? '-rotate-90' : ''"
+                      />
+                      <span class="text-xs font-medium">全局模型</span>
+                      <span class="text-xs text-muted-foreground">
+                        ({{ availableGlobalModels.length }})
+                      </span>
+                    </button>
+                    <button
+                      v-if="availableGlobalModels.length > 0"
+                      type="button"
+                      class="text-xs text-primary hover:underline shrink-0"
+                      @click.stop="selectAllGlobalModels"
+                    >
+                      {{ isAllGlobalModelsSelected ? '取消' : '全选' }}
+                    </button>
+                  </div>
+                  <div
+                    v-show="!collapsedGroups.has('global')"
+                    class="p-2 space-y-1 border-t"
+                  >
+                    <div
+                      v-if="availableGlobalModels.length === 0"
+                      class="py-4 text-center text-xs text-muted-foreground"
+                    >
+                      所有全局模型均已添加
+                    </div>
+                    <div
+                      v-for="model in availableGlobalModels"
+                      v-else
+                      :key="model.name"
+                      class="flex items-center gap-2 p-2 rounded-lg border transition-colors cursor-pointer"
+                      :class="selectedLeftIds.includes(model.name)
+                        ? 'border-primary bg-primary/10'
+                        : 'hover:bg-muted/50'"
+                      @click="toggleLeftSelection(model.name)"
+                    >
+                      <Checkbox
+                        :checked="selectedLeftIds.includes(model.name)"
+                        @update:checked="toggleLeftSelection(model.name)"
+                        @click.stop
+                      />
+                      <div class="flex-1 min-w-0">
+                        <p class="font-medium text-sm truncate">{{ model.display_name }}</p>
+                        <p class="text-xs text-muted-foreground truncate font-mono">{{ model.name }}</p>
+                      </div>
+                    </div>
+                  </div>
+                </div>
+
+                <!-- 从提供商获取的模型折叠组 -->
+                <div
+                  v-for="group in upstreamModelGroups"
+                  :key="group.api_format"
+                  class="border rounded-lg overflow-hidden"
+                >
+                  <div class="flex items-center gap-2 px-3 py-2 bg-muted/30">
+                    <button
+                      type="button"
+                      class="flex items-center gap-2 flex-1 hover:bg-muted/50 -mx-1 px-1 rounded transition-colors"
+                      @click="toggleGroupCollapse(group.api_format)"
+                    >
+                      <ChevronDown
+                        class="w-4 h-4 transition-transform shrink-0"
+                        :class="collapsedGroups.has(group.api_format) ? '-rotate-90' : ''"
+                      />
+                      <span class="text-xs font-medium">
+                        {{ API_FORMAT_LABELS[group.api_format] || group.api_format }}
+                      </span>
+                      <span class="text-xs text-muted-foreground">
+                        ({{ group.models.length }})
+                      </span>
+                    </button>
+                    <button
+                      type="button"
+                      class="text-xs text-primary hover:underline shrink-0"
+                      @click.stop="selectAllUpstreamModels(group.api_format)"
+                    >
+                      {{ isUpstreamGroupAllSelected(group.api_format) ? '取消' : '全选' }}
+                    </button>
+                  </div>
+                  <div
+                    v-show="!collapsedGroups.has(group.api_format)"
+                    class="p-2 space-y-1 border-t"
+                  >
+                    <div
+                      v-for="model in group.models"
+                      :key="model.id"
+                      class="flex items-center gap-2 p-2 rounded-lg border transition-colors cursor-pointer"
+                      :class="selectedLeftIds.includes(model.id)
+                        ? 'border-primary bg-primary/10'
+                        : 'hover:bg-muted/50'"
+                      @click="toggleLeftSelection(model.id)"
+                    >
+                      <Checkbox
+                        :checked="selectedLeftIds.includes(model.id)"
+                        @update:checked="toggleLeftSelection(model.id)"
+                        @click.stop
+                      />
+                      <div class="flex-1 min-w-0">
+                        <p class="font-medium text-sm truncate">{{ model.id }}</p>
+                        <p class="text-xs text-muted-foreground truncate font-mono">
+                          {{ model.owned_by || model.id }}
+                        </p>
+                      </div>
+                    </div>
+                  </div>
+                </div>
+              </div>
+            </div>
+          </div>
+
+          <!-- 中间：操作按钮 -->
+          <div class="flex flex-col items-center justify-center w-12 shrink-0 gap-2">
+            <Button
+              variant="outline"
+              size="sm"
+              class="w-9 h-8"
+              :class="selectedLeftIds.length > 0 ? 'border-primary' : ''"
+              :disabled="selectedLeftIds.length === 0"
+              title="添加选中"
+              @click="addSelected"
+            >
+              <ChevronRight
+                class="w-6 h-6 stroke-[3]"
+                :class="selectedLeftIds.length > 0 ? 'text-primary' : ''"
+              />
+            </Button>
+            <Button
+              variant="outline"
+              size="sm"
+              class="w-9 h-8"
+              :class="selectedRightIds.length > 0 ? 'border-primary' : ''"
+              :disabled="selectedRightIds.length === 0"
+              title="移除选中"
+              @click="removeSelected"
+            >
+              <ChevronLeft
+                class="w-6 h-6 stroke-[3]"
+                :class="selectedRightIds.length > 0 ? 'text-primary' : ''"
+              />
+            </Button>
+          </div>
+
+          <!-- 右侧：已添加的允许模型 -->
+          <div class="flex-1 space-y-2">
+            <div class="flex items-center justify-between">
+              <p class="text-sm font-medium">已添加</p>
+              <Button
+                v-if="allowedModels.length > 0"
+                variant="ghost"
+                size="sm"
+                class="h-6 px-2 text-xs"
+                @click="toggleSelectAllRight"
+              >
+                {{ isAllRightSelected ? '取消' : '全选' }}
+              </Button>
+            </div>
+            <div class="border rounded-lg h-80 overflow-y-auto">
+              <div
+                v-if="allowedModels.length === 0"
+                class="flex flex-col items-center justify-center h-full text-muted-foreground"
+              >
+                <Shield class="w-10 h-10 mb-2 opacity-30" />
+                <p class="text-sm">允许访问全部模型</p>
+                <p class="text-xs mt-1">添加模型以限制访问范围</p>
+              </div>
+              <div v-else class="p-2 space-y-1">
+                <div
+                  v-for="modelName in allowedModels"
+                  :key="'allowed-' + modelName"
+                  class="flex items-center gap-2 p-2 rounded-lg border transition-colors cursor-pointer"
+                  :class="selectedRightIds.includes(modelName)
+                    ? 'border-primary bg-primary/10'
+                    : 'hover:bg-muted/50'"
+                  @click="toggleRightSelection(modelName)"
+                >
+                  <Checkbox
+                    :checked="selectedRightIds.includes(modelName)"
+                    @update:checked="toggleRightSelection(modelName)"
+                    @click.stop
+                  />
+                  <div class="flex-1 min-w-0">
+                    <p class="font-medium text-sm truncate">
+                      {{ getModelDisplayName(modelName) }}
+                    </p>
+                    <p class="text-xs text-muted-foreground truncate font-mono">
+                      {{ modelName }}
+                    </p>
+                  </div>
+                </div>
+              </div>
+            </div>
+          </div>
+        </div>
+      </div>
+    </template>
+
+    <template #footer>
+      <div class="flex items-center justify-between w-full">
+        <p class="text-xs text-muted-foreground">
+          {{ hasChanges ? '有未保存的更改' : '' }}
+        </p>
+        <div class="flex items-center gap-2">
+          <Button variant="outline" @click="handleCancel">取消</Button>
+          <Button :disabled="saving || !hasChanges" @click="handleSave">
+            {{ saving ? '保存中...' : '保存' }}
+          </Button>
+        </div>
+      </div>
+    </template>
+  </Dialog>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, watch, onUnmounted } from 'vue'
+import {
+  Shield,
+  Search,
+  RefreshCw,
+  Loader2,
+  Zap,
+  ChevronRight,
+  ChevronLeft,
+  ChevronDown
+} from 'lucide-vue-next'
+import { Dialog, Button, Input, Checkbox, Badge } from '@/components/ui'
+import { useToast } from '@/composables/useToast'
+import { parseApiError } from '@/utils/errorParser'
+import {
+  updateProviderKey,
+  API_FORMAT_LABELS,
+  type EndpointAPIKey,
+  type AllowedModels,
+} from '@/api/endpoints'
+import { getGlobalModels, type GlobalModelResponse } from '@/api/global-models'
+import { adminApi } from '@/api/admin'
+import type { UpstreamModel } from '@/api/endpoints/types'
+
+interface AvailableModel {
+  name: string
+  display_name: string
+}
+
+const props = defineProps<{
+  open: boolean
+  apiKey: EndpointAPIKey | null
+  providerId: string
+}>()
+
+const emit = defineEmits<{
+  close: []
+  saved: []
+}>()
+
+const { success, error: showError } = useToast()
+
+const isOpen = computed(() => props.open)
+const saving = ref(false)
+const loadingGlobalModels = ref(false)
+const fetchingUpstreamModels = ref(false)
+const upstreamModelsLoaded = ref(false)
+
+// 用于取消异步操作的标志
+let loadingCancelled = false
+
+// 搜索
+const searchQuery = ref('')
+
+// 折叠状态
+const collapsedGroups = ref<Set<string>>(new Set())
+
+// 可用模型列表（全局模型）
+const allGlobalModels = ref<AvailableModel[]>([])
+// 上游模型列表
+const upstreamModels = ref<UpstreamModel[]>([])
+
+// 已添加的允许模型（右侧）
+const allowedModels = ref<string[]>([])
+const initialAllowedModels = ref<string[]>([])
+
+// 选中状态
+const selectedLeftIds = ref<string[]>([])
+const selectedRightIds = ref<string[]>([])
+
+// 是否有更改
+const hasChanges = computed(() => {
+  if (allowedModels.value.length !== initialAllowedModels.value.length) return true
+  const sorted1 = [...allowedModels.value].sort()
+  const sorted2 = [...initialAllowedModels.value].sort()
+  return sorted1.some((v, i) => v !== sorted2[i])
+})
+
+// 计算可添加的全局模型（排除已添加的）
+const availableGlobalModelsBase = computed(() => {
+  const allowedSet = new Set(allowedModels.value)
+  return allGlobalModels.value.filter(m => !allowedSet.has(m.name))
+})
+
+// 搜索过滤后的全局模型
+const availableGlobalModels = computed(() => {
+  if (!searchQuery.value.trim()) return availableGlobalModelsBase.value
+  const query = searchQuery.value.toLowerCase()
+  return availableGlobalModelsBase.value.filter(m =>
+    m.name.toLowerCase().includes(query) ||
+    m.display_name.toLowerCase().includes(query)
+  )
+})
+
+// 计算可添加的上游模型（排除已添加的）
+const availableUpstreamModelsBase = computed(() => {
+  const allowedSet = new Set(allowedModels.value)
+  return upstreamModels.value.filter(m => !allowedSet.has(m.id))
+})
+
+// 搜索过滤后的上游模型
+const availableUpstreamModels = computed(() => {
+  if (!searchQuery.value.trim()) return availableUpstreamModelsBase.value
+  const query = searchQuery.value.toLowerCase()
+  return availableUpstreamModelsBase.value.filter(m =>
+    m.id.toLowerCase().includes(query) ||
+    (m.owned_by && m.owned_by.toLowerCase().includes(query))
+  )
+})
+
+// 按 API 格式分组的上游模型
+const upstreamModelGroups = computed(() => {
+  const groups: Record<string, UpstreamModel[]> = {}
+  for (const model of availableUpstreamModels.value) {
+    const format = model.api_format || 'unknown'
+    if (!groups[format]) groups[format] = []
+    groups[format].push(model)
+  }
+  const order = Object.keys(API_FORMAT_LABELS)
+  return Object.entries(groups)
+    .map(([api_format, models]) => ({ api_format, models }))
+    .sort((a, b) => {
+      const aIndex = order.indexOf(a.api_format)
+      const bIndex = order.indexOf(b.api_format)
+      if (aIndex === -1 && bIndex === -1) return a.api_format.localeCompare(b.api_format)
+      if (aIndex === -1) return 1
+      if (bIndex === -1) return -1
+      return aIndex - bIndex
+    })
+})
+
+// 总可添加数量
+const totalAvailableCount = computed(() => {
+  return availableGlobalModels.value.length + availableUpstreamModels.value.length
+})
+
+// 右侧全选状态
+const isAllRightSelected = computed(() =>
+  allowedModels.value.length > 0 &&
+  selectedRightIds.value.length === allowedModels.value.length
+)
+
+// 全局模型是否全选
+const isAllGlobalModelsSelected = computed(() => {
+  if (availableGlobalModels.value.length === 0) return false
+  return availableGlobalModels.value.every(m => selectedLeftIds.value.includes(m.name))
+})
+
+// 检查某个上游组是否全选
+function isUpstreamGroupAllSelected(apiFormat: string): boolean {
+  const group = upstreamModelGroups.value.find(g => g.api_format === apiFormat)
+  if (!group || group.models.length === 0) return false
+  return group.models.every(m => selectedLeftIds.value.includes(m.id))
+}
+
+// 获取模型显示名称
+function getModelDisplayName(name: string): string {
+  const globalModel = allGlobalModels.value.find(m => m.name === name)
+  if (globalModel) return globalModel.display_name
+  const upstreamModel = upstreamModels.value.find(m => m.id === name)
+  if (upstreamModel) return upstreamModel.id
+  return name
+}
+
+// 加载全局模型
+async function loadGlobalModels() {
+  loadingGlobalModels.value = true
+  try {
+    const response = await getGlobalModels({ limit: 1000 })
+    // 检查是否已取消（dialog 已关闭）
+    if (loadingCancelled) return
+    allGlobalModels.value = response.models.map((m: GlobalModelResponse) => ({
+      name: m.name,
+      display_name: m.display_name
+    }))
+  } catch (err) {
+    if (loadingCancelled) return
+    showError('加载全局模型失败', '错误')
+  } finally {
+    loadingGlobalModels.value = false
+  }
+}
+
+// 从提供商获取模型（使用当前 key）
+async function fetchUpstreamModels() {
+  if (!props.providerId || !props.apiKey) return
+  try {
+    fetchingUpstreamModels.value = true
+    // 使用当前 key 的 ID 来查询上游模型
+    const response = await adminApi.queryProviderModels(props.providerId, props.apiKey.id)
+    // 检查是否已取消
+    if (loadingCancelled) return
+    if (response.success && response.data?.models) {
+      upstreamModels.value = response.data.models
+      upstreamModelsLoaded.value = true
+      const allGroups = new Set(['global'])
+      for (const model of response.data.models) {
+        if (model.api_format) allGroups.add(model.api_format)
+      }
+      collapsedGroups.value = allGroups
+    } else {
+      showError(response.data?.error || '获取上游模型失败', '错误')
+    }
+  } catch (err: any) {
+    if (loadingCancelled) return
+    showError(err.response?.data?.detail || '获取上游模型失败', '错误')
+  } finally {
+    fetchingUpstreamModels.value = false
+  }
+}
+
+// 切换折叠状态
+function toggleGroupCollapse(group: string) {
+  if (collapsedGroups.value.has(group)) {
+    collapsedGroups.value.delete(group)
+  } else {
+    collapsedGroups.value.add(group)
+  }
+  collapsedGroups.value = new Set(collapsedGroups.value)
+}
+
+// 是否为字典模式（按 API 格式区分）
+const isDictMode = ref(false)
+
+// 解析 allowed_models
+function parseAllowedModels(allowed: AllowedModels): string[] {
+  if (allowed === null || allowed === undefined) {
+    isDictMode.value = false
+    return []
+  }
+  if (Array.isArray(allowed)) {
+    isDictMode.value = false
+    return [...allowed]
+  }
+  // 字典模式：合并所有格式的模型，并设置警告标志
+  isDictMode.value = true
+  const all = new Set<string>()
+  for (const models of Object.values(allowed)) {
+    models.forEach(m => all.add(m))
+  }
+  return Array.from(all)
+}
+
+// 左侧选择
+function toggleLeftSelection(name: string) {
+  const idx = selectedLeftIds.value.indexOf(name)
+  if (idx === -1) {
+    selectedLeftIds.value.push(name)
+  } else {
+    selectedLeftIds.value.splice(idx, 1)
+  }
+}
+
+// 右侧选择
+function toggleRightSelection(name: string) {
+  const idx = selectedRightIds.value.indexOf(name)
+  if (idx === -1) {
+    selectedRightIds.value.push(name)
+  } else {
+    selectedRightIds.value.splice(idx, 1)
+  }
+}
+
+// 右侧全选切换
+function toggleSelectAllRight() {
+  if (isAllRightSelected.value) {
+    selectedRightIds.value = []
+  } else {
+    selectedRightIds.value = [...allowedModels.value]
+  }
+}
+
+// 全选全局模型
+function selectAllGlobalModels() {
+  const allNames = availableGlobalModels.value.map(m => m.name)
+  const allSelected = allNames.every(name => selectedLeftIds.value.includes(name))
+  if (allSelected) {
+    selectedLeftIds.value = selectedLeftIds.value.filter(id => !allNames.includes(id))
+  } else {
+    const newNames = allNames.filter(name => !selectedLeftIds.value.includes(name))
+    selectedLeftIds.value.push(...newNames)
+  }
+}
+
+// 全选某个 API 格式的上游模型
+function selectAllUpstreamModels(apiFormat: string) {
+  const group = upstreamModelGroups.value.find(g => g.api_format === apiFormat)
+  if (!group) return
+  const allIds = group.models.map(m => m.id)
+  const allSelected = allIds.every(id => selectedLeftIds.value.includes(id))
+  if (allSelected) {
+    selectedLeftIds.value = selectedLeftIds.value.filter(id => !allIds.includes(id))
+  } else {
+    const newIds = allIds.filter(id => !selectedLeftIds.value.includes(id))
+    selectedLeftIds.value.push(...newIds)
+  }
+}
+
+// 添加选中的模型到右侧
+function addSelected() {
+  for (const name of selectedLeftIds.value) {
+    if (!allowedModels.value.includes(name)) {
+      allowedModels.value.push(name)
+    }
+  }
+  selectedLeftIds.value = []
+}
+
+// 从右侧移除选中的模型
+function removeSelected() {
+  allowedModels.value = allowedModels.value.filter(
+    name => !selectedRightIds.value.includes(name)
+  )
+  selectedRightIds.value = []
+}
+
+// 监听对话框打开
+watch(() => props.open, async (open) => {
+  if (open && props.apiKey) {
+    // 重置取消标志
+    loadingCancelled = false
+
+    const parsed = parseAllowedModels(props.apiKey.allowed_models ?? null)
+    allowedModels.value = [...parsed]
+    initialAllowedModels.value = [...parsed]
+    selectedLeftIds.value = []
+    selectedRightIds.value = []
+    searchQuery.value = ''
+    upstreamModels.value = []
+    upstreamModelsLoaded.value = false
+    collapsedGroups.value = new Set()
+
+    await loadGlobalModels()
+  } else {
+    // dialog 关闭时设置取消标志
+    loadingCancelled = true
+  }
+})
+
+// 组件卸载时取消所有异步操作
+onUnmounted(() => {
+  loadingCancelled = true
+})
+
+function handleDialogUpdate(value: boolean) {
+  if (!value) emit('close')
+}
+
+function handleCancel() {
+  emit('close')
+}
+
+async function handleSave() {
+  if (!props.apiKey) return
+
+  saving.value = true
+  try {
+    // 空列表 = null（允许全部）
+    const newAllowed: AllowedModels = allowedModels.value.length > 0
+      ? [...allowedModels.value]
+      : null
+
+    await updateProviderKey(props.apiKey.id, { allowed_models: newAllowed })
+    success('模型权限已更新', '成功')
+    emit('saved')
+    emit('close')
+  } catch (err: any) {
+    showError(parseApiError(err, '保存失败'), '错误')
+  } finally {
+    saving.value = false
+  }
+}
+</script>
--- a/frontend/src/features/providers/components/KeyFormDialog.vue
+++ b/frontend/src/features/providers/components/KeyFormDialog.vue
@@ -2,57 +2,36 @@
  <Dialog
    :model-value="isOpen"
    :title="isEditMode ? '编辑密钥' : '添加密钥'"
-    :description="isEditMode ? '修改 API 密钥配置' : '为端点添加新的 API 密钥'"
+    :description="isEditMode ? '修改 API 密钥配置' : '为提供商添加新的 API 密钥'"
    :icon="isEditMode ? SquarePen : Key"
    size="2xl"
    @update:model-value="handleDialogUpdate"
  >
    <form
-      class="space-y-5"
+      class="space-y-4"
      autocomplete="off"
      @submit.prevent="handleSave"
    >
      <!-- 基本信息 -->
-      <div class="space-y-3">
-        <h3 class="text-sm font-medium border-b pb-2">
-          基本信息
-        </h3>
-        <div class="grid grid-cols-2 gap-4">
-          <div>
-            <Label :for="keyNameInputId">密钥名称 *</Label>
-            <Input
-              :id="keyNameInputId"
-              v-model="form.name"
-              :name="keyNameFieldName"
-              required
-              placeholder="例如：主 Key、备用 Key 1"
-              maxlength="100"
-              autocomplete="off"
-              autocapitalize="none"
-              autocorrect="off"
-              spellcheck="false"
-              data-form-type="other"
-              data-lpignore="true"
-              data-1p-ignore="true"
-            />
-          </div>
-          <div>
-            <Label for="rate_multiplier">成本倍率 *</Label>
-            <Input
-              id="rate_multiplier"
-              v-model.number="form.rate_multiplier"
-              type="number"
-              step="0.01"
-              min="0.01"
-              required
-              placeholder="1.0"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              真实成本 = 表面成本 × 倍率
-            </p>
-          </div>
+      <div class="grid grid-cols-2 gap-3">
+        <div>
+          <Label :for="keyNameInputId">密钥名称 *</Label>
+          <Input
+            :id="keyNameInputId"
+            v-model="form.name"
+            :name="keyNameFieldName"
+            required
+            placeholder="例如：主 Key、备用 Key 1"
+            maxlength="100"
+            autocomplete="off"
+            autocapitalize="none"
+            autocorrect="off"
+            spellcheck="false"
+            data-form-type="other"
+            data-lpignore="true"
+            data-1p-ignore="true"
+          />
        </div>
-
        <div>
          <Label :for="apiKeyInputId">API 密钥 {{ editingKey ? '' : '*' }}</Label>
          <Input
@@ -83,148 +62,161 @@
            v-else-if="editingKey"
            class="text-xs text-muted-foreground mt-1"
          >
-            留空表示不修改，输入新值则覆盖
+            留空表示不修改
          </p>
        </div>
+      </div>

+      <!-- 备注 -->
+      <div>
+        <Label for="note">备注</Label>
+        <Input
+          id="note"
+          v-model="form.note"
+          placeholder="可选的备注信息"
+        />
+      </div>
+
+      <!-- API 格式选择 -->
+      <div v-if="sortedApiFormats.length > 0">
+        <Label class="mb-1.5 block">支持的 API 格式 *</Label>
+        <div class="grid grid-cols-2 sm:grid-cols-3 gap-2">
+          <div
+            v-for="format in sortedApiFormats"
+            :key="format"
+            class="flex items-center justify-between rounded-md border px-2 py-1.5 transition-colors cursor-pointer"
+            :class="form.api_formats.includes(format)
+              ? 'bg-primary/5 border-primary/30'
+              : 'bg-muted/30 border-border hover:border-muted-foreground/30'"
+            @click="toggleApiFormat(format)"
+          >
+            <div class="flex items-center gap-1.5 min-w-0">
+              <span
+                class="w-4 h-4 rounded border flex items-center justify-center text-xs shrink-0"
+                :class="form.api_formats.includes(format)
+                  ? 'bg-primary border-primary text-primary-foreground'
+                  : 'border-muted-foreground/30'"
+              >
+                <span v-if="form.api_formats.includes(format)">✓</span>
+              </span>
+              <span
+                class="text-sm whitespace-nowrap"
+                :class="form.api_formats.includes(format) ? 'text-primary' : 'text-muted-foreground'"
+              >{{ API_FORMAT_LABELS[format] || format }}</span>
+            </div>
+            <div
+              class="flex items-center shrink-0 ml-2 text-xs text-muted-foreground gap-1"
+              @click.stop
+            >
+              <span>×</span>
+              <input
+                :value="form.rate_multipliers[format] ?? ''"
+                type="number"
+                step="0.01"
+                min="0.01"
+                placeholder="1"
+                class="w-9 bg-transparent text-right outline-none [appearance:textfield] [&::-webkit-outer-spin-button]:appearance-none [&::-webkit-inner-spin-button]:appearance-none"
+                :class="form.api_formats.includes(format) ? 'text-primary' : 'text-muted-foreground'"
+                title="成本倍率"
+                @input="(e) => updateRateMultiplier(format, (e.target as HTMLInputElement).value)"
+              >
+            </div>
+          </div>
+        </div>
+      </div>
+
+      <!-- 配置项 -->
+      <div class="grid grid-cols-4 gap-3">
        <div>
-          <Label for="note">备注</Label>
+          <Label
+            for="internal_priority"
+            class="text-xs"
+          >优先级</Label>
          <Input
-            id="note"
-            v-model="form.note"
-            placeholder="可选的备注信息"
+            id="internal_priority"
+            v-model.number="form.internal_priority"
+            type="number"
+            min="0"
+            class="h-8"
          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            越小越优先
+          </p>
+        </div>
+        <div>
+          <Label
+            for="rpm_limit"
+            class="text-xs"
+          >RPM 限制</Label>
+          <Input
+            id="rpm_limit"
+            :model-value="form.rpm_limit ?? ''"
+            type="number"
+            min="1"
+            max="10000"
+            placeholder="自适应"
+            class="h-8"
+            @update:model-value="(v) => form.rpm_limit = parseNullableNumberInput(v, { min: 1, max: 10000 })"
+          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            留空自适应
+          </p>
+        </div>
+        <div>
+          <Label
+            for="cache_ttl_minutes"
+            class="text-xs"
+          >缓存 TTL</Label>
+          <Input
+            id="cache_ttl_minutes"
+            :model-value="form.cache_ttl_minutes ?? ''"
+            type="number"
+            min="0"
+            max="60"
+            class="h-8"
+            @update:model-value="(v) => form.cache_ttl_minutes = parseNumberInput(v, { min: 0, max: 60 }) ?? 5"
+          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            分钟，0禁用
+          </p>
+        </div>
+        <div>
+          <Label
+            for="max_probe_interval_minutes"
+            class="text-xs"
+          >熔断探测</Label>
+          <Input
+            id="max_probe_interval_minutes"
+            :model-value="form.max_probe_interval_minutes ?? ''"
+            type="number"
+            min="2"
+            max="32"
+            placeholder="32"
+            class="h-8"
+            @update:model-value="(v) => form.max_probe_interval_minutes = parseNumberInput(v, { min: 2, max: 32 }) ?? 32"
+          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            分钟，2-32
+          </p>
        </div>
      </div>

-      <!-- 调度与限流 -->
-      <div class="space-y-3">
-        <h3 class="text-sm font-medium border-b pb-2">
-          调度与限流
-        </h3>
-        <div class="grid grid-cols-2 gap-4">
-          <div>
-            <Label for="internal_priority">内部优先级</Label>
-            <Input
-              id="internal_priority"
-              v-model.number="form.internal_priority"
-              type="number"
-              min="0"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              数字越小越优先
-            </p>
-          </div>
-          <div>
-            <Label for="max_concurrent">最大并发</Label>
-            <Input
-              id="max_concurrent"
-              :model-value="form.max_concurrent ?? ''"
-              type="number"
-              min="1"
-              placeholder="留空启用自适应"
-              @update:model-value="(v) => form.max_concurrent = parseNumberInput(v)"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              留空 = 自适应模式
-            </p>
-          </div>
-        </div>
-
-        <div class="grid grid-cols-3 gap-4">
-          <div>
-            <Label for="rate_limit">速率限制(/分钟)</Label>
-            <Input
-              id="rate_limit"
-              :model-value="form.rate_limit ?? ''"
-              type="number"
-              min="1"
-              @update:model-value="(v) => form.rate_limit = parseNumberInput(v)"
-            />
-          </div>
-          <div>
-            <Label for="daily_limit">每日限制</Label>
-            <Input
-              id="daily_limit"
-              :model-value="form.daily_limit ?? ''"
-              type="number"
-              min="1"
-              @update:model-value="(v) => form.daily_limit = parseNumberInput(v)"
-            />
-          </div>
-          <div>
-            <Label for="monthly_limit">每月限制</Label>
-            <Input
-              id="monthly_limit"
-              :model-value="form.monthly_limit ?? ''"
-              type="number"
-              min="1"
-              @update:model-value="(v) => form.monthly_limit = parseNumberInput(v)"
-            />
-          </div>
-        </div>
-      </div>
-
-      <!-- 缓存与熔断 -->
-      <div class="space-y-3">
-        <h3 class="text-sm font-medium border-b pb-2">
-          缓存与熔断
-        </h3>
-        <div class="grid grid-cols-2 gap-4">
-          <div>
-            <Label for="cache_ttl_minutes">缓存 TTL (分钟)</Label>
-            <Input
-              id="cache_ttl_minutes"
-              :model-value="form.cache_ttl_minutes ?? ''"
-              type="number"
-              min="0"
-              max="60"
-              @update:model-value="(v) => form.cache_ttl_minutes = parseNumberInput(v, { min: 0, max: 60 }) ?? 5"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              0 = 禁用缓存亲和性
-            </p>
-          </div>
-          <div>
-            <Label for="max_probe_interval_minutes">熔断探测间隔 (分钟)</Label>
-            <Input
-              id="max_probe_interval_minutes"
-              :model-value="form.max_probe_interval_minutes ?? ''"
-              type="number"
-              min="2"
-              max="32"
-              placeholder="32"
-              @update:model-value="(v) => form.max_probe_interval_minutes = parseNumberInput(v, { min: 2, max: 32 }) ?? 32"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              范围 2-32 分钟
-            </p>
-          </div>
-        </div>
-      </div>
-
-      <!-- 能力标签配置 -->
-      <div
-        v-if="availableCapabilities.length > 0"
-        class="space-y-3"
-      >
-        <h3 class="text-sm font-medium border-b pb-2">
-          能力标签
-        </h3>
-        <div class="flex flex-wrap gap-2">
-          <label
+      <!-- 能力标签 -->
+      <div v-if="availableCapabilities.length > 0">
+        <Label class="text-xs mb-1.5 block">能力标签</Label>
+        <div class="flex flex-wrap gap-1.5">
+          <button
            v-for="cap in availableCapabilities"
            :key="cap.name"
-            class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm"
+            type="button"
+            class="inline-flex items-center gap-1.5 px-2.5 py-1 rounded-md border text-sm transition-colors"
+            :class="form.capabilities[cap.name]
+              ? 'bg-primary/10 border-primary/50 text-primary'
+              : 'bg-card border-border hover:bg-muted/50 text-muted-foreground'"
+            @click="form.capabilities[cap.name] = !form.capabilities[cap.name]"
          >
-            <input
-              type="checkbox"
-              :checked="form.capabilities[cap.name] || false"
-              class="rounded"
-              @change="form.capabilities[cap.name] = !form.capabilities[cap.name]"
-            >
-            <span>{{ cap.display_name }}</span>
-          </label>
+            {{ cap.display_name }}
+          </button>
        </div>
      </div>
    </form>
@@ -240,25 +232,27 @@
        :disabled="saving"
        @click="handleSave"
      >
-        {{ saving ? '保存中...' : '保存' }}
+        {{ saving ? (isEditMode ? '保存中...' : '添加中...') : (isEditMode ? '保存' : '添加') }}
      </Button>
    </template>
  </Dialog>
 </template>

 <script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
+import { ref, computed, onMounted, watch } from 'vue'
 import { Dialog, Button, Input, Label } from '@/components/ui'
 import { Key, SquarePen } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
 import { useFormDialog } from '@/composables/useFormDialog'
 import { parseApiError } from '@/utils/errorParser'
-import { parseNumberInput } from '@/utils/form'
+import { parseNumberInput, parseNullableNumberInput } from '@/utils/form'
 import { log } from '@/utils/logger'
 import {
-  addEndpointKey,
-  updateEndpointKey,
+  addProviderKey,
+  updateProviderKey,
  getAllCapabilities,
+  API_FORMAT_LABELS,
+  sortApiFormats,
  type EndpointAPIKey,
  type EndpointAPIKeyUpdate,
  type ProviderEndpoint,
@@ -270,6 +264,7 @@ const props = defineProps<{
  endpoint: ProviderEndpoint | null
  editingKey: EndpointAPIKey | null
  providerId: string | null
+  availableApiFormats: string[]  // Provider 支持的所有 API 格式
 }>()

 const emit = defineEmits<{
@@ -279,6 +274,9 @@ const emit = defineEmits<{

 const { success, error: showError } = useToast()

+// 排序后的可用 API 格式列表
+const sortedApiFormats = computed(() => sortApiFormats(props.availableApiFormats))
+
 const isOpen = computed(() => props.open)
 const saving = ref(false)
 const formNonce = ref(createFieldNonce())
@@ -297,12 +295,10 @@ const availableCapabilities = ref<CapabilityDefinition[]>([])
 const form = ref({
  name: '',
  api_key: '',
-  rate_multiplier: 1.0,
-  internal_priority: 50,
-  max_concurrent: undefined as number | undefined,
-  rate_limit: undefined as number | undefined,
-  daily_limit: undefined as number | undefined,
-  monthly_limit: undefined as number | undefined,
+  api_formats: [] as string[],  // 支持的 API 格式列表
+  rate_multipliers: {} as Record<string, number>,  // 按 API 格式的成本倍率
+  internal_priority: 10,
+  rpm_limit: undefined as number | null | undefined,  // RPM 限制（null=自适应，undefined=保持原值）
  cache_ttl_minutes: 5,
  max_probe_interval_minutes: 32,
  note: '',
@@ -323,6 +319,43 @@ onMounted(() => {
  loadCapabilities()
 })

+// API 格式切换
+function toggleApiFormat(format: string) {
+  const index = form.value.api_formats.indexOf(format)
+  if (index === -1) {
+    // 添加格式
+    form.value.api_formats.push(format)
+  } else {
+    // 移除格式前检查：至少保留一个格式
+    if (form.value.api_formats.length <= 1) {
+      showError('至少需要选择一个 API 格式', '验证失败')
+      return
+    }
+    // 移除格式，但保留倍率配置（用户可能只是临时取消）
+    form.value.api_formats.splice(index, 1)
+  }
+}
+
+// 更新指定格式的成本倍率
+function updateRateMultiplier(format: string, value: string | number) {
+  // 使用对象替换以确保 Vue 3 响应性
+  const newMultipliers = { ...form.value.rate_multipliers }
+
+  if (value === '' || value === null || value === undefined) {
+    // 清空时删除该格式的配置（使用默认倍率）
+    delete newMultipliers[format]
+  } else {
+    const numValue = typeof value === 'string' ? parseFloat(value) : value
+    // 限制倍率范围：0.01 - 100
+    if (!isNaN(numValue) && numValue >= 0.01 && numValue <= 100) {
+      newMultipliers[format] = numValue
+    }
+  }
+
+  // 替换整个对象以触发响应式更新
+  form.value.rate_multipliers = newMultipliers
+}
+
 // API 密钥输入框样式计算
 function getApiKeyInputClass(): string {
  const classes = []
@@ -363,12 +396,10 @@ function resetForm() {
  form.value = {
    name: '',
    api_key: '',
-    rate_multiplier: 1.0,
-    internal_priority: 50,
-    max_concurrent: undefined,
-    rate_limit: undefined,
-    daily_limit: undefined,
-    monthly_limit: undefined,
+    api_formats: [],  // 默认不选中任何格式
+    rate_multipliers: {},
+    internal_priority: 10,
+    rpm_limit: undefined,
    cache_ttl_minutes: 5,
    max_probe_interval_minutes: 32,
    note: '',
@@ -377,6 +408,14 @@ function resetForm() {
  }
 }

+// 添加成功后清除部分字段以便继续添加
+function clearForNextAdd() {
+  formNonce.value = createFieldNonce()
+  apiKeyFocused.value = false
+  form.value.name = ''
+  form.value.api_key = ''
+}
+
 // 加载密钥数据（编辑模式）
 function loadKeyData() {
  if (!props.editingKey) return
@@ -385,13 +424,13 @@ function loadKeyData() {
  form.value = {
    name: props.editingKey.name,
    api_key: '',
-    rate_multiplier: props.editingKey.rate_multiplier || 1.0,
-    internal_priority: props.editingKey.internal_priority ?? 50,
+    api_formats: props.editingKey.api_formats?.length > 0
+      ? [...props.editingKey.api_formats]
+      : [],  // 编辑模式下保持原有选择，不默认全选
+    rate_multipliers: { ...(props.editingKey.rate_multipliers || {}) },
+    internal_priority: props.editingKey.internal_priority ?? 10,
    // 保留原始的 null/undefined 状态，null 表示自适应模式
-    max_concurrent: props.editingKey.max_concurrent ?? undefined,
-    rate_limit: props.editingKey.rate_limit ?? undefined,
-    daily_limit: props.editingKey.daily_limit ?? undefined,
-    monthly_limit: props.editingKey.monthly_limit ?? undefined,
+    rpm_limit: props.editingKey.rpm_limit ?? undefined,
    cache_ttl_minutes: props.editingKey.cache_ttl_minutes ?? 5,
    max_probe_interval_minutes: props.editingKey.max_probe_interval_minutes ?? 32,
    note: props.editingKey.note || '',
@@ -415,7 +454,11 @@ function createFieldNonce(): string {
 }

 async function handleSave() {
-  if (!props.endpoint) return
+  // 必须有 providerId
+  if (!props.providerId) {
+    showError('无法保存：缺少提供商信息', '错误')
+    return
+  }

  // 提交前验证
  if (apiKeyError.value) {
@@ -429,6 +472,12 @@ async function handleSave() {
    return
  }

+  // 验证至少选择一个 API 格式
+  if (form.value.api_formats.length === 0) {
+    showError('请至少选择一个 API 格式', '验证失败')
+    return
+  }
+
  // 过滤出有效的能力配置（只包含值为 true 的）
  const activeCapabilities: Record<string, boolean> = {}
  for (const [key, value] of Object.entries(form.value.capabilities)) {
@@ -440,21 +489,27 @@ async function handleSave() {

  saving.value = true
  try {
+    // 准备 rate_multipliers 数据：只保留已选中格式的倍率配置
+    const filteredMultipliers: Record<string, number> = {}
+    for (const format of form.value.api_formats) {
+      if (form.value.rate_multipliers[format] !== undefined) {
+        filteredMultipliers[format] = form.value.rate_multipliers[format]
+      }
+    }
+    const rateMultipliersData = Object.keys(filteredMultipliers).length > 0
+      ? filteredMultipliers
+      : null
+
    if (props.editingKey) {
      // 更新模式
-      // 注意：max_concurrent 需要显式发送 null 来切换到自适应模式
-      // undefined 会在 JSON 中被忽略，所以用 null 表示"清空/自适应"
+      // 注意：rpm_limit 使用 null 表示自适应模式
+      // undefined 表示"保持原值不变"（会在 JSON 序列化时被忽略）
      const updateData: EndpointAPIKeyUpdate = {
+        api_formats: form.value.api_formats,
        name: form.value.name,
-        rate_multiplier: form.value.rate_multiplier,
+        rate_multipliers: rateMultipliersData,
        internal_priority: form.value.internal_priority,
-        // 显式使用 null 表示自适应模式，这样后端能区分"未提供"和"设置为 null"
-        // 注意：只有 max_concurrent 需要这种处理，因为它有"自适应模式"的概念
-        // 其他限制字段（rate_limit 等）不支持"清空"操作，undefined 会被 JSON 忽略即不更新
-        max_concurrent: form.value.max_concurrent === undefined ? null : form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        daily_limit: form.value.daily_limit,
-        monthly_limit: form.value.monthly_limit,
+        rpm_limit: form.value.rpm_limit,
        cache_ttl_minutes: form.value.cache_ttl_minutes,
        max_probe_interval_minutes: form.value.max_probe_interval_minutes,
        note: form.value.note,
@@ -466,26 +521,27 @@ async function handleSave() {
        updateData.api_key = form.value.api_key
      }

-      await updateEndpointKey(props.editingKey.id, updateData)
+      await updateProviderKey(props.editingKey.id, updateData)
      success('密钥已更新', '成功')
    } else {
-      // 新增
-      await addEndpointKey(props.endpoint.id, {
-        endpoint_id: props.endpoint.id,
+      // 新增模式
+      await addProviderKey(props.providerId, {
+        api_formats: form.value.api_formats,
        api_key: form.value.api_key,
        name: form.value.name,
-        rate_multiplier: form.value.rate_multiplier,
+        rate_multipliers: rateMultipliersData,
        internal_priority: form.value.internal_priority,
-        max_concurrent: form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        daily_limit: form.value.daily_limit,
-        monthly_limit: form.value.monthly_limit,
+        rpm_limit: form.value.rpm_limit,
        cache_ttl_minutes: form.value.cache_ttl_minutes,
        max_probe_interval_minutes: form.value.max_probe_interval_minutes,
        note: form.value.note,
        capabilities: capabilitiesData || undefined
      })
      success('密钥已添加', '成功')
+      // 添加模式：不关闭对话框，只清除名称和密钥以便继续添加
+      emit('saved')
+      clearForNextAdd()
+      return
    }

    emit('saved')
--- a/frontend/src/features/providers/components/PriorityManagementDialog.vue
+++ b/frontend/src/features/providers/components/PriorityManagementDialog.vue
@@ -95,7 +95,7 @@

              <!-- 提供商信息 -->
              <div class="flex-1 min-w-0 flex items-center gap-2">
-                <span class="font-medium text-sm truncate">{{ provider.display_name }}</span>
+                <span class="font-medium text-sm truncate">{{ provider.name }}</span>
                <Badge
                  v-if="!provider.is_active"
                  variant="secondary"
@@ -395,7 +395,7 @@ import { Dialog } from '@/components/ui'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import { useToast } from '@/composables/useToast'
-import { updateProvider, updateEndpointKey } from '@/api/endpoints'
+import { updateProvider, updateProviderKey } from '@/api/endpoints'
 import type { ProviderWithEndpointsSummary } from '@/api/endpoints'
 import { adminApi } from '@/api/admin'

@@ -696,7 +696,7 @@ async function save() {
      const keys = keysByFormat.value[format]
      keys.forEach((key) => {
        // 使用用户设置的 priority 值，相同 priority 会做负载均衡
-        keyUpdates.push(updateEndpointKey(key.id, { global_priority: key.priority }))
+        keyUpdates.push(updateProviderKey(key.id, { global_priority: key.priority }))
      })
    }

--- a/frontend/src/features/providers/components/ProviderDetailDrawer.vue
+++ b/frontend/src/features/providers/components/ProviderDetailDrawer.vue
--- a/frontend/src/features/providers/components/ProviderFormDialog.vue
+++ b/frontend/src/features/providers/components/ProviderFormDialog.vue
@@ -4,47 +4,29 @@
    :title="isEditMode ? '编辑提供商' : '添加提供商'"
    :description="isEditMode ? '更新提供商配置。API 端点和密钥需在详情页面单独管理。' : '创建新的提供商配置。创建后可以为其添加 API 端点和密钥。'"
    :icon="isEditMode ? SquarePen : Server"
-    size="2xl"
+    size="xl"
    @update:model-value="handleDialogUpdate"
  >
    <form
-      class="space-y-6"
+      class="space-y-5"
      @submit.prevent="handleSubmit"
    >
      <!-- 基本信息 -->
-      <div class="space-y-4">
+      <div class="space-y-3">
        <h3 class="text-sm font-medium border-b pb-2">
          基本信息
        </h3>

-        <!-- 添加模式显示提供商标识 -->
-        <div
-          v-if="!isEditMode"
-          class="space-y-2"
-        >
-          <Label for="name">提供商标识 *</Label>
-          <Input
-            id="name"
-            v-model="form.name"
-            placeholder="例如: openai-primary"
-            required
-          />
-          <p class="text-xs text-muted-foreground">
-            唯一ID，创建后不可修改
-          </p>
-        </div>
-
        <div class="grid grid-cols-2 gap-4">
-          <div class="space-y-2">
-            <Label for="display_name">显示名称 *</Label>
+          <div class="space-y-1.5">
+            <Label for="name">名称 *</Label>
            <Input
-              id="display_name"
-              v-model="form.display_name"
+              id="name"
+              v-model="form.name"
              placeholder="例如: OpenAI 主账号"
-              required
            />
          </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
            <Label for="website">主站链接</Label>
            <Input
              id="website"
@@ -55,24 +37,28 @@
          </div>
        </div>

-        <div class="space-y-2">
+        <div class="space-y-1.5">
          <Label for="description">描述</Label>
-          <Textarea
+          <Input
            id="description"
            v-model="form.description"
            placeholder="提供商描述（可选）"
-            rows="2"
          />
        </div>
      </div>

-      <!-- 计费与限流 -->
-      <div class="space-y-4">
-        <h3 class="text-sm font-medium border-b pb-2">
-          计费与限流
-        </h3>
+      <!-- 计费与限流 / 请求配置 -->
+      <div class="space-y-3">
        <div class="grid grid-cols-2 gap-4">
-          <div class="space-y-2">
+          <h3 class="text-sm font-medium border-b pb-2">
+            计费与限流
+          </h3>
+          <h3 class="text-sm font-medium border-b pb-2">
+            请求配置
+          </h3>
+        </div>
+        <div class="grid grid-cols-2 gap-4">
+          <div class="space-y-1.5">
            <Label>计费类型</Label>
            <Select
              v-model="form.billing_type"
@@ -82,27 +68,35 @@
                <SelectValue />
              </SelectTrigger>
              <SelectContent>
-                <SelectItem value="monthly_quota">
-                  月卡额度
-                </SelectItem>
-                <SelectItem value="pay_as_you_go">
-                  按量付费
-                </SelectItem>
-                <SelectItem value="free_tier">
-                  免费套餐
-                </SelectItem>
+                <SelectItem value="monthly_quota">月卡额度</SelectItem>
+                <SelectItem value="pay_as_you_go">按量付费</SelectItem>
+                <SelectItem value="free_tier">免费套餐</SelectItem>
              </SelectContent>
            </Select>
          </div>
-          <div class="space-y-2">
-            <Label>RPM 限制</Label>
-            <Input
-              :model-value="form.rpm_limit ?? ''"
-              type="number"
-              min="0"
-              placeholder="不限制请留空"
-              @update:model-value="(v) => form.rpm_limit = parseNumberInput(v)"
-            />
+          <div class="grid grid-cols-2 gap-4">
+            <div class="space-y-1.5">
+              <Label>超时时间 (秒)</Label>
+              <Input
+                :model-value="form.timeout ?? ''"
+                type="number"
+                min="1"
+                max="600"
+                placeholder="默认 300"
+                @update:model-value="(v) => form.timeout = parseNumberInput(v)"
+              />
+            </div>
+            <div class="space-y-1.5">
+              <Label>最大重试次数</Label>
+              <Input
+                :model-value="form.max_retries ?? ''"
+                type="number"
+                min="0"
+                max="10"
+                placeholder="默认 2"
+                @update:model-value="(v) => form.max_retries = parseNumberInput(v)"
+              />
+            </div>
          </div>
        </div>

@@ -111,52 +105,94 @@
          v-if="form.billing_type === 'monthly_quota'"
          class="grid grid-cols-2 gap-4 p-3 border rounded-lg bg-muted/50"
        >
-          <div class="space-y-2">
+          <div class="space-y-1.5">
            <Label class="text-xs">周期额度 (USD)</Label>
            <Input
              :model-value="form.monthly_quota_usd ?? ''"
              type="number"
              step="0.01"
              min="0"
-              class="h-9"
              @update:model-value="(v) => form.monthly_quota_usd = parseNumberInput(v, { allowFloat: true })"
            />
          </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
            <Label class="text-xs">重置周期 (天)</Label>
            <Input
              :model-value="form.quota_reset_day ?? ''"
              type="number"
              min="1"
              max="365"
-              class="h-9"
              @update:model-value="(v) => form.quota_reset_day = parseNumberInput(v) ?? 30"
            />
          </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
            <Label class="text-xs">
-              周期开始时间
-              <span class="text-red-500">*</span>
+              周期开始时间 <span class="text-red-500">*</span>
            </Label>
            <Input
              v-model="form.quota_last_reset_at"
              type="datetime-local"
-              class="h-9"
            />
-            <p class="text-xs text-muted-foreground">
-              系统会自动统计从该时间点开始的使用量
-            </p>
          </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
            <Label class="text-xs">过期时间</Label>
            <Input
              v-model="form.quota_expires_at"
              type="datetime-local"
-              class="h-9"
            />
-            <p class="text-xs text-muted-foreground">
-              留空表示永久有效
-            </p>
+          </div>
+        </div>
+      </div>
+
+      <!-- 代理配置 -->
+      <div class="space-y-3">
+        <div class="flex items-center justify-between">
+          <h3 class="text-sm font-medium">
+            代理配置
+          </h3>
+          <div class="flex items-center gap-2">
+            <Switch
+              :model-value="form.proxy_enabled"
+              @update:model-value="(v: boolean) => form.proxy_enabled = v"
+            />
+            <span class="text-sm text-muted-foreground">启用代理</span>
+          </div>
+        </div>
+        <div
+          v-if="form.proxy_enabled"
+          class="grid grid-cols-2 gap-4 p-3 border rounded-lg bg-muted/50"
+        >
+          <div class="space-y-1.5">
+            <Label class="text-xs">代理地址 *</Label>
+            <Input
+              v-model="form.proxy_url"
+              placeholder="http://proxy:port 或 socks5://proxy:port"
+            />
+          </div>
+          <div class="grid grid-cols-2 gap-3">
+            <div class="space-y-1.5">
+              <Label class="text-xs">用户名</Label>
+              <Input
+                v-model="form.proxy_username"
+                placeholder="可选"
+                autocomplete="off"
+                data-form-type="other"
+                data-lpignore="true"
+                data-1p-ignore="true"
+              />
+            </div>
+            <div class="space-y-1.5">
+              <Label class="text-xs">密码</Label>
+              <Input
+                v-model="form.proxy_password"
+                type="password"
+                placeholder="可选"
+                autocomplete="new-password"
+                data-form-type="other"
+                data-lpignore="true"
+                data-1p-ignore="true"
+              />
+            </div>
          </div>
        </div>
      </div>
@@ -172,7 +208,7 @@
        取消
      </Button>
      <Button
-        :disabled="loading || !form.display_name || (!isEditMode && !form.name)"
+        :disabled="loading || !form.name"
        @click="handleSubmit"
      >
        {{ loading ? (isEditMode ? '保存中...' : '创建中...') : (isEditMode ? '保存' : '创建') }}
@@ -187,13 +223,13 @@ import {
  Dialog,
  Button,
  Input,
-  Textarea,
  Label,
  Select,
  SelectTrigger,
  SelectValue,
  SelectContent,
  SelectItem,
+  Switch,
 } from '@/components/ui'
 import { Server, SquarePen } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
@@ -223,7 +259,6 @@ const internalOpen = computed(() => props.modelValue)
 // 表单数据
 const form = ref({
  name: '',
-  display_name: '',
  description: '',
  website: '',
  // 计费配置
@@ -232,19 +267,25 @@ const form = ref({
  quota_reset_day: 30,
  quota_last_reset_at: '',  // 周期开始时间
  quota_expires_at: '',
-  rpm_limit: undefined as string | number | undefined,
  provider_priority: 999,
  // 状态配置
  is_active: true,
  rate_limit: undefined as number | undefined,
  concurrent_limit: undefined as number | undefined,
+  // 请求配置
+  timeout: undefined as number | undefined,
+  max_retries: undefined as number | undefined,
+  // 代理配置（扁平化便于表单绑定）
+  proxy_enabled: false,
+  proxy_url: '',
+  proxy_username: '',
+  proxy_password: '',
 })

 // 重置表单
 function resetForm() {
  form.value = {
    name: '',
-    display_name: '',
    description: '',
    website: '',
    billing_type: 'pay_as_you_go',
@@ -252,11 +293,18 @@ function resetForm() {
    quota_reset_day: 30,
    quota_last_reset_at: '',
    quota_expires_at: '',
-    rpm_limit: undefined,
    provider_priority: 999,
    is_active: true,
    rate_limit: undefined,
    concurrent_limit: undefined,
+    // 请求配置
+    timeout: undefined,
+    max_retries: undefined,
+    // 代理配置
+    proxy_enabled: false,
+    proxy_url: '',
+    proxy_username: '',
+    proxy_password: '',
  }
 }

@@ -264,9 +312,9 @@ function resetForm() {
 function loadProviderData() {
  if (!props.provider) return

+  const proxy = props.provider.proxy
  form.value = {
    name: props.provider.name,
-    display_name: props.provider.display_name,
    description: props.provider.description || '',
    website: props.provider.website || '',
    billing_type: (props.provider.billing_type as 'monthly_quota' | 'pay_as_you_go' | 'free_tier') || 'pay_as_you_go',
@@ -276,11 +324,18 @@ function loadProviderData() {
      new Date(props.provider.quota_last_reset_at).toISOString().slice(0, 16) : '',
    quota_expires_at: props.provider.quota_expires_at ?
      new Date(props.provider.quota_expires_at).toISOString().slice(0, 16) : '',
-    rpm_limit: props.provider.rpm_limit ?? undefined,
    provider_priority: props.provider.provider_priority || 999,
    is_active: props.provider.is_active,
    rate_limit: undefined,
    concurrent_limit: undefined,
+    // 请求配置
+    timeout: props.provider.timeout ?? undefined,
+    max_retries: props.provider.max_retries ?? undefined,
+    // 代理配置
+    proxy_enabled: proxy?.enabled ?? false,
+    proxy_url: proxy?.url || '',
+    proxy_username: proxy?.username || '',
+    proxy_password: proxy?.password || '',
  }
 }

@@ -302,17 +357,37 @@ const handleSubmit = async () => {
    return
  }

+  // 启用代理时必须填写代理地址
+  if (form.value.proxy_enabled && !form.value.proxy_url) {
+    showError('启用代理时必须填写代理地址', '验证失败')
+    return
+  }
+
  loading.value = true
  try {
+    // 构建代理配置
+    const proxy = form.value.proxy_enabled ? {
+      url: form.value.proxy_url,
+      username: form.value.proxy_username || undefined,
+      password: form.value.proxy_password || undefined,
+      enabled: true,
+    } : null
+
    const payload = {
-      ...form.value,
-      rpm_limit:
-        form.value.rpm_limit === undefined || form.value.rpm_limit === ''
-          ? null
-          : Number(form.value.rpm_limit),
-      // 空字符串时不发送
+      name: form.value.name,
+      description: form.value.description || undefined,
+      website: form.value.website || undefined,
+      billing_type: form.value.billing_type,
+      monthly_quota_usd: form.value.monthly_quota_usd,
+      quota_reset_day: form.value.quota_reset_day,
      quota_last_reset_at: form.value.quota_last_reset_at || undefined,
      quota_expires_at: form.value.quota_expires_at || undefined,
+      provider_priority: form.value.provider_priority,
+      is_active: form.value.is_active,
+      // 请求配置
+      timeout: form.value.timeout ?? undefined,
+      max_retries: form.value.max_retries ?? undefined,
+      proxy,
    }

    if (isEditMode.value && props.provider) {
--- a/frontend/src/features/providers/components/index.ts
+++ b/frontend/src/features/providers/components/index.ts
@@ -2,6 +2,7 @@ export { default as ProviderFormDialog } from './ProviderFormDialog.vue'
 export { default as EndpointFormDialog } from './EndpointFormDialog.vue'
 export { default as KeyFormDialog } from './KeyFormDialog.vue'
 export { default as KeyAllowedModelsDialog } from './KeyAllowedModelsDialog.vue'
+export { default as KeyAllowedModelsEditDialog } from './KeyAllowedModelsEditDialog.vue'
 export { default as PriorityManagementDialog } from './PriorityManagementDialog.vue'
 export { default as ProviderModelFormDialog } from './ProviderModelFormDialog.vue'
 export { default as ProviderDetailDrawer } from './ProviderDetailDrawer.vue'
--- a/frontend/src/features/providers/components/provider-tabs/ModelsTab.vue
+++ b/frontend/src/features/providers/components/provider-tabs/ModelsTab.vue
@@ -178,7 +178,7 @@
                <Button
                  variant="ghost"
                  size="icon"
-                  class="h-8 w-8 text-destructive hover:text-destructive"
+                  class="h-8 w-8 hover:text-destructive"
                  title="删除"
                  @click="deleteModel(model)"
                >
--- a/frontend/src/features/usage/components/RequestDetailDrawer.vue
+++ b/frontend/src/features/usage/components/RequestDetailDrawer.vue
@@ -289,14 +289,14 @@
                />
              </div>

-              <!-- 错误信息卡片 -->
+              <!-- 响应客户端错误卡片 -->
              <Card
                v-if="detail.error_message"
                class="border-red-200 dark:border-red-800"
              >
                <div class="p-4">
                  <h4 class="text-sm font-semibold text-red-600 dark:text-red-400 mb-2">
-                    错误信息
+                    响应客户端错误
                  </h4>
                  <div class="bg-red-50 dark:bg-red-900/20 rounded-lg p-3">
                    <p class="text-sm text-red-800 dark:text-red-300">
@@ -431,7 +431,7 @@

                    <TabsContent value="response-headers">
                      <JsonContent
-                        :data="detail.response_headers"
+                        :data="actualResponseHeaders"
                        :view-mode="viewMode"
                        :expand-depth="currentExpandDepth"
                        :is-dark="isDark"
@@ -614,6 +614,25 @@ const tabs = [
  { name: 'metadata', label: '元数据' },
 ]

+// 判断数据是否有实际内容（非空对象/数组）
+function hasContent(data: unknown): boolean {
+  if (data === null || data === undefined) return false
+  if (typeof data === 'object') {
+    return Object.keys(data as object).length > 0
+  }
+  return true
+}
+
+// 获取实际的响应头（优先 client_response_headers，回退到 response_headers）
+const actualResponseHeaders = computed(() => {
+  if (!detail.value) return null
+  // 优先返回客户端响应头，如果没有则回退到提供商响应头
+  if (hasContent(detail.value.client_response_headers)) {
+    return detail.value.client_response_headers
+  }
+  return detail.value.response_headers
+})
+
 // 根据实际数据决定显示哪些 Tab
 const visibleTabs = computed(() => {
  if (!detail.value) return []
@@ -621,15 +640,15 @@ const visibleTabs = computed(() => {
  return tabs.filter(tab => {
    switch (tab.name) {
      case 'request-headers':
-        return detail.value!.request_headers && Object.keys(detail.value!.request_headers).length > 0
+        return hasContent(detail.value!.request_headers)
      case 'request-body':
-        return detail.value!.request_body !== null && detail.value!.request_body !== undefined
+        return hasContent(detail.value!.request_body)
      case 'response-headers':
-        return detail.value!.response_headers && Object.keys(detail.value!.response_headers).length > 0
+        return hasContent(actualResponseHeaders.value)
      case 'response-body':
-        return detail.value!.response_body !== null && detail.value!.response_body !== undefined
+        return hasContent(detail.value!.response_body)
      case 'metadata':
-        return detail.value!.metadata && Object.keys(detail.value!.metadata).length > 0
+        return hasContent(detail.value!.metadata)
      default:
        return false
    }
@@ -775,7 +794,7 @@ function copyJsonToClipboard(tabName: string) {
      data = detail.value.request_body
      break
    case 'response-headers':
-      data = detail.value.response_headers
+      data = actualResponseHeaders.value
      break
    case 'response-body':
      data = detail.value.response_body
--- a/frontend/src/features/users/components/UserFormDialog.vue
+++ b/frontend/src/features/users/components/UserFormDialog.vue
@@ -252,7 +252,7 @@
                    @click.stop
                    @change="toggleSelection('allowed_providers', provider.id)"
                  >
-                  <span class="text-sm">{{ provider.display_name || provider.name }}</span>
+                  <span class="text-sm">{{ provider.name }}</span>
                </div>
                <div
                  v-if="providers.length === 0"
--- a/frontend/src/mocks/data.ts
+++ b/frontend/src/mocks/data.ts
@@ -424,8 +424,7 @@ export const MOCK_ADMIN_API_KEYS: AdminApiKeysResponse = {
 export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  {
    id: 'provider-001',
-    name: 'duck_coding_free',
-    display_name: 'DuckCodingFree',
+    name: 'DuckCodingFree',
    description: '',
    website: 'https://duckcoding.com',
    provider_priority: 1,
@@ -451,8 +450,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  },
  {
    id: 'provider-002',
-    name: 'open_claude_code',
-    display_name: 'OpenClaudeCode',
+    name: 'OpenClaudeCode',
    description: '',
    website: 'https://www.openclaudecode.cn',
    provider_priority: 2,
@@ -477,8 +475,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  },
  {
    id: 'provider-003',
-    name: '88_code',
-    display_name: '88Code',
+    name: '88Code',
    description: '',
    website: 'https://www.88code.org/',
    provider_priority: 3,
@@ -503,8 +500,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  },
  {
    id: 'provider-004',
-    name: 'ikun_code',
-    display_name: 'IKunCode',
+    name: 'IKunCode',
    description: '',
    website: 'https://api.ikuncode.cc',
    provider_priority: 4,
@@ -531,8 +527,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  },
  {
    id: 'provider-005',
-    name: 'duck_coding',
-    display_name: 'DuckCoding',
+    name: 'DuckCoding',
    description: '',
    website: 'https://duckcoding.com',
    provider_priority: 5,
@@ -561,8 +556,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  },
  {
    id: 'provider-006',
-    name: 'privnode',
-    display_name: 'Privnode',
+    name: 'Privnode',
    description: '',
    website: 'https://privnode.com',
    provider_priority: 6,
@@ -584,8 +578,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
  },
  {
    id: 'provider-007',
-    name: 'undying_api',
-    display_name: 'UndyingAPI',
+    name: 'UndyingAPI',
    description: '',
    website: 'https://vip.undyingapi.com',
    provider_priority: 7,
--- a/frontend/src/mocks/handler.ts
+++ b/frontend/src/mocks/handler.ts
@@ -418,16 +418,16 @@ const MOCK_ALIASES = [

 // Mock Endpoint Keys
 const MOCK_ENDPOINT_KEYS = [
-  { id: 'ekey-001', endpoint_id: 'ep-001', api_key_masked: 'sk-ant...abc1', name: 'Primary Key', rate_multiplier: 1.0, internal_priority: 1, health_score: 98, consecutive_failures: 0, request_count: 5000, success_count: 4950, error_count: 50, success_rate: 99, avg_response_time_ms: 1200, is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ekey-002', endpoint_id: 'ep-001', api_key_masked: 'sk-ant...def2', name: 'Backup Key', rate_multiplier: 1.0, internal_priority: 2, health_score: 95, consecutive_failures: 1, request_count: 2000, success_count: 1950, error_count: 50, success_rate: 97.5, avg_response_time_ms: 1350, is_active: true, created_at: '2024-02-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ekey-003', endpoint_id: 'ep-002', api_key_masked: 'sk-oai...ghi3', name: 'OpenAI Main', rate_multiplier: 1.0, internal_priority: 1, health_score: 97, consecutive_failures: 0, request_count: 3500, success_count: 3450, error_count: 50, success_rate: 98.6, avg_response_time_ms: 900, is_active: true, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
+  { id: 'ekey-001', provider_id: 'provider-001', api_formats: ['CLAUDE'], api_key_masked: 'sk-ant...abc1', name: 'Primary Key', rate_multiplier: 1.0, internal_priority: 1, health_score: 0.98, consecutive_failures: 0, request_count: 5000, success_count: 4950, error_count: 50, success_rate: 0.99, avg_response_time_ms: 1200, cache_ttl_minutes: 5, max_probe_interval_minutes: 32, is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ekey-002', provider_id: 'provider-001', api_formats: ['CLAUDE'], api_key_masked: 'sk-ant...def2', name: 'Backup Key', rate_multiplier: 1.0, internal_priority: 2, health_score: 0.95, consecutive_failures: 1, request_count: 2000, success_count: 1950, error_count: 50, success_rate: 0.975, avg_response_time_ms: 1350, cache_ttl_minutes: 5, max_probe_interval_minutes: 32, is_active: true, created_at: '2024-02-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ekey-003', provider_id: 'provider-002', api_formats: ['OPENAI'], api_key_masked: 'sk-oai...ghi3', name: 'OpenAI Main', rate_multiplier: 1.0, internal_priority: 1, health_score: 0.97, consecutive_failures: 0, request_count: 3500, success_count: 3450, error_count: 50, success_rate: 0.986, avg_response_time_ms: 900, cache_ttl_minutes: 5, max_probe_interval_minutes: 32, is_active: true, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
 ]

 // Mock Endpoints
 const MOCK_ENDPOINTS = [
-  { id: 'ep-001', provider_id: 'provider-001', provider_name: 'anthropic', api_format: 'claude', base_url: 'https://api.anthropic.com', auth_type: 'bearer', timeout: 120, max_retries: 2, priority: 100, weight: 100, health_score: 98, consecutive_failures: 0, is_active: true, total_keys: 2, active_keys: 2, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ep-002', provider_id: 'provider-002', provider_name: 'openai', api_format: 'openai', base_url: 'https://api.openai.com', auth_type: 'bearer', timeout: 60, max_retries: 2, priority: 90, weight: 100, health_score: 97, consecutive_failures: 0, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ep-003', provider_id: 'provider-003', provider_name: 'google', api_format: 'gemini', base_url: 'https://generativelanguage.googleapis.com', auth_type: 'api_key', timeout: 60, max_retries: 2, priority: 80, weight: 100, health_score: 96, consecutive_failures: 0, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
+  { id: 'ep-001', provider_id: 'provider-001', provider_name: 'anthropic', api_format: 'CLAUDE', base_url: 'https://api.anthropic.com', timeout: 300, max_retries: 2, is_active: true, total_keys: 2, active_keys: 2, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ep-002', provider_id: 'provider-002', provider_name: 'openai', api_format: 'OPENAI', base_url: 'https://api.openai.com', timeout: 60, max_retries: 2, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ep-003', provider_id: 'provider-003', provider_name: 'google', api_format: 'GEMINI', base_url: 'https://generativelanguage.googleapis.com', timeout: 60, max_retries: 2, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
 ]

 // Mock 能力定义
@@ -581,7 +581,6 @@ const mockHandlers: Record<string, (config: AxiosRequestConfig) => Promise<Axios
    return createMockResponse(MOCK_PROVIDERS.map(p => ({
      id: p.id,
      name: p.name,
-      display_name: p.display_name,
      is_active: p.is_active
    })))
  },
@@ -1222,13 +1221,8 @@ function generateMockEndpointsForProvider(providerId: string) {
      base_url: format.includes('CLAUDE') ? 'https://api.anthropic.com' :
                format.includes('OPENAI') ? 'https://api.openai.com' :
                'https://generativelanguage.googleapis.com',
-      auth_type: format.includes('GEMINI') ? 'api_key' : 'bearer',
-      timeout: 120,
+      timeout: 300,
      max_retries: 2,
-      priority: 100 - index * 10,
-      weight: 100,
-      health_score: healthDetail?.health_score ?? 1.0,
-      consecutive_failures: healthDetail?.health_score && healthDetail.health_score < 0.7 ? 2 : 0,
      is_active: healthDetail?.is_active ?? true,
      total_keys: Math.ceil(Math.random() * 3) + 1,
      active_keys: Math.ceil(Math.random() * 2) + 1,
@@ -1238,11 +1232,16 @@ function generateMockEndpointsForProvider(providerId: string) {
  })
 }

-// 为 endpoint 生成 keys
-function generateMockKeysForEndpoint(endpointId: string, count: number = 2) {
+// 为 provider 生成 keys（Key 归属 Provider，通过 api_formats 关联）
+const PROVIDER_KEYS_CACHE: Record<string, any[]> = {}
+function generateMockKeysForProvider(providerId: string, count: number = 2) {
+  const provider = MOCK_PROVIDERS.find(p => p.id === providerId)
+  const formats = provider?.api_formats || []
+
  return Array.from({ length: count }, (_, i) => ({
-    id: `key-${endpointId}-${i + 1}`,
-    endpoint_id: endpointId,
+    id: `key-${providerId}-${i + 1}`,
+    provider_id: providerId,
+    api_formats: i === 0 ? formats : formats.slice(0, 1),
    api_key_masked: `sk-***...${Math.random().toString(36).substring(2, 6)}`,
    name: i === 0 ? 'Primary Key' : `Backup Key ${i}`,
    rate_multiplier: 1.0,
@@ -1254,6 +1253,8 @@ function generateMockKeysForEndpoint(endpointId: string, count: number = 2) {
    error_count: Math.floor(Math.random() * 100),
    success_rate: 0.95 + Math.random() * 0.04,  // 0.95-0.99
    avg_response_time_ms: 800 + Math.floor(Math.random() * 600),
+    cache_ttl_minutes: 5,
+    max_probe_interval_minutes: 32,
    is_active: true,
    created_at: '2024-01-01T00:00:00Z',
    updated_at: new Date().toISOString()
@@ -1463,29 +1464,63 @@ registerDynamicRoute('PUT', '/api/admin/endpoints/:endpointId', async (config, p
 registerDynamicRoute('DELETE', '/api/admin/endpoints/:endpointId', async (_config, _params) => {
  await delay()
  requireAdmin()
-  return createMockResponse({ message: '删除成功（演示模式）' })
+  return createMockResponse({ message: '删除成功（演示模式）', affected_keys_count: 0 })
 })

-// Endpoint Keys 列表
-registerDynamicRoute('GET', '/api/admin/endpoints/:endpointId/keys', async (_config, params) => {
+// Provider Keys 列表
+registerDynamicRoute('GET', '/api/admin/endpoints/providers/:providerId/keys', async (_config, params) => {
  await delay()
  requireAdmin()
-  const keys = generateMockKeysForEndpoint(params.endpointId, 2)
-  return createMockResponse(keys)
+  if (!PROVIDER_KEYS_CACHE[params.providerId]) {
+    PROVIDER_KEYS_CACHE[params.providerId] = generateMockKeysForProvider(params.providerId, 2)
+  }
+  return createMockResponse(PROVIDER_KEYS_CACHE[params.providerId])
 })

-// 创建 Key
-registerDynamicRoute('POST', '/api/admin/endpoints/:endpointId/keys', async (config, params) => {
+// 为 Provider 创建 Key
+registerDynamicRoute('POST', '/api/admin/endpoints/providers/:providerId/keys', async (config, params) => {
  await delay()
  requireAdmin()
  const body = JSON.parse(config.data || '{}')
-  return createMockResponse({
+  const apiKeyPlain = body.api_key || 'sk-demo'
+  const masked = apiKeyPlain.length >= 12
+    ? `${apiKeyPlain.slice(0, 8)}***${apiKeyPlain.slice(-4)}`
+    : 'sk-***...demo'
+
+  const newKey = {
    id: `key-demo-${Date.now()}`,
-    endpoint_id: params.endpointId,
-    api_key_masked: 'sk-***...demo',
-    ...body,
-    created_at: new Date().toISOString()
-  })
+    provider_id: params.providerId,
+    api_formats: body.api_formats || [],
+    api_key_masked: masked,
+    api_key_plain: null,
+    name: body.name || 'New Key',
+    note: body.note,
+    rate_multiplier: body.rate_multiplier ?? 1.0,
+    rate_multipliers: body.rate_multipliers ?? null,
+    internal_priority: body.internal_priority ?? 50,
+    global_priority: body.global_priority ?? null,
+    rpm_limit: body.rpm_limit ?? null,
+    allowed_models: body.allowed_models ?? null,
+    capabilities: body.capabilities ?? null,
+    cache_ttl_minutes: body.cache_ttl_minutes ?? 5,
+    max_probe_interval_minutes: body.max_probe_interval_minutes ?? 32,
+    health_score: 1.0,
+    consecutive_failures: 0,
+    request_count: 0,
+    success_count: 0,
+    error_count: 0,
+    success_rate: 0.0,
+    avg_response_time_ms: 0.0,
+    is_active: true,
+    created_at: new Date().toISOString(),
+    updated_at: new Date().toISOString(),
+  }
+
+  if (!PROVIDER_KEYS_CACHE[params.providerId]) {
+    PROVIDER_KEYS_CACHE[params.providerId] = []
+  }
+  PROVIDER_KEYS_CACHE[params.providerId].push(newKey)
+  return createMockResponse(newKey)
 })

 // Key 更新
@@ -1503,6 +1538,50 @@ registerDynamicRoute('DELETE', '/api/admin/endpoints/keys/:keyId', async (_confi
  return createMockResponse({ message: '删除成功（演示模式）' })
 })

+// Key Reveal
+registerDynamicRoute('GET', '/api/admin/endpoints/keys/:keyId/reveal', async (_config, _params) => {
+  await delay()
+  requireAdmin()
+  return createMockResponse({ api_key: 'sk-demo-reveal' })
+})
+
+// Keys grouped by format
+mockHandlers['GET /api/admin/endpoints/keys/grouped-by-format'] = async () => {
+  await delay()
+  requireAdmin()
+
+  // 确保每个 provider 都有 key 数据
+  for (const provider of MOCK_PROVIDERS) {
+    if (!PROVIDER_KEYS_CACHE[provider.id]) {
+      PROVIDER_KEYS_CACHE[provider.id] = generateMockKeysForProvider(provider.id, 2)
+    }
+  }
+
+  const grouped: Record<string, any[]> = {}
+  for (const provider of MOCK_PROVIDERS) {
+    const endpoints = generateMockEndpointsForProvider(provider.id)
+    const baseUrlByFormat = Object.fromEntries(endpoints.map(e => [e.api_format, e.base_url]))
+    const keys = PROVIDER_KEYS_CACHE[provider.id] || []
+    for (const key of keys) {
+      const formats: string[] = key.api_formats || []
+      for (const fmt of formats) {
+        if (!grouped[fmt]) grouped[fmt] = []
+        grouped[fmt].push({
+          ...key,
+          api_format: fmt,
+          provider_name: provider.name,
+          endpoint_base_url: baseUrlByFormat[fmt],
+          global_priority: key.global_priority ?? null,
+          circuit_breaker_open: false,
+          capabilities: [],
+        })
+      }
+    }
+  }
+
+  return createMockResponse(grouped)
+}
+
 // Provider Models 列表
 registerDynamicRoute('GET', '/api/admin/providers/:providerId/models', async (_config, params) => {
  await delay()
--- a/frontend/src/utils/errorParser.ts
+++ b/frontend/src/utils/errorParser.ts
@@ -20,7 +20,7 @@ interface ValidationError {
 const fieldNameMap: Record<string, string> = {
  'api_key': 'API 密钥',
  'priority': '优先级',
-  'max_concurrent': '最大并发',
+  'rpm_limit': 'RPM 限制',
  'rate_limit': '速率限制',
  'daily_limit': '每日限制',
  'monthly_limit': '每月限制',
@@ -44,7 +44,6 @@ const fieldNameMap: Record<string, string> = {
  'monthly_quota_usd': '月度配额',
  'quota_reset_day': '配额重置日',
  'quota_expires_at': '配额过期时间',
-  'rpm_limit': 'RPM 限制',
  'cache_ttl_minutes': '缓存 TTL',
  'max_probe_interval_minutes': '最大探测间隔',
 }
@@ -151,11 +150,18 @@ export function parseApiError(err: unknown, defaultMessage: string = '操作失
    return '无法连接到服务器，请检查网络连接'
  }

-  const detail = err.response?.data?.detail
+  const data = err.response?.data
+
+  // 1. 处理 {error: {type, message}} 格式（ProxyException 返回格式）
+  if (data?.error?.message) {
+    return data.error.message
+  }
+
+  const detail = data?.detail

  // 如果没有 detail 字段
  if (!detail) {
-    return err.response?.data?.message || err.message || defaultMessage
+    return data?.message || err.message || defaultMessage
  }

  // 1. 处理 Pydantic 验证错误（数组格式）
--- a/frontend/src/utils/form.ts
+++ b/frontend/src/utils/form.ts
@@ -54,6 +54,57 @@ export function parseNumberInput(
  return result
 }

+/**
+ * Parse number input value for nullable fields (like rpm_limit)
+ * Returns `null` when empty (to signal "use adaptive/default mode")
+ * Returns `undefined` when not provided (to signal "keep original value")
+ *
+ * @param value - Input value (string or number)
+ * @param options - Parse options
+ * @returns Parsed number, null (for empty/adaptive), or undefined
+ */
+export function parseNullableNumberInput(
+  value: string | number | null | undefined,
+  options: {
+    allowFloat?: boolean
+    min?: number
+    max?: number
+  } = {}
+): number | null | undefined {
+  const { allowFloat = false, min, max } = options
+
+  // Empty string means "null" (adaptive mode)
+  if (value === '') {
+    return null
+  }
+
+  // null/undefined means "keep original value"
+  if (value === null || value === undefined) {
+    return undefined
+  }
+
+  // Parse the value
+  const num = typeof value === 'string'
+    ? (allowFloat ? parseFloat(value) : parseInt(value, 10))
+    : value
+
+  // Handle NaN - treat as null (adaptive mode)
+  if (isNaN(num)) {
+    return null
+  }
+
+  // Apply min/max constraints
+  let result = num
+  if (min !== undefined && result < min) {
+    result = min
+  }
+  if (max !== undefined && result > max) {
+    result = max
+  }
+
+  return result
+}
+
 /**
 * Create a handler function for number input with specific field
 * Useful for creating inline handlers in templates
--- a/frontend/src/views/admin/ModelManagement.vue
+++ b/frontend/src/views/admin/ModelManagement.vue
@@ -530,9 +530,6 @@
                    />
                    <div class="flex-1 min-w-0">
                      <p class="font-medium text-sm truncate">
-                        {{ provider.display_name || provider.name }}
-                      </p>
-                      <p class="text-xs text-muted-foreground truncate">
                        {{ provider.name }}
                      </p>
                    </div>
@@ -645,10 +642,7 @@
                    />
                    <div class="flex-1 min-w-0">
                      <p class="font-medium text-sm truncate">
-                        {{ provider.display_name }}
-                      </p>
-                      <p class="text-xs text-muted-foreground truncate">
-                        {{ provider.identifier }}
+                        {{ provider.name }}
                      </p>
                    </div>
                    <Badge
@@ -679,7 +673,7 @@
    <ProviderModelFormDialog
      :open="editProviderDialogOpen"
      :provider-id="editingProvider?.id || ''"
-      :provider-name="editingProvider?.display_name || ''"
+      :provider-name="editingProvider?.name || ''"
      :editing-model="editingProviderModel"
      @update:open="handleEditProviderDialogUpdate"
      @saved="handleEditProviderSaved"
@@ -939,7 +933,7 @@ async function batchAddSelectedProviders() {
      const errorMessages = result.errors
        .map(e => {
          const provider = providerOptions.value.find(p => p.id === e.provider_id)
-          const providerName = provider?.display_name || provider?.name || e.provider_id
+          const providerName = provider?.name || e.provider_id
          return `${providerName}: ${e.error}`
        })
        .join('\n')
@@ -977,7 +971,7 @@ async function batchRemoveSelectedProviders() {
        await deleteModel(providerId, provider.model_id)
        successCount++
      } catch (err: any) {
-        errors.push(`${provider.display_name}: ${parseApiError(err, '删除失败')}`)
+        errors.push(`${provider.name}: ${parseApiError(err, '删除失败')}`)
      }
    }

@@ -1088,8 +1082,7 @@ async function loadModelProviders(_globalModelId: string) {
    selectedModelProviders.value = response.providers.map(p => ({
      id: p.provider_id,
      model_id: p.model_id,
-      display_name: p.provider_display_name || p.provider_name,
-      identifier: p.provider_name,
+      name: p.provider_name,
      provider_type: 'API',
      target_model: p.target_model,
      is_active: p.is_active,
@@ -1219,7 +1212,7 @@ async function confirmDeleteProviderImplementation(provider: any) {
  }

  const confirmed = await confirmDanger(
-    `确定要删除 ${provider.display_name} 的模型关联吗？\n\n模型: ${provider.target_model}\n\n此操作不可恢复！`,
+    `确定要删除 ${provider.name} 的模型关联吗？\n\n模型: ${provider.target_model}\n\n此操作不可恢复！`,
    '删除关联提供商'
  )
  if (!confirmed) return
@@ -1227,7 +1220,7 @@ async function confirmDeleteProviderImplementation(provider: any) {
  try {
    const { deleteModel } = await import('@/api/endpoints')
    await deleteModel(provider.id, provider.model_id)
-    success(`已删除 ${provider.display_name} 的模型实现`)
+    success(`已删除 ${provider.name} 的模型实现`)
    // 重新加载 Provider 列表
    if (selectedModel.value) {
      await loadModelProviders(selectedModel.value.id)
--- a/frontend/src/views/admin/ProviderManagement.vue
+++ b/frontend/src/views/admin/ProviderManagement.vue
@@ -134,10 +134,7 @@
              @click="handleRowClick($event, provider.id)"
            >
              <TableCell class="py-3.5">
-                <div class="flex flex-col gap-0.5">
-                  <span class="text-sm font-medium text-foreground">{{ provider.display_name }}</span>
-                  <span class="text-xs text-muted-foreground/70 font-mono">{{ provider.name }}</span>
-                </div>
+                <span class="text-sm font-medium text-foreground">{{ provider.name }}</span>
              </TableCell>
              <TableCell class="py-3.5">
                <Badge
@@ -219,17 +216,10 @@
                    >${{ (provider.monthly_used_usd ?? 0).toFixed(2) }}</span> / <span class="font-medium">${{ (provider.monthly_quota_usd ?? 0).toFixed(2) }}</span>
                  </div>
                  <div
-                    v-if="rpmUsage(provider)"
-                    class="flex items-center gap-1"
-                  >
-                    <span class="text-muted-foreground/70">RPM:</span>
-                    <span class="font-medium text-foreground/80">{{ rpmUsage(provider) }}</span>
-                  </div>
-                  <div
-                    v-if="provider.billing_type !== 'monthly_quota' && !rpmUsage(provider)"
+                    v-else
                    class="text-muted-foreground/50"
                  >
-                    无限制
+                    按量付费
                  </div>
                </div>
              </TableCell>
@@ -304,7 +294,7 @@
          <div class="flex items-start justify-between gap-3">
            <div class="flex-1 min-w-0">
              <div class="flex items-center gap-2">
-                <span class="font-medium text-foreground truncate">{{ provider.display_name }}</span>
+                <span class="font-medium text-foreground truncate">{{ provider.name }}</span>
                <Badge
                  :variant="provider.is_active ? 'success' : 'secondary'"
                  class="text-xs shrink-0"
@@ -312,7 +302,6 @@
                  {{ provider.is_active ? '活跃' : '停用' }}
                </Badge>
              </div>
-              <span class="text-xs text-muted-foreground/70 font-mono">{{ provider.name }}</span>
            </div>
            <div
              class="flex items-center gap-0.5 shrink-0"
@@ -383,20 +372,17 @@
            </span>
          </div>

-          <!-- 第四行：配额/限流 -->
+          <!-- 第四行：配额 -->
          <div
-            v-if="provider.billing_type === 'monthly_quota' || rpmUsage(provider)"
+            v-if="provider.billing_type === 'monthly_quota'"
            class="flex items-center gap-3 text-xs text-muted-foreground"
          >
-            <span v-if="provider.billing_type === 'monthly_quota'">
+            <span>
              配额: <span
                class="font-semibold"
                :class="getQuotaUsedColorClass(provider)"
              >${{ (provider.monthly_used_usd ?? 0).toFixed(2) }}</span> / ${{ (provider.monthly_quota_usd ?? 0).toFixed(2) }}
            </span>
-            <span v-if="rpmUsage(provider)">
-              RPM: {{ rpmUsage(provider) }}
-            </span>
          </div>
        </div>
      </div>
@@ -509,7 +495,7 @@ const filteredProviders = computed(() => {
  if (searchQuery.value.trim()) {
    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
    result = result.filter(p => {
-      const searchableText = `${p.display_name} ${p.name}`.toLowerCase()
+      const searchableText = `${p.name}`.toLowerCase()
      return keywords.every(keyword => searchableText.includes(keyword))
    })
  }
@@ -525,7 +511,7 @@ const filteredProviders = computed(() => {
      return a.provider_priority - b.provider_priority
    }
    // 3. 按名称排序
-    return a.display_name.localeCompare(b.display_name)
+    return a.name.localeCompare(b.name)
  })
 })

@@ -586,7 +572,10 @@ function sortEndpoints(endpoints: any[]) {

 // 判断端点是否可用（有 key）
 function isEndpointAvailable(endpoint: any, _provider: ProviderWithEndpointsSummary): boolean {
-  // 检查该端点是否有活跃的密钥
+  // 检查端点是否启用，以及是否有活跃的密钥
+  if (endpoint.is_active === false) {
+    return false
+  }
  return (endpoint.active_keys ?? 0) > 0
 }

@@ -639,21 +628,6 @@ function getQuotaUsedColorClass(provider: ProviderWithEndpointsSummary): string
  return 'text-foreground'
 }

-function rpmUsage(provider: ProviderWithEndpointsSummary): string | null {
-  const rpmLimit = provider.rpm_limit
-  const rpmUsed = provider.rpm_used ?? 0
-
-  if (rpmLimit === null || rpmLimit === undefined) {
-    return rpmUsed > 0 ? `${rpmUsed}` : null
-  }
-
-  if (rpmLimit === 0) {
-    return '已完全禁止'
-  }
-
-  return `${rpmUsed} / ${rpmLimit}`
-}
-
 // 使用复用的行点击逻辑
 const { handleMouseDown, shouldTriggerRowClick } = useRowClick()

@@ -706,7 +680,7 @@ function handleProviderAdded() {
 async function handleDeleteProvider(provider: ProviderWithEndpointsSummary) {
  const confirmed = await confirmDanger(
    '删除提供商',
-    `确定要删除提供商 "${provider.display_name}" 吗？\n\n这将同时删除其所有端点、密钥和配置。此操作不可恢复！`
+    `确定要删除提供商 "${provider.name}" 吗？\n\n这将同时删除其所有端点、密钥和配置。此操作不可恢复！`
  )

  if (!confirmed) return
--- a/frontend/src/views/admin/SystemSettings.vue
+++ b/frontend/src/views/admin/SystemSettings.vue
@@ -511,7 +511,7 @@
              端点: {{ importPreview.providers?.reduce((sum: number, p: any) => sum + (p.endpoints?.length || 0), 0) }} 个
            </li>
            <li>
-              API Keys: {{ importPreview.providers?.reduce((sum: number, p: any) => sum + p.endpoints?.reduce((s: number, e: any) => s + (e.keys?.length || 0), 0), 0) }} 个
+              API Keys: {{ importPreview.providers?.reduce((sum: number, p: any) => sum + (p.api_keys?.length || 0), 0) }} 个
            </li>
          </ul>
        </div>
@@ -1144,7 +1144,7 @@ function handleConfigFileSelect(event: Event) {
      const data = JSON.parse(content) as ConfigExportData

      // 验证版本
-      if (data.version !== '1.0') {
+      if (data.version !== '2.0') {
        error(`不支持的配置版本: ${data.version}`)
        return
      }
--- a/frontend/src/views/shared/Dashboard.vue
+++ b/frontend/src/views/shared/Dashboard.vue
@@ -16,7 +16,8 @@

        <!-- 主要统计卡片 -->
        <div class="grid grid-cols-2 gap-3 sm:gap-4 xl:grid-cols-4">
-          <template v-if="loading && stats.length === 0">
+          <!-- 加载中骨架屏 -->
+          <template v-if="loading">
            <Card
              v-for="i in 4"
              :key="'skeleton-' + i"
@@ -27,62 +28,98 @@
              <Skeleton class="h-4 w-16" />
            </Card>
          </template>
-          <Card
-            v-for="(stat, index) in stats"
-            v-else
-            :key="stat.name"
-            class="relative overflow-hidden p-3 sm:p-5"
-            :class="statCardBorders[index % statCardBorders.length]"
-          >
-            <div
-              class="pointer-events-none absolute -right-4 -top-6 h-28 w-28 rounded-full blur-3xl opacity-40"
-              :class="statCardGlows[index % statCardGlows.length]"
-            />
-            <!-- 图标固定在右上角 -->
-            <div
-              class="absolute top-3 right-3 sm:top-5 sm:right-5 rounded-xl sm:rounded-2xl border border-border bg-card/50 p-2 sm:p-3 shadow-inner backdrop-blur-sm"
-              :class="getStatIconColor(index)"
+          <!-- 有数据时显示统计卡片 -->
+          <template v-else-if="stats.length > 0">
+            <Card
+              v-for="(stat, index) in stats"
+              :key="stat.name"
+              class="relative overflow-hidden p-3 sm:p-5"
+              :class="statCardBorders[index % statCardBorders.length]"
            >
-              <component
-                :is="stat.icon"
-                class="h-4 w-4 sm:h-5 sm:w-5"
-              />
-            </div>
-            <!-- 内容区域 -->
-            <div>
-              <p class="text-[9px] sm:text-[11px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.4em] text-muted-foreground pr-10 sm:pr-14">
-                {{ stat.name }}
-              </p>
-              <p class="mt-2 sm:mt-4 text-xl sm:text-3xl font-semibold text-foreground">
-                {{ stat.value }}
-              </p>
-              <p
-                v-if="stat.subValue"
-                class="mt-0.5 sm:mt-1 text-[10px] sm:text-sm text-muted-foreground"
-              >
-                {{ stat.subValue }}
-              </p>
              <div
-                v-if="stat.change || stat.extraBadge"
-                class="mt-1.5 sm:mt-2 flex items-center gap-1 sm:gap-1.5 flex-wrap"
+                class="pointer-events-none absolute -right-4 -top-6 h-28 w-28 rounded-full blur-3xl opacity-40"
+                :class="statCardGlows[index % statCardGlows.length]"
+              />
+              <!-- 图标固定在右上角 -->
+              <div
+                class="absolute top-3 right-3 sm:top-5 sm:right-5 rounded-xl sm:rounded-2xl border border-border bg-card/50 p-2 sm:p-3 shadow-inner backdrop-blur-sm"
+                :class="getStatIconColor(index)"
              >
-                <Badge
-                  v-if="stat.change"
-                  variant="secondary"
-                  class="text-[9px] sm:text-xs"
-                >
-                  {{ stat.change }}
-                </Badge>
-                <Badge
-                  v-if="stat.extraBadge"
-                  variant="secondary"
-                  class="text-[9px] sm:text-xs"
-                >
-                  {{ stat.extraBadge }}
-                </Badge>
+                <component
+                  :is="stat.icon"
+                  class="h-4 w-4 sm:h-5 sm:w-5"
+                />
              </div>
-            </div>
-          </Card>
+              <!-- 内容区域 -->
+              <div>
+                <p class="text-[9px] sm:text-[11px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.4em] text-muted-foreground pr-10 sm:pr-14">
+                  {{ stat.name }}
+                </p>
+                <p class="mt-2 sm:mt-4 text-xl sm:text-3xl font-semibold text-foreground">
+                  {{ stat.value }}
+                </p>
+                <p
+                  v-if="stat.subValue"
+                  class="mt-0.5 sm:mt-1 text-[10px] sm:text-sm text-muted-foreground"
+                >
+                  {{ stat.subValue }}
+                </p>
+                <div
+                  v-if="stat.change || stat.extraBadge"
+                  class="mt-1.5 sm:mt-2 flex items-center gap-1 sm:gap-1.5 flex-wrap"
+                >
+                  <Badge
+                    v-if="stat.change"
+                    variant="secondary"
+                    class="text-[9px] sm:text-xs"
+                  >
+                    {{ stat.change }}
+                  </Badge>
+                  <Badge
+                    v-if="stat.extraBadge"
+                    variant="secondary"
+                    class="text-[9px] sm:text-xs"
+                  >
+                    {{ stat.extraBadge }}
+                  </Badge>
+                </div>
+              </div>
+            </Card>
+          </template>
+          <!-- 无数据时显示占位卡片 -->
+          <template v-else>
+            <Card
+              v-for="(placeholder, index) in emptyStatPlaceholders"
+              :key="'empty-' + index"
+              class="relative overflow-hidden p-3 sm:p-5"
+              :class="statCardBorders[index % statCardBorders.length]"
+            >
+              <div
+                class="pointer-events-none absolute -right-4 -top-6 h-28 w-28 rounded-full blur-3xl opacity-20"
+                :class="statCardGlows[index % statCardGlows.length]"
+              />
+              <div
+                class="absolute top-3 right-3 sm:top-5 sm:right-5 rounded-xl sm:rounded-2xl border border-border bg-card/50 p-2 sm:p-3 shadow-inner backdrop-blur-sm"
+                :class="getStatIconColor(index)"
+              >
+                <component
+                  :is="placeholder.icon"
+                  class="h-4 w-4 sm:h-5 sm:w-5"
+                />
+              </div>
+              <div>
+                <p class="text-[9px] sm:text-[11px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.4em] text-muted-foreground pr-10 sm:pr-14">
+                  {{ placeholder.name }}
+                </p>
+                <p class="mt-2 sm:mt-4 text-xl sm:text-3xl font-semibold text-muted-foreground/50">
+                  --
+                </p>
+                <p class="mt-0.5 sm:mt-1 text-[10px] sm:text-sm text-muted-foreground/50">
+                  暂无数据
+                </p>
+              </div>
+            </Card>
+          </template>
        </div>

        <!-- 管理员：系统健康摘要 -->
@@ -872,6 +909,24 @@ const iconMap: Record<string, any> = {
  Users, Activity, TrendingUp, DollarSign, Key, Hash, Database
 }

+// 空状态占位卡片
+const emptyStatPlaceholders = computed(() => {
+  if (isAdmin.value) {
+    return [
+      { name: '今日请求', icon: Activity },
+      { name: '今日 Tokens', icon: Hash },
+      { name: '活跃用户', icon: Users },
+      { name: '今日费用', icon: DollarSign }
+    ]
+  }
+  return [
+    { name: '今日请求', icon: Activity },
+    { name: '今日 Tokens', icon: Hash },
+    { name: 'API Keys', icon: Key },
+    { name: '今日费用', icon: DollarSign }
+  ]
+})
+
 const totalStats = computed(() => {
  if (dailyStats.value.length === 0) {
    return { requests: 0, tokens: 0, cost: 0, avgResponseTime: 0 }
--- a/frontend/tailwind.config.js
+++ b/frontend/tailwind.config.js
@@ -78,6 +78,20 @@ export default {
        md: "calc(var(--radius) - 2px)",
        sm: "calc(var(--radius) - 4px)",
      },
+      keyframes: {
+        "collapsible-down": {
+          from: { height: "0" },
+          to: { height: "var(--radix-collapsible-content-height)" },
+        },
+        "collapsible-up": {
+          from: { height: "var(--radix-collapsible-content-height)" },
+          to: { height: "0" },
+        },
+      },
+      animation: {
+        "collapsible-down": "collapsible-down 0.2s ease-out",
+        "collapsible-up": "collapsible-up 0.2s ease-out",
+      },
    },
  },
  plugins: [require("tailwindcss-animate")],
--- a/src/api/admin/adaptive.py
+++ b/src/api/admin/adaptive.py
@@ -1,12 +1,12 @@
 """
-自适应并发管理 API 端点
+自适应 RPM 管理 API 端点

 设计原则：
- 自适应模式由 max_concurrent 字段决定：
-  - max_concurrent = NULL：启用自适应模式，系统自动学习并调整并发限制
-  - max_concurrent = 数字：固定限制模式，使用用户指定的并发限制
- learned_max_concurrent：自适应模式下学习到的并发限制值
- adaptive_mode 是计算字段，基于 max_concurrent 是否为 NULL
+- 自适应模式由 rpm_limit 字段决定：
+  - rpm_limit = NULL：启用自适应模式，系统自动学习并调整 RPM 限制
+  - rpm_limit = 数字：固定限制模式，使用用户指定的 RPM 限制
+- learned_rpm_limit：自适应模式下学习到的 RPM 限制值
+- adaptive_mode 是计算字段，基于 rpm_limit 是否为 NULL
 """

 from dataclasses import dataclass
@@ -18,12 +18,13 @@ from sqlalchemy.orm import Session

 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import RPMDefaults
 from src.core.exceptions import InvalidRequestException, translate_pydantic_error
 from src.database import get_db
 from src.models.database import ProviderAPIKey
-from src.services.rate_limit.adaptive_concurrency import get_adaptive_manager
+from src.services.rate_limit.adaptive_rpm import get_adaptive_rpm_manager

-router = APIRouter(prefix="/api/admin/adaptive", tags=["Adaptive Concurrency"])
+router = APIRouter(prefix="/api/admin/adaptive", tags=["Adaptive RPM"])
 pipeline = ApiRequestPipeline()


@@ -35,19 +36,19 @@ class EnableAdaptiveRequest(BaseModel):

    enabled: bool = Field(..., description="是否启用自适应模式（true=自适应，false=固定限制）")
    fixed_limit: Optional[int] = Field(
-        None, ge=1, le=100, description="固定并发限制（仅当 enabled=false 时生效）"
+        None, ge=1, le=100, description="固定 RPM 限制（仅当 enabled=false 时生效，1-100）"
    )


 class AdaptiveStatsResponse(BaseModel):
    """自适应统计响应"""

-    adaptive_mode: bool = Field(..., description="是否为自适应模式（max_concurrent=NULL）")
-    max_concurrent: Optional[int] = Field(None, description="用户配置的固定限制（NULL=自适应）")
+    adaptive_mode: bool = Field(..., description="是否为自适应模式（rpm_limit=NULL）")
+    rpm_limit: Optional[int] = Field(None, description="用户配置的固定限制（NULL=自适应）")
    effective_limit: Optional[int] = Field(
        None, description="当前有效限制（自适应使用学习值，固定使用配置值）"
    )
-    learned_limit: Optional[int] = Field(None, description="学习到的并发限制")
+    learned_limit: Optional[int] = Field(None, description="学习到的 RPM 限制")
    concurrent_429_count: int
    rpm_429_count: int
    last_429_at: Optional[str]
@@ -61,11 +62,12 @@ class KeyListItem(BaseModel):

    id: str
    name: Optional[str]
-    endpoint_id: str
-    is_adaptive: bool = Field(..., description="是否为自适应模式（max_concurrent=NULL）")
-    max_concurrent: Optional[int] = Field(None, description="固定并发限制（NULL=自适应）")
+    provider_id: str
+    api_formats: List[str] = Field(default_factory=list)
+    is_adaptive: bool = Field(..., description="是否为自适应模式（rpm_limit=NULL）")
+    rpm_limit: Optional[int] = Field(None, description="固定 RPM 限制（NULL=自适应）")
    effective_limit: Optional[int] = Field(None, description="当前有效限制")
-    learned_max_concurrent: Optional[int] = Field(None, description="学习到的并发限制")
+    learned_rpm_limit: Optional[int] = Field(None, description="学习到的 RPM 限制")
    concurrent_429_count: int
    rpm_429_count: int

@@ -80,22 +82,22 @@ class KeyListItem(BaseModel):
 )
 async def list_adaptive_keys(
    request: Request,
-    endpoint_id: Optional[str] = Query(None, description="按 Endpoint 过滤"),
+    provider_id: Optional[str] = Query(None, description="按 Provider 过滤"),
    db: Session = Depends(get_db),
 ):
    """
    获取所有启用自适应模式的Key列表

    可选参数：
-    - endpoint_id: 按 Endpoint 过滤
+    - provider_id: 按 Provider 过滤
    """
-    adapter = ListAdaptiveKeysAdapter(endpoint_id=endpoint_id)
+    adapter = ListAdaptiveKeysAdapter(provider_id=provider_id)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


@router.patch(
    "/keys/{key_id}/mode",
-    summary="Toggle key's concurrency control mode",
+    summary="Toggle key's RPM control mode",
 )
 async def toggle_adaptive_mode(
    key_id: str,
@@ -103,10 +105,10 @@ async def toggle_adaptive_mode(
    db: Session = Depends(get_db),
 ):
    """
-    Toggle the concurrency control mode for a specific key
+    Toggle the RPM control mode for a specific key

    Parameters:
-    - enabled: true=adaptive mode (max_concurrent=NULL), false=fixed limit mode
+    - enabled: true=adaptive mode (rpm_limit=NULL), false=fixed limit mode
    - fixed_limit: fixed limit value (required when enabled=false)
    """
    adapter = ToggleAdaptiveModeAdapter(key_id=key_id)
@@ -124,7 +126,7 @@ async def get_adaptive_stats(
    db: Session = Depends(get_db),
 ):
    """
-    获取指定Key的自适应并发统计信息
+    获取指定Key的自适应 RPM 统计信息

    包括：
    - 当前配置
@@ -149,12 +151,12 @@ async def reset_adaptive_learning(
    Reset the adaptive learning state for a specific key

    Clears:
-    - Learned concurrency limit (learned_max_concurrent)
+    - Learned RPM limit (learned_rpm_limit)
    - 429 error counts
    - Adjustment history

    Does not change:
-    - max_concurrent config (determines adaptive mode)
+    - rpm_limit config (determines adaptive mode)
    """
    adapter = ResetAdaptiveLearningAdapter(key_id=key_id)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -162,40 +164,40 @@ async def reset_adaptive_learning(

@router.patch(
    "/keys/{key_id}/limit",
-    summary="Set key to fixed concurrency limit mode",
+    summary="Set key to fixed RPM limit mode",
 )
-async def set_concurrent_limit(
+async def set_rpm_limit(
    key_id: str,
    request: Request,
-    limit: int = Query(..., ge=1, le=100, description="Concurrency limit value"),
+    limit: int = Query(..., ge=1, le=100, description="RPM limit value (1-100)"),
    db: Session = Depends(get_db),
 ):
    """
-    Set key to fixed concurrency limit mode
+    Set key to fixed RPM limit mode

    Note:
    - After setting this value, key switches to fixed limit mode and won't auto-adjust
    - To restore adaptive mode, use PATCH /keys/{key_id}/mode
    """
-    adapter = SetConcurrentLimitAdapter(key_id=key_id, limit=limit)
+    adapter = SetRPMLimitAdapter(key_id=key_id, limit=limit)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


@router.get(
    "/summary",
-    summary="获取自适应并发的全局统计",
+    summary="获取自适应 RPM 的全局统计",
 )
 async def get_adaptive_summary(
    request: Request,
    db: Session = Depends(get_db),
 ):
    """
-    获取自适应并发的全局统计摘要
+    获取自适应 RPM 的全局统计摘要

    包括：
    - 启用自适应模式的Key数量
    - 总429错误数
-    - 并发限制调整次数
+    - RPM 限制调整次数
    """
    adapter = AdaptiveSummaryAdapter()
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -206,26 +208,29 @@ async def get_adaptive_summary(

@dataclass
 class ListAdaptiveKeysAdapter(AdminApiAdapter):
-    endpoint_id: Optional[str] = None
+    provider_id: Optional[str] = None

    async def handle(self, context):  # type: ignore[override]
-        # 自适应模式：max_concurrent = NULL
-        query = context.db.query(ProviderAPIKey).filter(ProviderAPIKey.max_concurrent.is_(None))
-        if self.endpoint_id:
-            query = query.filter(ProviderAPIKey.endpoint_id == self.endpoint_id)
+        # 自适应模式：rpm_limit = NULL
+        query = context.db.query(ProviderAPIKey).filter(ProviderAPIKey.rpm_limit.is_(None))
+        if self.provider_id:
+            query = query.filter(ProviderAPIKey.provider_id == self.provider_id)

        keys = query.all()
        return [
            KeyListItem(
                id=key.id,
                name=key.name,
-                endpoint_id=key.endpoint_id,
-                is_adaptive=key.max_concurrent is None,
-                max_concurrent=key.max_concurrent,
+                provider_id=key.provider_id,
+                api_formats=key.api_formats or [],
+                is_adaptive=key.rpm_limit is None,
+                rpm_limit=key.rpm_limit,
                effective_limit=(
-                    key.learned_max_concurrent if key.max_concurrent is None else key.max_concurrent
+                    (key.learned_rpm_limit if key.learned_rpm_limit is not None else RPMDefaults.INITIAL_LIMIT)
+                    if key.rpm_limit is None
+                    else key.rpm_limit
                ),
-                learned_max_concurrent=key.learned_max_concurrent,
+                learned_rpm_limit=key.learned_rpm_limit,
                concurrent_429_count=key.concurrent_429_count or 0,
                rpm_429_count=key.rpm_429_count or 0,
            )
@@ -252,28 +257,32 @@ class ToggleAdaptiveModeAdapter(AdminApiAdapter):
            raise InvalidRequestException("请求数据验证失败")

        if body.enabled:
-            # 启用自适应模式：将 max_concurrent 设为 NULL
-            key.max_concurrent = None
-            message = "已切换为自适应模式，系统将自动学习并调整并发限制"
+            # 启用自适应模式：将 rpm_limit 设为 NULL
+            key.rpm_limit = None
+            message = "已切换为自适应模式，系统将自动学习并调整 RPM 限制"
        else:
            # 禁用自适应模式：设置固定限制
            if body.fixed_limit is None:
                raise HTTPException(
                    status_code=400, detail="禁用自适应模式时必须提供 fixed_limit 参数"
                )
-            key.max_concurrent = body.fixed_limit
-            message = f"已切换为固定限制模式，并发限制设为 {body.fixed_limit}"
+            key.rpm_limit = body.fixed_limit
+            message = f"已切换为固定限制模式，RPM 限制设为 {body.fixed_limit}"

        context.db.commit()
        context.db.refresh(key)

-        is_adaptive = key.max_concurrent is None
+        is_adaptive = key.rpm_limit is None
        return {
            "message": message,
            "key_id": key.id,
            "is_adaptive": is_adaptive,
-            "max_concurrent": key.max_concurrent,
-            "effective_limit": key.learned_max_concurrent if is_adaptive else key.max_concurrent,
+            "rpm_limit": key.rpm_limit,
+            "effective_limit": (
+                (key.learned_rpm_limit if key.learned_rpm_limit is not None else RPMDefaults.INITIAL_LIMIT)
+                if is_adaptive
+                else key.rpm_limit
+            ),
        }


@@ -286,13 +295,13 @@ class GetAdaptiveStatsAdapter(AdminApiAdapter):
        if not key:
            raise HTTPException(status_code=404, detail="Key not found")

-        adaptive_manager = get_adaptive_manager()
+        adaptive_manager = get_adaptive_rpm_manager()
        stats = adaptive_manager.get_adjustment_stats(key)

        # 转换字段名以匹配响应模型
        return AdaptiveStatsResponse(
            adaptive_mode=stats["adaptive_mode"],
-            max_concurrent=stats["max_concurrent"],
+            rpm_limit=stats["rpm_limit"],
            effective_limit=stats["effective_limit"],
            learned_limit=stats["learned_limit"],
            concurrent_429_count=stats["concurrent_429_count"],
@@ -313,13 +322,13 @@ class ResetAdaptiveLearningAdapter(AdminApiAdapter):
        if not key:
            raise HTTPException(status_code=404, detail="Key not found")

-        adaptive_manager = get_adaptive_manager()
+        adaptive_manager = get_adaptive_rpm_manager()
        adaptive_manager.reset_learning(context.db, key)
        return {"message": "学习状态已重置", "key_id": key.id}


@dataclass
-class SetConcurrentLimitAdapter(AdminApiAdapter):
+class SetRPMLimitAdapter(AdminApiAdapter):
    key_id: str
    limit: int

@@ -328,25 +337,25 @@ class SetConcurrentLimitAdapter(AdminApiAdapter):
        if not key:
            raise HTTPException(status_code=404, detail="Key not found")

-        was_adaptive = key.max_concurrent is None
-        key.max_concurrent = self.limit
+        was_adaptive = key.rpm_limit is None
+        key.rpm_limit = self.limit
        context.db.commit()
        context.db.refresh(key)

        return {
-            "message": f"已设置为固定限制模式，并发限制为 {self.limit}",
+            "message": f"已设置为固定限制模式，RPM 限制为 {self.limit}",
            "key_id": key.id,
            "is_adaptive": False,
-            "max_concurrent": key.max_concurrent,
+            "rpm_limit": key.rpm_limit,
            "previous_mode": "adaptive" if was_adaptive else "fixed",
        }


 class AdaptiveSummaryAdapter(AdminApiAdapter):
    async def handle(self, context):  # type: ignore[override]
-        # 自适应模式：max_concurrent = NULL
+        # 自适应模式：rpm_limit = NULL
        adaptive_keys = (
-            context.db.query(ProviderAPIKey).filter(ProviderAPIKey.max_concurrent.is_(None)).all()
+            context.db.query(ProviderAPIKey).filter(ProviderAPIKey.rpm_limit.is_(None)).all()
        )

        total_keys = len(adaptive_keys)
--- a/src/api/admin/endpoints/concurrency.py
+++ b/src/api/admin/endpoints/concurrency.py
@@ -1,9 +1,8 @@
 """
-Endpoint 并发控制管理 API
+Key RPM 限制管理 API
 """

 from dataclasses import dataclass
-from typing import Optional

 from fastapi import APIRouter, Depends, Request
 from sqlalchemy.orm import Session
@@ -12,83 +11,56 @@ from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
 from src.core.exceptions import NotFoundException
 from src.database import get_db
-from src.models.database import ProviderAPIKey, ProviderEndpoint
-from src.models.endpoint_models import (
-    ConcurrencyStatusResponse,
-    ResetConcurrencyRequest,
-)
+from src.models.database import ProviderAPIKey
+from src.models.endpoint_models import KeyRpmStatusResponse
 from src.services.rate_limit.concurrency_manager import get_concurrency_manager

-router = APIRouter(tags=["Concurrency Control"])
+router = APIRouter(tags=["RPM Control"])
 pipeline = ApiRequestPipeline()


-@router.get("/concurrency/endpoint/{endpoint_id}", response_model=ConcurrencyStatusResponse)
-async def get_endpoint_concurrency(
-    endpoint_id: str,
-    request: Request,
-    db: Session = Depends(get_db),
-) -> ConcurrencyStatusResponse:
-    """
-    获取 Endpoint 当前并发状态
-
-    查询指定 Endpoint 的实时并发使用情况，包括当前并发数和最大并发限制。
-
-    **路径参数**:
-    - `endpoint_id`: Endpoint ID
-
-    **返回字段**:
-    - `endpoint_id`: Endpoint ID
-    - `endpoint_current_concurrency`: 当前并发数
-    - `endpoint_max_concurrent`: 最大并发限制
-    """
-    adapter = AdminEndpointConcurrencyAdapter(endpoint_id=endpoint_id)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
-@router.get("/concurrency/key/{key_id}", response_model=ConcurrencyStatusResponse)
-async def get_key_concurrency(
+@router.get("/rpm/key/{key_id}", response_model=KeyRpmStatusResponse)
+async def get_key_rpm(
    key_id: str,
    request: Request,
    db: Session = Depends(get_db),
-) -> ConcurrencyStatusResponse:
+) -> KeyRpmStatusResponse:
    """
-    获取 Key 当前并发状态
+    获取 Key 当前 RPM 状态

-    查询指定 API Key 的实时并发使用情况，包括当前并发数和最大并发限制。
+    查询指定 API Key 的实时 RPM 使用情况，包括当前 RPM 计数和最大 RPM 限制。

    **路径参数**:
    - `key_id`: API Key ID

    **返回字段**:
    - `key_id`: API Key ID
-    - `key_current_concurrency`: 当前并发数
-    - `key_max_concurrent`: 最大并发限制
+    - `current_rpm`: 当前 RPM 计数
+    - `rpm_limit`: RPM 限制
    """
-    adapter = AdminKeyConcurrencyAdapter(key_id=key_id)
+    adapter = AdminKeyRpmAdapter(key_id=key_id)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


-@router.delete("/concurrency")
-async def reset_concurrency(
-    request: ResetConcurrencyRequest,
+@router.delete("/rpm/key/{key_id}")
+async def reset_key_rpm(
+    key_id: str,
    http_request: Request,
    db: Session = Depends(get_db),
 ) -> dict:
    """
-    重置并发计数器
+    重置 Key RPM 计数器

-    重置指定 Endpoint 或 Key 的并发计数器，用于解决计数不准确的问题。
+    重置指定 API Key 的 RPM 计数器，用于解决计数不准确的问题。
    管理员功能，请谨慎使用。

-    **请求体字段**:
-    - `endpoint_id`: Endpoint ID（可选）
-    - `key_id`: API Key ID（可选）
+    **路径参数**:
+    - `key_id`: API Key ID

    **返回字段**:
    - `message`: 操作结果消息
    """
-    adapter = AdminResetConcurrencyAdapter(endpoint_id=request.endpoint_id, key_id=request.key_id)
+    adapter = AdminResetKeyRpmAdapter(key_id=key_id)
    return await pipeline.run(adapter=adapter, http_request=http_request, db=db, mode=adapter.mode)


@@ -96,31 +68,7 @@ async def reset_concurrency(


@dataclass
-class AdminEndpointConcurrencyAdapter(AdminApiAdapter):
-    endpoint_id: str
-
-    async def handle(self, context):  # type: ignore[override]
-        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
-
-        concurrency_manager = await get_concurrency_manager()
-        endpoint_count, _ = await concurrency_manager.get_current_concurrency(
-            endpoint_id=self.endpoint_id
-        )
-
-        return ConcurrencyStatusResponse(
-            endpoint_id=self.endpoint_id,
-            endpoint_current_concurrency=endpoint_count,
-            endpoint_max_concurrent=endpoint.max_concurrent,
-        )
-
-
-@dataclass
-class AdminKeyConcurrencyAdapter(AdminApiAdapter):
+class AdminKeyRpmAdapter(AdminApiAdapter):
    key_id: str

    async def handle(self, context):  # type: ignore[override]
@@ -130,23 +78,20 @@ class AdminKeyConcurrencyAdapter(AdminApiAdapter):
            raise NotFoundException(f"Key {self.key_id} 不存在")

        concurrency_manager = await get_concurrency_manager()
-        _, key_count = await concurrency_manager.get_current_concurrency(key_id=self.key_id)
+        key_count = await concurrency_manager.get_key_rpm_count(key_id=self.key_id)

-        return ConcurrencyStatusResponse(
+        return KeyRpmStatusResponse(
            key_id=self.key_id,
-            key_current_concurrency=key_count,
-            key_max_concurrent=key.max_concurrent,
+            current_rpm=key_count,
+            rpm_limit=key.rpm_limit,
        )


@dataclass
-class AdminResetConcurrencyAdapter(AdminApiAdapter):
-    endpoint_id: Optional[str]
-    key_id: Optional[str]
+class AdminResetKeyRpmAdapter(AdminApiAdapter):
+    key_id: str

    async def handle(self, context):  # type: ignore[override]
        concurrency_manager = await get_concurrency_manager()
-        await concurrency_manager.reset_concurrency(
-            endpoint_id=self.endpoint_id, key_id=self.key_id
-        )
-        return {"message": "并发计数已重置"}
+        await concurrency_manager.reset_key_rpm(key_id=self.key_id)
+        return {"message": "RPM 计数已重置"}
--- a/src/api/admin/endpoints/health.py
+++ b/src/api/admin/endpoints/health.py
@@ -5,7 +5,7 @@ Endpoint 健康监控 API
 from collections import defaultdict
 from dataclasses import dataclass
 from datetime import datetime, timedelta, timezone
-from typing import Dict, List
+from typing import Dict, List, Optional

 from fastapi import APIRouter, Depends, Query, Request
 from sqlalchemy import func
@@ -128,29 +128,32 @@ async def get_api_format_health_monitor(
 async def get_key_health(
    key_id: str,
    request: Request,
+    api_format: Optional[str] = Query(None, description="API 格式（可选，如 CLAUDE、OPENAI）"),
    db: Session = Depends(get_db),
 ) -> HealthStatusResponse:
    """
    获取 Key 健康状态

    获取指定 API Key 的健康状态详情，包括健康分数、连续失败次数、
-    熔断器状态等信息。
+    熔断器状态等信息。支持按 API 格式查询。

    **路径参数**:
    - `key_id`: API Key ID

+    **查询参数**:
+    - `api_format`: 可选，指定 API 格式（如 CLAUDE、OPENAI）。
+      - 指定时返回该格式的健康度详情
+      - 不指定时返回所有格式的健康度摘要
+
    **返回字段**:
    - `key_id`: API Key ID
    - `key_health_score`: 健康分数（0.0-1.0）
-    - `key_consecutive_failures`: 连续失败次数
-    - `key_last_failure_at`: 最后失败时间
    - `key_is_active`: 是否活跃
    - `key_statistics`: 统计信息
-    - `circuit_breaker_open`: 熔断器是否打开
-    - `circuit_breaker_open_at`: 熔断器打开时间
-    - `next_probe_at`: 下次探测时间
+    - `health_by_format`: 按格式的健康度数据（无 api_format 参数时）
+    - `circuit_breaker_open`: 熔断器是否打开（有 api_format 参数时）
    """
-    adapter = AdminKeyHealthAdapter(key_id=key_id)
+    adapter = AdminKeyHealthAdapter(key_id=key_id, api_format=api_format)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


@@ -158,17 +161,23 @@ async def get_key_health(
 async def recover_key_health(
    key_id: str,
    request: Request,
+    api_format: Optional[str] = Query(None, description="API 格式（可选，不指定则恢复所有格式）"),
    db: Session = Depends(get_db),
 ) -> dict:
    """
    恢复 Key 健康状态

    手动恢复指定 Key 的健康状态，将健康分数重置为 1.0，关闭熔断器，
-    取消自动禁用，并重置所有失败计数。
+    取消自动禁用，并重置所有失败计数。支持按 API 格式恢复。

    **路径参数**:
    - `key_id`: API Key ID

+    **查询参数**:
+    - `api_format`: 可选，指定 API 格式（如 CLAUDE、OPENAI）
+      - 指定时仅恢复该格式的健康度
+      - 不指定时恢复所有格式
+
    **返回字段**:
    - `message`: 操作结果消息
    - `details`: 详细信息
@@ -176,7 +185,7 @@ async def recover_key_health(
      - `circuit_breaker_open`: 熔断器状态
      - `is_active`: 是否活跃
    """
-    adapter = AdminRecoverKeyHealthAdapter(key_id=key_id)
+    adapter = AdminRecoverKeyHealthAdapter(key_id=key_id, api_format=api_format)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


@@ -276,34 +285,9 @@ class AdminApiFormatHealthMonitorAdapter(AdminApiAdapter):
            )
            all_formats[api_format] = provider_count

-        # 1.1 获取所有活跃的 API 格式及其 API Key 数量
-        active_keys = (
-            db.query(
-                ProviderEndpoint.api_format,
-                func.count(ProviderAPIKey.id).label("key_count"),
-            )
-            .join(ProviderAPIKey, ProviderEndpoint.id == ProviderAPIKey.endpoint_id)
-            .join(Provider, ProviderEndpoint.provider_id == Provider.id)
-            .filter(
-                ProviderEndpoint.is_active.is_(True),
-                Provider.is_active.is_(True),
-                ProviderAPIKey.is_active.is_(True),
-            )
-            .group_by(ProviderEndpoint.api_format)
-            .all()
-        )
-
-        # 构建所有格式的 key_count 映射
-        key_counts: Dict[str, int] = {}
-        for api_format_enum, key_count in active_keys:
-            api_format = (
-                api_format_enum.value if hasattr(api_format_enum, "value") else str(api_format_enum)
-            )
-            key_counts[api_format] = key_count
-
-        # 1.2 建立每个 API 格式对应的 Endpoint ID 列表，供 Usage 时间线生成使用
+        # 1.1 建立每个 API 格式对应的 Endpoint ID 列表（用于时间线生成），并收集活跃的 provider+format 组合
        endpoint_rows = (
-            db.query(ProviderEndpoint.api_format, ProviderEndpoint.id)
+            db.query(ProviderEndpoint.api_format, ProviderEndpoint.id, ProviderEndpoint.provider_id)
            .join(Provider, ProviderEndpoint.provider_id == Provider.id)
            .filter(
                ProviderEndpoint.is_active.is_(True),
@@ -312,11 +296,32 @@ class AdminApiFormatHealthMonitorAdapter(AdminApiAdapter):
            .all()
        )
        endpoint_map: Dict[str, List[str]] = defaultdict(list)
-        for api_format_enum, endpoint_id in endpoint_rows:
+        active_provider_formats: set[tuple[str, str]] = set()
+        for api_format_enum, endpoint_id, provider_id in endpoint_rows:
            api_format = (
                api_format_enum.value if hasattr(api_format_enum, "value") else str(api_format_enum)
            )
            endpoint_map[api_format].append(endpoint_id)
+            active_provider_formats.add((str(provider_id), api_format))
+
+        # 1.2 统计每个 API 格式可用的活跃 Key 数量（Key 属于 Provider，通过 api_formats 关联格式）
+        key_counts: Dict[str, int] = {}
+        if active_provider_formats:
+            active_provider_keys = (
+                db.query(ProviderAPIKey.provider_id, ProviderAPIKey.api_formats)
+                .join(Provider, ProviderAPIKey.provider_id == Provider.id)
+                .filter(
+                    Provider.is_active.is_(True),
+                    ProviderAPIKey.is_active.is_(True),
+                )
+                .all()
+            )
+            for provider_id, api_formats in active_provider_keys:
+                pid = str(provider_id)
+                for fmt in (api_formats or []):
+                    if (pid, fmt) not in active_provider_formats:
+                        continue
+                    key_counts[fmt] = key_counts.get(fmt, 0) + 1

        # 2. 统计窗口内每个 API 格式的请求状态分布（真实统计）
        # 只统计最终状态：success, failed, skipped
@@ -457,28 +462,45 @@ class AdminApiFormatHealthMonitorAdapter(AdminApiAdapter):
@dataclass
 class AdminKeyHealthAdapter(AdminApiAdapter):
    key_id: str
+    api_format: Optional[str] = None

    async def handle(self, context):  # type: ignore[override]
-        health_data = health_monitor.get_key_health(context.db, self.key_id)
+        health_data = health_monitor.get_key_health(context.db, self.key_id, self.api_format)
        if not health_data:
            raise NotFoundException(f"Key {self.key_id} 不存在")

-        return HealthStatusResponse(
-            key_id=health_data["key_id"],
-            key_health_score=health_data["health_score"],
-            key_consecutive_failures=health_data["consecutive_failures"],
-            key_last_failure_at=health_data["last_failure_at"],
-            key_is_active=health_data["is_active"],
-            key_statistics=health_data["statistics"],
-            circuit_breaker_open=health_data["circuit_breaker_open"],
-            circuit_breaker_open_at=health_data["circuit_breaker_open_at"],
-            next_probe_at=health_data["next_probe_at"],
-        )
+        # 构建响应
+        response_data = {
+            "key_id": health_data["key_id"],
+            "key_is_active": health_data["is_active"],
+            "key_statistics": health_data.get("statistics"),
+            "key_health_score": health_data.get("health_score", 1.0),
+        }
+
+        if self.api_format:
+            # 单格式查询
+            response_data["api_format"] = self.api_format
+            response_data["key_consecutive_failures"] = health_data.get("consecutive_failures")
+            response_data["key_last_failure_at"] = health_data.get("last_failure_at")
+            circuit = health_data.get("circuit_breaker", {})
+            response_data["circuit_breaker_open"] = circuit.get("open", False)
+            response_data["circuit_breaker_open_at"] = circuit.get("open_at")
+            response_data["next_probe_at"] = circuit.get("next_probe_at")
+            response_data["half_open_until"] = circuit.get("half_open_until")
+            response_data["half_open_successes"] = circuit.get("half_open_successes", 0)
+            response_data["half_open_failures"] = circuit.get("half_open_failures", 0)
+        else:
+            # 全格式查询
+            response_data["any_circuit_open"] = health_data.get("any_circuit_open", False)
+            response_data["health_by_format"] = health_data.get("health_by_format")
+
+        return HealthStatusResponse(**response_data)


@dataclass
 class AdminRecoverKeyHealthAdapter(AdminApiAdapter):
    key_id: str
+    api_format: Optional[str] = None

    async def handle(self, context):  # type: ignore[override]
        db = context.db
@@ -486,28 +508,38 @@ class AdminRecoverKeyHealthAdapter(AdminApiAdapter):
        if not key:
            raise NotFoundException(f"Key {self.key_id} 不存在")

-        key.health_score = 1.0
-        key.consecutive_failures = 0
-        key.last_failure_at = None
-        key.circuit_breaker_open = False
-        key.circuit_breaker_open_at = None
-        key.next_probe_at = None
+        # 使用 health_monitor.reset_health 重置健康度
+        success = health_monitor.reset_health(db, key_id=self.key_id, api_format=self.api_format)
+        if not success:
+            raise Exception("重置健康度失败")
+
+        # 如果 Key 被禁用，重新启用
        if not key.is_active:
-            key.is_active = True
+            key.is_active = True  # type: ignore[assignment]

        db.commit()

-        admin_name = context.user.username if context.user else "admin"
-        logger.info(f"管理员恢复Key健康状态: {self.key_id} (health_score: 1.0, circuit_breaker: closed)")
-
-        return {
-            "message": "Key已完全恢复",
-            "details": {
-                "health_score": 1.0,
-                "circuit_breaker_open": False,
-                "is_active": True,
-            },
-        }
+        if self.api_format:
+            logger.info(f"管理员恢复Key健康状态: {self.key_id}/{self.api_format}")
+            return {
+                "message": f"Key 的 {self.api_format} 格式已恢复",
+                "details": {
+                    "api_format": self.api_format,
+                    "health_score": 1.0,
+                    "circuit_breaker_open": False,
+                    "is_active": True,
+                },
+            }
+        else:
+            logger.info(f"管理员恢复Key健康状态: {self.key_id} (所有格式)")
+            return {
+                "message": "Key 所有格式已恢复",
+                "details": {
+                    "health_score": 1.0,
+                    "circuit_breaker_open": False,
+                    "is_active": True,
+                },
+            }


 class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
@@ -516,10 +548,17 @@ class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
    async def handle(self, context):  # type: ignore[override]
        db = context.db

-        # 查找所有熔断的 Key
-        circuit_open_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.circuit_breaker_open == True).all()
-        )
+        # 查找所有有熔断格式的 Key（检查 circuit_breaker_by_format JSON 字段）
+        all_keys = db.query(ProviderAPIKey).all()
+
+        # 筛选出有任何格式熔断的 Key
+        circuit_open_keys = []
+        for key in all_keys:
+            circuit_by_format = key.circuit_breaker_by_format or {}
+            for fmt, circuit_data in circuit_by_format.items():
+                if circuit_data.get("open"):
+                    circuit_open_keys.append(key)
+                    break

        if not circuit_open_keys:
            return {
@@ -530,17 +569,15 @@ class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):

        recovered_keys = []
        for key in circuit_open_keys:
-            key.health_score = 1.0
-            key.consecutive_failures = 0
-            key.last_failure_at = None
-            key.circuit_breaker_open = False
-            key.circuit_breaker_open_at = None
-            key.next_probe_at = None
+            # 重置所有格式的健康度
+            key.health_by_format = {}  # type: ignore[assignment]
+            key.circuit_breaker_by_format = {}  # type: ignore[assignment]
            recovered_keys.append(
                {
                    "key_id": key.id,
                    "key_name": key.name,
-                    "endpoint_id": key.endpoint_id,
+                    "provider_id": key.provider_id,
+                    "api_formats": key.api_formats,
                }
            )

@@ -552,7 +589,6 @@ class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
        HealthMonitor._open_circuit_keys = 0
        health_open_circuits.set(0)

-        admin_name = context.user.username if context.user else "admin"
        logger.info(f"管理员批量恢复 {len(recovered_keys)} 个 Key 的健康状态")

        return {
--- a/src/api/admin/endpoints/keys.py
+++ b/src/api/admin/endpoints/keys.py
@@ -1,5 +1,5 @@
 """
-Endpoint API Keys 管理
+Provider API Keys 管理
 """

 import uuid
@@ -12,6 +12,7 @@ from sqlalchemy.orm import Session

 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import RPMDefaults
 from src.core.crypto import crypto_service
 from src.core.exceptions import InvalidRequestException, NotFoundException
 from src.core.key_capabilities import get_capability
@@ -20,96 +21,14 @@ from src.database import get_db
 from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint
 from src.services.cache.provider_cache import ProviderCacheService
 from src.models.endpoint_models import (
-    BatchUpdateKeyPriorityRequest,
    EndpointAPIKeyCreate,
    EndpointAPIKeyResponse,
    EndpointAPIKeyUpdate,
 )

-router = APIRouter(tags=["Endpoint Keys"])
+router = APIRouter(tags=["Provider Keys"])
 pipeline = ApiRequestPipeline()

-
-@router.get("/{endpoint_id}/keys", response_model=List[EndpointAPIKeyResponse])
-async def list_endpoint_keys(
-    endpoint_id: str,
-    request: Request,
-    skip: int = Query(0, ge=0, description="跳过的记录数"),
-    limit: int = Query(100, ge=1, le=1000, description="返回的最大记录数"),
-    db: Session = Depends(get_db),
-) -> List[EndpointAPIKeyResponse]:
-    """
-    获取 Endpoint 的所有 Keys
-
-    获取指定 Endpoint 下的所有 API Key 列表，包括 Key 的配置、统计信息等。
-    结果按优先级和创建时间排序。
-
-    **路径参数**:
-    - `endpoint_id`: Endpoint ID
-
-    **查询参数**:
-    - `skip`: 跳过的记录数，用于分页（默认 0）
-    - `limit`: 返回的最大记录数（1-1000，默认 100）
-
-    **返回字段**:
-    - `id`: Key ID
-    - `name`: Key 名称
-    - `api_key_masked`: 脱敏后的 API Key
-    - `internal_priority`: 内部优先级
-    - `global_priority`: 全局优先级
-    - `rate_multiplier`: 速率倍数
-    - `max_concurrent`: 最大并发数（null 表示自适应模式）
-    - `is_adaptive`: 是否为自适应并发模式
-    - `effective_limit`: 有效并发限制
-    - `success_rate`: 成功率
-    - `avg_response_time_ms`: 平均响应时间（毫秒）
-    - 其他配置和统计字段
-    """
-    adapter = AdminListEndpointKeysAdapter(
-        endpoint_id=endpoint_id,
-        skip=skip,
-        limit=limit,
-    )
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
-@router.post("/{endpoint_id}/keys", response_model=EndpointAPIKeyResponse)
-async def add_endpoint_key(
-    endpoint_id: str,
-    key_data: EndpointAPIKeyCreate,
-    request: Request,
-    db: Session = Depends(get_db),
-) -> EndpointAPIKeyResponse:
-    """
-    为 Endpoint 添加 Key
-
-    为指定 Endpoint 添加新的 API Key，支持配置并发限制、速率倍数、
-    优先级、配额限制、能力限制等。
-
-    **路径参数**:
-    - `endpoint_id`: Endpoint ID
-
-    **请求体字段**:
-    - `endpoint_id`: Endpoint ID（必须与路径参数一致）
-    - `api_key`: API Key 原文（将被加密存储）
-    - `name`: Key 名称
-    - `note`: 备注（可选）
-    - `rate_multiplier`: 速率倍数（默认 1.0）
-    - `internal_priority`: 内部优先级（默认 100）
-    - `max_concurrent`: 最大并发数（null 表示自适应模式）
-    - `rate_limit`: 每分钟请求限制（可选）
-    - `daily_limit`: 每日请求限制（可选）
-    - `monthly_limit`: 每月请求限制（可选）
-    - `allowed_models`: 允许的模型列表（可选）
-    - `capabilities`: 能力配置（可选）
-
-    **返回字段**:
-    - 包含完整的 Key 信息，其中 `api_key_plain` 为原文（仅在创建时返回）
-    """
-    adapter = AdminCreateEndpointKeyAdapter(endpoint_id=endpoint_id, key_data=key_data)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
@router.put("/keys/{key_id}", response_model=EndpointAPIKeyResponse)
 async def update_endpoint_key(
    key_id: str,
@@ -118,7 +37,7 @@ async def update_endpoint_key(
    db: Session = Depends(get_db),
 ) -> EndpointAPIKeyResponse:
    """
-    更新 Endpoint Key
+    更新 Provider Key

    更新指定 Key 的配置，支持修改并发限制、速率倍数、优先级、
    配额限制、能力限制等。支持部分更新。
@@ -132,10 +51,7 @@ async def update_endpoint_key(
    - `note`: 备注
    - `rate_multiplier`: 速率倍数
    - `internal_priority`: 内部优先级
-    - `max_concurrent`: 最大并发数（设置为 null 可切换到自适应模式）
-    - `rate_limit`: 每分钟请求限制
-    - `daily_limit`: 每日请求限制
-    - `monthly_limit`: 每月请求限制
+    - `rpm_limit`: RPM 限制（设置为 null 可切换到自适应模式）
    - `allowed_models`: 允许的模型列表
    - `capabilities`: 能力配置
    - `is_active`: 是否活跃
@@ -210,7 +126,7 @@ async def delete_endpoint_key(
    db: Session = Depends(get_db),
 ) -> dict:
    """
-    删除 Endpoint Key
+    删除 Provider Key

    删除指定的 API Key。此操作不可逆，请谨慎使用。

@@ -224,163 +140,66 @@ async def delete_endpoint_key(
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


-@router.put("/{endpoint_id}/keys/batch-priority")
-async def batch_update_key_priority(
-    endpoint_id: str,
-    request: Request,
-    priority_data: BatchUpdateKeyPriorityRequest,
-    db: Session = Depends(get_db),
-) -> dict:
-    """
-    批量更新 Endpoint 下 Keys 的优先级
+# ========== Provider Keys API ==========

-    批量更新指定 Endpoint 下多个 Key 的内部优先级，用于拖动排序。
-    所有 Key 必须属于指定的 Endpoint。
+
+@router.get("/providers/{provider_id}/keys", response_model=List[EndpointAPIKeyResponse])
+async def list_provider_keys(
+    provider_id: str,
+    request: Request,
+    skip: int = Query(0, ge=0, description="跳过的记录数"),
+    limit: int = Query(100, ge=1, le=1000, description="返回的最大记录数"),
+    db: Session = Depends(get_db),
+) -> List[EndpointAPIKeyResponse]:
+    """
+    获取 Provider 的所有 Keys
+
+    获取指定 Provider 下的所有 API Key 列表，支持多 API 格式。
+    结果按优先级和创建时间排序。

    **路径参数**:
-    - `endpoint_id`: Endpoint ID
+    - `provider_id`: Provider ID
+
+    **查询参数**:
+    - `skip`: 跳过的记录数，用于分页（默认 0）
+    - `limit`: 返回的最大记录数（1-1000，默认 100）
+    """
+    adapter = AdminListProviderKeysAdapter(
+        provider_id=provider_id,
+        skip=skip,
+        limit=limit,
+    )
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.post("/providers/{provider_id}/keys", response_model=EndpointAPIKeyResponse)
+async def add_provider_key(
+    provider_id: str,
+    key_data: EndpointAPIKeyCreate,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> EndpointAPIKeyResponse:
+    """
+    为 Provider 添加 Key
+
+    为指定 Provider 添加新的 API Key，支持配置多个 API 格式。
+
+    **路径参数**:
+    - `provider_id`: Provider ID

    **请求体字段**:
-    - `priorities`: 优先级列表
-      - `key_id`: Key ID
-      - `internal_priority`: 新的内部优先级
-
-    **返回字段**:
-    - `message`: 操作结果消息
-    - `updated_count`: 实际更新的 Key 数量
+    - `api_formats`: 支持的 API 格式列表（必填）
+    - `api_key`: API Key 原文（将被加密存储）
+    - `name`: Key 名称
+    - 其他配置字段同 Key
    """
-    adapter = AdminBatchUpdateKeyPriorityAdapter(endpoint_id=endpoint_id, priority_data=priority_data)
+    adapter = AdminCreateProviderKeyAdapter(provider_id=provider_id, key_data=key_data)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)


 # -------- Adapters --------


-@dataclass
-class AdminListEndpointKeysAdapter(AdminApiAdapter):
-    endpoint_id: str
-    skip: int
-    limit: int
-
-    async def handle(self, context):  # type: ignore[override]
-        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
-
-        keys = (
-            db.query(ProviderAPIKey)
-            .filter(ProviderAPIKey.endpoint_id == self.endpoint_id)
-            .order_by(ProviderAPIKey.internal_priority.asc(), ProviderAPIKey.created_at.asc())
-            .offset(self.skip)
-            .limit(self.limit)
-            .all()
-        )
-
-        result: List[EndpointAPIKeyResponse] = []
-        for key in keys:
-            try:
-                decrypted_key = crypto_service.decrypt(key.api_key)
-                masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
-            except Exception:
-                masked_key = "***ERROR***"
-
-            success_rate = key.success_count / key.request_count if key.request_count > 0 else 0.0
-            avg_response_time_ms = (
-                key.total_response_time_ms / key.success_count if key.success_count > 0 else 0.0
-            )
-
-            is_adaptive = key.max_concurrent is None
-            key_dict = key.__dict__.copy()
-            key_dict.pop("_sa_instance_state", None)
-            key_dict.update(
-                {
-                    "api_key_masked": masked_key,
-                    "api_key_plain": None,
-                    "success_rate": success_rate,
-                    "avg_response_time_ms": round(avg_response_time_ms, 2),
-                    "is_adaptive": is_adaptive,
-                    "effective_limit": (
-                        key.learned_max_concurrent if is_adaptive else key.max_concurrent
-                    ),
-                }
-            )
-            result.append(EndpointAPIKeyResponse(**key_dict))
-
-        return result
-
-
-@dataclass
-class AdminCreateEndpointKeyAdapter(AdminApiAdapter):
-    endpoint_id: str
-    key_data: EndpointAPIKeyCreate
-
-    async def handle(self, context):  # type: ignore[override]
-        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
-
-        if self.key_data.endpoint_id != self.endpoint_id:
-            raise InvalidRequestException("endpoint_id 不匹配")
-
-        encrypted_key = crypto_service.encrypt(self.key_data.api_key)
-        now = datetime.now(timezone.utc)
-        # max_concurrent=NULL 表示自适应模式，数字表示固定限制
-        new_key = ProviderAPIKey(
-            id=str(uuid.uuid4()),
-            endpoint_id=self.endpoint_id,
-            api_key=encrypted_key,
-            name=self.key_data.name,
-            note=self.key_data.note,
-            rate_multiplier=self.key_data.rate_multiplier,
-            internal_priority=self.key_data.internal_priority,
-            max_concurrent=self.key_data.max_concurrent,  # NULL=自适应模式
-            rate_limit=self.key_data.rate_limit,
-            daily_limit=self.key_data.daily_limit,
-            monthly_limit=self.key_data.monthly_limit,
-            allowed_models=self.key_data.allowed_models if self.key_data.allowed_models else None,
-            capabilities=self.key_data.capabilities if self.key_data.capabilities else None,
-            request_count=0,
-            success_count=0,
-            error_count=0,
-            total_response_time_ms=0,
-            is_active=True,
-            last_used_at=None,
-            created_at=now,
-            updated_at=now,
-        )
-
-        db.add(new_key)
-        db.commit()
-        db.refresh(new_key)
-
-        logger.info(f"[OK] 添加 Key: Endpoint={self.endpoint_id}, Key=***{self.key_data.api_key[-4:]}, ID={new_key.id}")
-
-        masked_key = f"{self.key_data.api_key[:8]}***{self.key_data.api_key[-4:]}"
-        is_adaptive = new_key.max_concurrent is None
-        response_dict = new_key.__dict__.copy()
-        response_dict.pop("_sa_instance_state", None)
-        response_dict.update(
-            {
-                "api_key_masked": masked_key,
-                "api_key_plain": self.key_data.api_key,
-                "success_rate": 0.0,
-                "avg_response_time_ms": 0.0,
-                "is_adaptive": is_adaptive,
-                "effective_limit": (
-                    new_key.learned_max_concurrent if is_adaptive else new_key.max_concurrent
-                ),
-            }
-        )
-
-        return EndpointAPIKeyResponse(**response_dict)
-
-
@dataclass
 class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
    key_id: str
@@ -396,14 +215,21 @@ class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
        if "api_key" in update_data:
            update_data["api_key"] = crypto_service.encrypt(update_data["api_key"])

-        # 特殊处理 max_concurrent：需要区分"未提供"和"显式设置为 null"
-        # 当 max_concurrent 被显式设置时（在 model_fields_set 中），即使值为 None 也应该更新
-        if "max_concurrent" in self.key_data.model_fields_set:
-            update_data["max_concurrent"] = self.key_data.max_concurrent
-            # 切换到自适应模式时，清空学习到的并发限制，让系统重新学习
-            if self.key_data.max_concurrent is None:
-                update_data["learned_max_concurrent"] = None
-                logger.info("Key %s 切换为自适应并发模式", self.key_id)
+        # 特殊处理 rpm_limit：需要区分"未提供"和"显式设置为 null"
+        if "rpm_limit" in self.key_data.model_fields_set:
+            update_data["rpm_limit"] = self.key_data.rpm_limit
+            if self.key_data.rpm_limit is None:
+                update_data["learned_rpm_limit"] = None
+                logger.info("Key %s 切换为自适应 RPM 模式", self.key_id)
+
+        # 统一处理 allowed_models：空列表/空字典 -> None（表示不限制）
+        if "allowed_models" in update_data:
+            am = update_data["allowed_models"]
+            if am is not None and (
+                (isinstance(am, list) and len(am) == 0)
+                or (isinstance(am, dict) and len(am) == 0)
+            ):
+                update_data["allowed_models"] = None

        for field, value in update_data.items():
            setattr(key, field, value)
@@ -412,39 +238,13 @@ class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
        db.commit()
        db.refresh(key)

-        # 如果更新了 rate_multiplier，清除缓存
-        if "rate_multiplier" in update_data:
-            await ProviderCacheService.invalidate_provider_api_key_cache(self.key_id)
+        # 任何字段更新都清除缓存，确保缓存一致性
+        # 包括 is_active、allowed_models、capabilities 等影响权限和行为的字段
+        await ProviderCacheService.invalidate_provider_api_key_cache(self.key_id)

        logger.info("[OK] 更新 Key: ID=%s, Updates=%s", self.key_id, list(update_data.keys()))

-        try:
-            decrypted_key = crypto_service.decrypt(key.api_key)
-            masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
-        except Exception:
-            masked_key = "***ERROR***"
-
-        success_rate = key.success_count / key.request_count if key.request_count > 0 else 0.0
-        avg_response_time_ms = (
-            key.total_response_time_ms / key.success_count if key.success_count > 0 else 0.0
-        )
-
-        is_adaptive = key.max_concurrent is None
-        response_dict = key.__dict__.copy()
-        response_dict.pop("_sa_instance_state", None)
-        response_dict.update(
-            {
-                "api_key_masked": masked_key,
-                "api_key_plain": None,
-                "success_rate": success_rate,
-                "avg_response_time_ms": round(avg_response_time_ms, 2),
-                "is_adaptive": is_adaptive,
-                "effective_limit": (
-                    key.learned_max_concurrent if is_adaptive else key.max_concurrent
-                ),
-            }
-        )
-        return EndpointAPIKeyResponse(**response_dict)
+        return _build_key_response(key)


@dataclass
@@ -481,7 +281,7 @@ class AdminDeleteEndpointKeyAdapter(AdminApiAdapter):
        if not key:
            raise NotFoundException(f"Key {self.key_id} 不存在")

-        endpoint_id = key.endpoint_id
+        provider_id = key.provider_id
        try:
            db.delete(key)
            db.commit()
@@ -490,7 +290,7 @@ class AdminDeleteEndpointKeyAdapter(AdminApiAdapter):
            logger.error(f"删除 Key 失败: ID={self.key_id}, Error={exc}")
            raise

-        logger.warning(f"[DELETE] 删除 Key: ID={self.key_id}, Endpoint={endpoint_id}")
+        logger.warning(f"[DELETE] 删除 Key: ID={self.key_id}, Provider={provider_id}")
        return {"message": f"Key {self.key_id} 已删除"}


@@ -498,31 +298,51 @@ class AdminGetKeysGroupedByFormatAdapter(AdminApiAdapter):
    async def handle(self, context):  # type: ignore[override]
        db = context.db

+        # Key 属于 Provider：按 key.api_formats 分组展示
        keys = (
-            db.query(ProviderAPIKey, ProviderEndpoint, Provider)
-            .join(ProviderEndpoint, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
-            .join(Provider, ProviderEndpoint.provider_id == Provider.id)
+            db.query(ProviderAPIKey, Provider)
+            .join(Provider, ProviderAPIKey.provider_id == Provider.id)
            .filter(
                ProviderAPIKey.is_active.is_(True),
-                ProviderEndpoint.is_active.is_(True),
                Provider.is_active.is_(True),
            )
            .order_by(
-                ProviderAPIKey.global_priority.asc().nullslast(), ProviderAPIKey.internal_priority.asc()
+                ProviderAPIKey.global_priority.asc().nullslast(),
+                ProviderAPIKey.internal_priority.asc(),
            )
            .all()
        )

+        provider_ids = {str(provider.id) for _key, provider in keys}
+        endpoints = (
+            db.query(
+                ProviderEndpoint.provider_id,
+                ProviderEndpoint.api_format,
+                ProviderEndpoint.base_url,
+            )
+            .filter(
+                ProviderEndpoint.provider_id.in_(provider_ids),
+                ProviderEndpoint.is_active.is_(True),
+            )
+            .all()
+        )
+        endpoint_base_url_map: Dict[tuple[str, str], str] = {}
+        for provider_id, api_format, base_url in endpoints:
+            fmt = api_format.value if hasattr(api_format, "value") else str(api_format)
+            endpoint_base_url_map[(str(provider_id), fmt)] = base_url
+
        grouped: Dict[str, List[dict]] = {}
-        for key, endpoint, provider in keys:
-            api_format = endpoint.api_format
-            if api_format not in grouped:
-                grouped[api_format] = []
+        for key, provider in keys:
+            api_formats = key.api_formats or []
+
+            if not api_formats:
+                continue  # 跳过没有 API 格式的 Key

            try:
                decrypted_key = crypto_service.decrypt(key.api_key)
                masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
-            except Exception:
+            except Exception as e:
+                logger.error(f"解密 Key 失败: key_id={key.id}, error={e}")
                masked_key = "***ERROR***"

            # 计算健康度指标
@@ -541,73 +361,209 @@ class AdminGetKeysGroupedByFormatAdapter(AdminApiAdapter):
                        cap_def = get_capability(cap_name)
                        caps_list.append(cap_def.short_name if cap_def else cap_name)

-            grouped[api_format].append(
-                {
-                    "id": key.id,
-                    "name": key.name,
-                    "api_key_masked": masked_key,
-                    "internal_priority": key.internal_priority,
-                    "global_priority": key.global_priority,
-                    "rate_multiplier": key.rate_multiplier,
-                    "is_active": key.is_active,
-                    "circuit_breaker_open": key.circuit_breaker_open,
-                    "provider_name": provider.display_name or provider.name,
-                    "endpoint_base_url": endpoint.base_url,
-                    "api_format": api_format,
-                    "capabilities": caps_list,
-                    "health_score": key.health_score,
-                    "success_rate": success_rate,
-                    "avg_response_time_ms": avg_response_time_ms,
-                    "request_count": key.request_count,
-                }
-            )
+            # 构建 Key 信息（基础数据）
+            key_info = {
+                "id": key.id,
+                "name": key.name,
+                "api_key_masked": masked_key,
+                "internal_priority": key.internal_priority,
+                "global_priority": key.global_priority,
+                "rate_multiplier": key.rate_multiplier,
+                "is_active": key.is_active,
+                "provider_name": provider.name,
+                "api_formats": api_formats,
+                "capabilities": caps_list,
+                "success_rate": success_rate,
+                "avg_response_time_ms": avg_response_time_ms,
+                "request_count": key.request_count,
+            }
+
+            # 将 Key 添加到每个支持的格式分组中，并附加格式特定的健康度数据
+            health_by_format = key.health_by_format or {}
+            circuit_by_format = key.circuit_breaker_by_format or {}
+            provider_id = str(provider.id)
+            for api_format in api_formats:
+                if api_format not in grouped:
+                    grouped[api_format] = []
+                # 为每个格式创建副本，设置当前格式
+                format_key_info = key_info.copy()
+                format_key_info["api_format"] = api_format
+                format_key_info["endpoint_base_url"] = endpoint_base_url_map.get(
+                    (provider_id, api_format)
+                )
+                # 添加格式特定的健康度数据
+                format_health = health_by_format.get(api_format, {})
+                format_circuit = circuit_by_format.get(api_format, {})
+                format_key_info["health_score"] = float(format_health.get("health_score") or 1.0)
+                format_key_info["circuit_breaker_open"] = bool(format_circuit.get("open", False))
+                grouped[api_format].append(format_key_info)

        # 直接返回分组对象，供前端使用
        return grouped


+# ========== Adapters ==========
+
+
+def _build_key_response(
+    key: ProviderAPIKey, api_key_plain: str | None = None
+) -> EndpointAPIKeyResponse:
+    """构建 Key 响应对象的辅助函数"""
+    try:
+        decrypted_key = crypto_service.decrypt(key.api_key)
+        masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
+    except Exception:
+        masked_key = "***ERROR***"
+
+    success_rate = key.success_count / key.request_count if key.request_count > 0 else 0.0
+    avg_response_time_ms = (
+        key.total_response_time_ms / key.success_count if key.success_count > 0 else 0.0
+    )
+
+    is_adaptive = key.rpm_limit is None
+    key_dict = key.__dict__.copy()
+    key_dict.pop("_sa_instance_state", None)
+
+    # 从 health_by_format 计算汇总字段（便于列表展示）
+    health_by_format = key.health_by_format or {}
+    circuit_by_format = key.circuit_breaker_by_format or {}
+
+    # 计算整体健康度（取所有格式中的最低值）
+    if health_by_format:
+        health_scores = [
+            float(h.get("health_score") or 1.0) for h in health_by_format.values()
+        ]
+        min_health_score = min(health_scores) if health_scores else 1.0
+        # 取最大的连续失败次数
+        max_consecutive = max(
+            (int(h.get("consecutive_failures") or 0) for h in health_by_format.values()),
+            default=0,
+        )
+        # 取最近的失败时间
+        failure_times = [
+            h.get("last_failure_at")
+            for h in health_by_format.values()
+            if h.get("last_failure_at")
+        ]
+        last_failure = max(failure_times) if failure_times else None
+    else:
+        min_health_score = 1.0
+        max_consecutive = 0
+        last_failure = None
+
+    # 检查是否有任何格式的熔断器打开
+    any_circuit_open = any(c.get("open", False) for c in circuit_by_format.values())
+
+    key_dict.update(
+        {
+            "api_key_masked": masked_key,
+            "api_key_plain": api_key_plain,
+            "success_rate": success_rate,
+            "avg_response_time_ms": round(avg_response_time_ms, 2),
+            "is_adaptive": is_adaptive,
+            "effective_limit": (
+                (key.learned_rpm_limit if key.learned_rpm_limit is not None else RPMDefaults.INITIAL_LIMIT)
+                if is_adaptive
+                else key.rpm_limit
+            ),
+            # 汇总字段
+            "health_score": min_health_score,
+            "consecutive_failures": max_consecutive,
+            "last_failure_at": last_failure,
+            "circuit_breaker_open": any_circuit_open,
+        }
+    )
+
+    # 防御性：确保 api_formats 存在（历史数据可能为空/缺失）
+    if "api_formats" not in key_dict or key_dict["api_formats"] is None:
+        key_dict["api_formats"] = []
+
+    return EndpointAPIKeyResponse(**key_dict)
+
+
@dataclass
-class AdminBatchUpdateKeyPriorityAdapter(AdminApiAdapter):
-    endpoint_id: str
-    priority_data: BatchUpdateKeyPriorityRequest
+class AdminListProviderKeysAdapter(AdminApiAdapter):
+    """获取 Provider 的所有 Keys"""
+
+    provider_id: str
+    skip: int
+    limit: int

    async def handle(self, context):  # type: ignore[override]
        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
+        provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
+        if not provider:
+            raise NotFoundException(f"Provider {self.provider_id} 不存在")

-        # 获取所有需要更新的 Key ID
-        key_ids = [item.key_id for item in self.priority_data.priorities]
-
-        # 验证所有 Key 都属于该 Endpoint
        keys = (
            db.query(ProviderAPIKey)
-            .filter(
-                ProviderAPIKey.id.in_(key_ids),
-                ProviderAPIKey.endpoint_id == self.endpoint_id,
-            )
+            .filter(ProviderAPIKey.provider_id == self.provider_id)
+            .order_by(ProviderAPIKey.internal_priority.asc(), ProviderAPIKey.created_at.asc())
+            .offset(self.skip)
+            .limit(self.limit)
            .all()
        )

-        if len(keys) != len(key_ids):
-            found_ids = {k.id for k in keys}
-            missing_ids = set(key_ids) - found_ids
-            raise InvalidRequestException(f"Keys 不属于该 Endpoint 或不存在: {missing_ids}")
+        return [_build_key_response(key) for key in keys]

-        # 批量更新优先级
-        key_map = {k.id: k for k in keys}
-        updated_count = 0
-        for item in self.priority_data.priorities:
-            key = key_map.get(item.key_id)
-            if key and key.internal_priority != item.internal_priority:
-                key.internal_priority = item.internal_priority
-                key.updated_at = datetime.now(timezone.utc)
-                updated_count += 1

+@dataclass
+class AdminCreateProviderKeyAdapter(AdminApiAdapter):
+    """为 Provider 添加 Key"""
+
+    provider_id: str
+    key_data: EndpointAPIKeyCreate
+
+    async def handle(self, context):  # type: ignore[override]
+        db = context.db
+        provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
+        if not provider:
+            raise NotFoundException(f"Provider {self.provider_id} 不存在")
+
+        # 验证 api_formats 必填
+        if not self.key_data.api_formats:
+            raise InvalidRequestException("api_formats 为必填字段")
+
+        # 允许同一个 API Key 在同一 Provider 下添加多次
+        # 用户可以为不同的 API 格式创建独立的配置记录，便于分开管理
+
+        encrypted_key = crypto_service.encrypt(self.key_data.api_key)
+        now = datetime.now(timezone.utc)
+
+        new_key = ProviderAPIKey(
+            id=str(uuid.uuid4()),
+            provider_id=self.provider_id,
+            api_formats=self.key_data.api_formats,
+            api_key=encrypted_key,
+            name=self.key_data.name,
+            note=self.key_data.note,
+            rate_multiplier=self.key_data.rate_multiplier,
+            rate_multipliers=self.key_data.rate_multipliers,  # 按 API 格式的成本倍率
+            internal_priority=self.key_data.internal_priority,
+            rpm_limit=self.key_data.rpm_limit,
+            allowed_models=self.key_data.allowed_models if self.key_data.allowed_models else None,
+            capabilities=self.key_data.capabilities if self.key_data.capabilities else None,
+            cache_ttl_minutes=self.key_data.cache_ttl_minutes,
+            max_probe_interval_minutes=self.key_data.max_probe_interval_minutes,
+            request_count=0,
+            success_count=0,
+            error_count=0,
+            total_response_time_ms=0,
+            health_by_format={},  # 按格式存储健康度
+            circuit_breaker_by_format={},  # 按格式存储熔断器状态
+            is_active=True,
+            last_used_at=None,
+            created_at=now,
+            updated_at=now,
+        )
+
+        db.add(new_key)
        db.commit()
+        db.refresh(new_key)

-        logger.info(f"[OK] 批量更新 Key 优先级: Endpoint={self.endpoint_id}, Updated={updated_count}/{len(key_ids)}")
-        return {"message": f"已更新 {updated_count} 个 Key 的优先级", "updated_count": updated_count}
+        logger.info(
+            f"[OK] 添加 Key: Provider={self.provider_id}, "
+            f"Formats={self.key_data.api_formats}, Key=***{self.key_data.api_key[-4:]}, ID={new_key.id}"
+        )
+
+        return _build_key_response(new_key, api_key_plain=self.key_data.api_key)
--- a/src/api/admin/endpoints/routes.py
+++ b/src/api/admin/endpoints/routes.py
@@ -67,8 +67,6 @@ async def list_provider_endpoints(
    - `custom_path`: 自定义路径
    - `timeout`: 超时时间（秒）
    - `max_retries`: 最大重试次数
-    - `max_concurrent`: 最大并发数
-    - `rate_limit`: 速率限制
    - `is_active`: 是否活跃
    - `total_keys`: Key 总数
    - `active_keys`: 活跃 Key 数量
@@ -107,8 +105,6 @@ async def create_provider_endpoint(
    - `headers`: 自定义请求头（可选）
    - `timeout`: 超时时间（秒，默认 300）
    - `max_retries`: 最大重试次数（默认 2）
-    - `max_concurrent`: 最大并发数（可选）
-    - `rate_limit`: 速率限制（可选）
    - `config`: 额外配置（可选）
    - `proxy`: 代理配置（可选）

@@ -145,8 +141,6 @@ async def get_endpoint(
    - `custom_path`: 自定义路径
    - `timeout`: 超时时间（秒）
    - `max_retries`: 最大重试次数
-    - `max_concurrent`: 最大并发数
-    - `rate_limit`: 速率限制
    - `is_active`: 是否活跃
    - `total_keys`: Key 总数
    - `active_keys`: 活跃 Key 数量
@@ -178,8 +172,6 @@ async def update_endpoint(
    - `headers`: 自定义请求头
    - `timeout`: 超时时间（秒）
    - `max_retries`: 最大重试次数
-    - `max_concurrent`: 最大并发数
-    - `rate_limit`: 速率限制
    - `is_active`: 是否活跃
    - `config`: 额外配置
    - `proxy`: 代理配置（设置为 null 可清除代理）
@@ -203,15 +195,15 @@ async def delete_endpoint(
    """
    删除 Endpoint

-    删除指定的 Endpoint，同时级联删除所有关联的 API Keys。
-    此操作不可逆，请谨慎使用。
+    删除指定的 Endpoint，会影响该 Provider 在该 API 格式下的路由能力。
+    Key 不会被删除，但包含该 API 格式的 Key 将无法被调度使用（直到重新创建该格式的 Endpoint）。

    **路径参数**:
    - `endpoint_id`: Endpoint ID

    **返回字段**:
    - `message`: 操作结果消息
-    - `deleted_keys_count`: 同时删除的 Key 数量
+    - `affected_keys_count`: 受影响的 Key 数量（包含该 API 格式）
    """
    adapter = AdminDeleteProviderEndpointAdapter(endpoint_id=endpoint_id)
    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -241,39 +233,33 @@ class AdminListProviderEndpointsAdapter(AdminApiAdapter):
            .all()
        )

-        endpoint_ids = [ep.id for ep in endpoints]
-        total_keys_map = {}
-        active_keys_map = {}
-        if endpoint_ids:
-            total_rows = (
-                db.query(ProviderAPIKey.endpoint_id, func.count(ProviderAPIKey.id).label("total"))
-                .filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids))
-                .group_by(ProviderAPIKey.endpoint_id)
-                .all()
-            )
-            total_keys_map = {row.endpoint_id: row.total for row in total_rows}
-
-            active_rows = (
-                db.query(ProviderAPIKey.endpoint_id, func.count(ProviderAPIKey.id).label("active"))
-                .filter(
-                    and_(
-                        ProviderAPIKey.endpoint_id.in_(endpoint_ids),
-                        ProviderAPIKey.is_active.is_(True),
-                    )
-                )
-                .group_by(ProviderAPIKey.endpoint_id)
-                .all()
-            )
-            active_keys_map = {row.endpoint_id: row.active for row in active_rows}
+        # Key 是 Provider 级别资源：按 key.api_formats 归类到各 Endpoint.api_format 下
+        keys = (
+            db.query(ProviderAPIKey.api_formats, ProviderAPIKey.is_active)
+            .filter(ProviderAPIKey.provider_id == self.provider_id)
+            .all()
+        )
+        total_keys_map: dict[str, int] = {}
+        active_keys_map: dict[str, int] = {}
+        for api_formats, is_active in keys:
+            for fmt in (api_formats or []):
+                total_keys_map[fmt] = total_keys_map.get(fmt, 0) + 1
+                if is_active:
+                    active_keys_map[fmt] = active_keys_map.get(fmt, 0) + 1

        result: List[ProviderEndpointResponse] = []
        for endpoint in endpoints:
+            endpoint_format = (
+                endpoint.api_format
+                if isinstance(endpoint.api_format, str)
+                else endpoint.api_format.value
+            )
            endpoint_dict = {
                **endpoint.__dict__,
                "provider_name": provider.name,
                "api_format": endpoint.api_format,
-                "total_keys": total_keys_map.get(endpoint.id, 0),
-                "active_keys": active_keys_map.get(endpoint.id, 0),
+                "total_keys": total_keys_map.get(endpoint_format, 0),
+                "active_keys": active_keys_map.get(endpoint_format, 0),
                "proxy": mask_proxy_password(endpoint.proxy),
            }
            endpoint_dict.pop("_sa_instance_state", None)
@@ -321,8 +307,6 @@ class AdminCreateProviderEndpointAdapter(AdminApiAdapter):
            headers=self.endpoint_data.headers,
            timeout=self.endpoint_data.timeout,
            max_retries=self.endpoint_data.max_retries,
-            max_concurrent=self.endpoint_data.max_concurrent,
-            rate_limit=self.endpoint_data.rate_limit,
            is_active=True,
            config=self.endpoint_data.config,
            proxy=self.endpoint_data.proxy.model_dump() if self.endpoint_data.proxy else None,
@@ -367,19 +351,23 @@ class AdminGetProviderEndpointAdapter(AdminApiAdapter):
            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")

        endpoint_obj, provider = endpoint
-        total_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == self.endpoint_id).count()
+        endpoint_format = (
+            endpoint_obj.api_format
+            if isinstance(endpoint_obj.api_format, str)
+            else endpoint_obj.api_format.value
        )
-        active_keys = (
-            db.query(ProviderAPIKey)
-            .filter(
-                and_(
-                    ProviderAPIKey.endpoint_id == self.endpoint_id,
-                    ProviderAPIKey.is_active.is_(True),
-                )
-            )
-            .count()
+        keys = (
+            db.query(ProviderAPIKey.api_formats, ProviderAPIKey.is_active)
+            .filter(ProviderAPIKey.provider_id == endpoint_obj.provider_id)
+            .all()
        )
+        total_keys = 0
+        active_keys = 0
+        for api_formats, is_active in keys:
+            if endpoint_format in (api_formats or []):
+                total_keys += 1
+                if is_active:
+                    active_keys += 1

        endpoint_dict = {
            k: v
@@ -431,19 +419,21 @@ class AdminUpdateProviderEndpointAdapter(AdminApiAdapter):
        provider = db.query(Provider).filter(Provider.id == endpoint.provider_id).first()
        logger.info(f"[OK] 更新 Endpoint: ID={self.endpoint_id}, Updates={list(update_data.keys())}")

-        total_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == self.endpoint_id).count()
+        endpoint_format = (
+            endpoint.api_format if isinstance(endpoint.api_format, str) else endpoint.api_format.value
        )
-        active_keys = (
-            db.query(ProviderAPIKey)
-            .filter(
-                and_(
-                    ProviderAPIKey.endpoint_id == self.endpoint_id,
-                    ProviderAPIKey.is_active.is_(True),
-                )
-            )
-            .count()
+        keys = (
+            db.query(ProviderAPIKey.api_formats, ProviderAPIKey.is_active)
+            .filter(ProviderAPIKey.provider_id == endpoint.provider_id)
+            .all()
        )
+        total_keys = 0
+        active_keys = 0
+        for api_formats, is_active in keys:
+            if endpoint_format in (api_formats or []):
+                total_keys += 1
+                if is_active:
+                    active_keys += 1

        endpoint_dict = {
            k: v
@@ -472,12 +462,26 @@ class AdminDeleteProviderEndpointAdapter(AdminApiAdapter):
        if not endpoint:
            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")

-        keys_count = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == self.endpoint_id).count()
+        endpoint_format = (
+            endpoint.api_format if isinstance(endpoint.api_format, str) else endpoint.api_format.value
+        )
+        keys = (
+            db.query(ProviderAPIKey.api_formats)
+            .filter(ProviderAPIKey.provider_id == endpoint.provider_id)
+            .all()
+        )
+        affected_keys_count = sum(
+            1 for (api_formats,) in keys if endpoint_format in (api_formats or [])
        )
        db.delete(endpoint)
        db.commit()

-        logger.warning(f"[DELETE] 删除 Endpoint: ID={self.endpoint_id}, 同时删除了 {keys_count} 个 Keys")
+        logger.warning(
+            f"[DELETE] 删除 Endpoint: ID={self.endpoint_id}, Format={endpoint_format}, "
+            f"AffectedKeys={affected_keys_count}"
+        )

-        return {"message": f"Endpoint {self.endpoint_id} 已删除", "deleted_keys_count": keys_count}
+        return {
+            "message": f"Endpoint {self.endpoint_id} 已删除",
+            "affected_keys_count": affected_keys_count,
+        }
--- a/src/api/admin/models/catalog.py
+++ b/src/api/admin/models/catalog.py
@@ -125,7 +125,6 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
                    ModelCatalogProviderDetail(
                        provider_id=provider.id,
                        provider_name=provider.name,
-                        provider_display_name=provider.display_name,
                        model_id=model.id,
                        target_model=model.provider_model_name,
                        # 显示有效价格
--- a/src/api/admin/models/global_models.py
+++ b/src/api/admin/models/global_models.py
@@ -452,7 +452,6 @@ class AdminGetGlobalModelProvidersAdapter(AdminApiAdapter):
                ModelCatalogProviderDetail(
                    provider_id=provider.id,
                    provider_name=provider.name,
-                    provider_display_name=provider.display_name,
                    model_id=model.id,
                    target_model=model.provider_model_name,
                    input_price_per_1m=model.get_effective_input_price(),
--- a/src/api/admin/monitoring/cache.py
+++ b/src/api/admin/monitoring/cache.py
@@ -819,7 +819,7 @@ class AdminListAffinitiesAdapter(AdminApiAdapter):
                "username": user.username if user else None,
                "email": user.email if user else None,
                "provider_id": provider_id,
-                "provider_name": provider.display_name if provider else None,
+                "provider_name": provider.name if provider else None,
                "endpoint_id": endpoint_id,
                "endpoint_api_format": (
                    endpoint.api_format if endpoint and endpoint.api_format else None
@@ -1369,9 +1369,7 @@ class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):
                                    for model, provider in models:
                                        # 检查是否是主模型名称
                                        if model.provider_model_name == mapping_name:
-                                            provider_names.append(
-                                                provider.display_name or provider.name
-                                            )
+                                            provider_names.append(provider.name)
                                            continue
                                        # 检查是否在映射列表中
                                        if model.provider_model_mappings:
@@ -1381,9 +1379,7 @@ class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):
                                                if isinstance(a, dict)
                                            ]
                                            if mapping_name in mapping_list:
-                                                provider_names.append(
-                                                    provider.display_name or provider.name
-                                                )
+                                                provider_names.append(provider.name)
                                    provider_names = sorted(list(set(provider_names)))

                                mappings.append({
@@ -1473,7 +1469,7 @@ class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):

                                    provider_model_mappings.append({
                                        "provider_id": provider_id,
-                                        "provider_name": provider.display_name or provider.name,
+                                        "provider_name": provider.name,
                                        "global_model_id": global_model_id,
                                        "global_model_name": global_model.name,
                                        "global_model_display_name": global_model.display_name,
--- a/src/api/admin/provider_query.py
+++ b/src/api/admin/provider_query.py
@@ -13,10 +13,11 @@ from sqlalchemy.orm import Session, joinedload

 from src.api.handlers.base.chat_adapter_base import get_adapter_class
 from src.api.handlers.base.cli_adapter_base import get_cli_adapter_class
+from src.config.constants import TimeoutDefaults
 from src.core.crypto import crypto_service
 from src.core.logger import logger
 from src.database.database import get_db
-from src.models.database import Provider, ProviderEndpoint, User
+from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint, User
 from src.utils.auth_utils import get_current_user

 router = APIRouter(prefix="/api/admin/provider-query", tags=["Provider Query"])
@@ -81,10 +82,13 @@ async def query_available_models(
    Returns:
        所有端点的模型列表（合并）
    """
-    # 获取提供商及其端点
+    # 获取提供商及其端点和 API Keys
    provider = (
        db.query(Provider)
-        .options(joinedload(Provider.endpoints).joinedload(ProviderEndpoint.api_keys))
+        .options(
+            joinedload(Provider.endpoints),
+            joinedload(Provider.api_keys),
+        )
        .filter(Provider.id == request.provider_id)
        .first()
    )
@@ -95,49 +99,70 @@ async def query_available_models(
    # 收集所有活跃端点的配置
    endpoint_configs: list[dict] = []

+    # 构建 api_format -> endpoint 映射
+    format_to_endpoint: dict[str, ProviderEndpoint] = {}
+    for endpoint in provider.endpoints:
+        if endpoint.is_active:
+            format_to_endpoint[endpoint.api_format] = endpoint
+
    if request.api_key_id:
-        # 指定了特定的 API Key，只使用该 Key 对应的端点
-        for endpoint in provider.endpoints:
-            for api_key in endpoint.api_keys:
-                if api_key.id == request.api_key_id:
-                    try:
-                        api_key_value = crypto_service.decrypt(api_key.api_key)
-                    except Exception as e:
-                        logger.error(f"Failed to decrypt API key: {e}")
-                        raise HTTPException(status_code=500, detail="Failed to decrypt API key")
-                    endpoint_configs.append({
-                        "api_key": api_key_value,
-                        "base_url": endpoint.base_url,
-                        "api_format": endpoint.api_format,
-                        "extra_headers": endpoint.headers,
-                    })
-                    break
-            if endpoint_configs:
-                break
+        # 指定了特定的 API Key（从 provider.api_keys 查找）
+        api_key = next(
+            (key for key in provider.api_keys if key.id == request.api_key_id),
+            None
+        )
+
+        if not api_key:
+            raise HTTPException(status_code=404, detail="API Key not found")
+
+        try:
+            api_key_value = crypto_service.decrypt(api_key.api_key)
+        except Exception as e:
+            logger.error(f"Failed to decrypt API key: {e}")
+            raise HTTPException(status_code=500, detail="Failed to decrypt API key")
+
+        # 根据 Key 的 api_formats 找对应的 Endpoint
+        key_formats = api_key.api_formats or []
+        for fmt in key_formats:
+            endpoint = format_to_endpoint.get(fmt)
+            if endpoint:
+                endpoint_configs.append({
+                    "api_key": api_key_value,
+                    "base_url": endpoint.base_url,
+                    "api_format": fmt,
+                    "extra_headers": endpoint.headers,
+                })

        if not endpoint_configs:
-            raise HTTPException(status_code=404, detail="API Key not found")
+            raise HTTPException(
+                status_code=400,
+                detail="No matching endpoint found for this API Key's formats"
+            )
    else:
-        # 遍历所有活跃端点，每个端点取第一个可用的 Key
+        # 遍历所有活跃端点，为每个端点找一个支持该格式的 Key
        for endpoint in provider.endpoints:
-            if not endpoint.is_active or not endpoint.api_keys:
+            if not endpoint.is_active:
                continue

-            # 找第一个可用的 Key
-            for api_key in endpoint.api_keys:
-                if api_key.is_active:
-                    try:
-                        api_key_value = crypto_service.decrypt(api_key.api_key)
-                    except Exception as e:
-                        logger.error(f"Failed to decrypt API key: {e}")
-                        continue  # 尝试下一个 Key
-                    endpoint_configs.append({
-                        "api_key": api_key_value,
-                        "base_url": endpoint.base_url,
-                        "api_format": endpoint.api_format,
-                        "extra_headers": endpoint.headers,
-                    })
-                    break  # 只取第一个可用的 Key
+            # 找第一个支持该格式的可用 Key
+            for api_key in provider.api_keys:
+                if not api_key.is_active:
+                    continue
+                key_formats = api_key.api_formats or []
+                if endpoint.api_format not in key_formats:
+                    continue
+                try:
+                    api_key_value = crypto_service.decrypt(api_key.api_key)
+                except Exception as e:
+                    logger.error(f"Failed to decrypt API key: {e}")
+                    continue
+                endpoint_configs.append({
+                    "api_key": api_key_value,
+                    "base_url": endpoint.base_url,
+                    "api_format": endpoint.api_format,
+                    "extra_headers": endpoint.headers,
+                })
+                break  # 只取第一个可用的 Key

        if not endpoint_configs:
            raise HTTPException(status_code=400, detail="No active API Key found for this provider")
@@ -214,7 +239,6 @@ async def query_available_models(
        "provider": {
            "id": provider.id,
            "name": provider.name,
-            "display_name": provider.display_name,
        },
    }

@@ -229,17 +253,14 @@ async def test_model(
    测试模型连接性

    向指定提供商的指定模型发送测试请求，验证模型是否可用
-
-    Args:
-        request: 测试请求
-
-    Returns:
-        测试结果
    """
-    # 获取提供商及其端点
+    # 获取提供商及其端点和 Keys
    provider = (
        db.query(Provider)
-        .options(joinedload(Provider.endpoints).joinedload(ProviderEndpoint.api_keys))
+        .options(
+            joinedload(Provider.endpoints),
+            joinedload(Provider.api_keys),
+        )
        .filter(Provider.id == request.provider_id)
        .first()
    )
@@ -247,28 +268,38 @@ async def test_model(
    if not provider:
        raise HTTPException(status_code=404, detail="Provider not found")

-    # 找到合适的端点和API Key
-    endpoint_config = None
+    # 构建 api_format -> endpoint 映射
+    format_to_endpoint: dict[str, ProviderEndpoint] = {}
+    for ep in provider.endpoints:
+        if ep.is_active:
+            format_to_endpoint[ep.api_format] = ep
+
+    # 找到合适的端点和 API Key
    endpoint = None
    api_key = None

    if request.api_key_id:
-        # 使用指定的API Key
-        for ep in provider.endpoints:
-            for key in ep.api_keys:
-                if key.id == request.api_key_id and key.is_active and ep.is_active:
-                    endpoint = ep
-                    api_key = key
+        # 使用指定的 API Key
+        api_key = next(
+            (key for key in provider.api_keys if key.id == request.api_key_id and key.is_active),
+            None
+        )
+        if api_key:
+            # 找到该 Key 支持的第一个活跃 Endpoint
+            for fmt in (api_key.api_formats or []):
+                if fmt in format_to_endpoint:
+                    endpoint = format_to_endpoint[fmt]
                    break
-            if endpoint:
-                break
    else:
        # 使用第一个可用的端点和密钥
        for ep in provider.endpoints:
-            if not ep.is_active or not ep.api_keys:
+            if not ep.is_active:
                continue
-            for key in ep.api_keys:
-                if key.is_active:
+            # 找支持该格式的第一个可用 Key
+            for key in provider.api_keys:
+                if not key.is_active:
+                    continue
+                if ep.api_format in (key.api_formats or []):
                    endpoint = ep
                    api_key = key
                    break
@@ -284,14 +315,14 @@ async def test_model(
        logger.error(f"[test-model] Failed to decrypt API key: {e}")
        raise HTTPException(status_code=500, detail="Failed to decrypt API key")

-    # 构建请求配置
+    # 构建请求配置（timeout 从 Provider 读取）
    endpoint_config = {
        "api_key": api_key_value,
        "api_key_id": api_key.id,  # 添加API Key ID用于用量记录
        "base_url": endpoint.base_url,
        "api_format": endpoint.api_format,
        "extra_headers": endpoint.headers,
-        "timeout": endpoint.timeout or 30.0,
+        "timeout": provider.timeout or TimeoutDefaults.HTTP_REQUEST,
    }

    try:
@@ -304,7 +335,6 @@ async def test_model(
                "provider": {
                    "id": provider.id,
                    "name": provider.name,
-                    "display_name": provider.display_name,
                },
                "model": request.model_name,
            }
@@ -325,7 +355,6 @@ async def test_model(
                    "provider": {
                        "id": provider.id,
                        "name": provider.name,
-                        "display_name": provider.display_name,
                    },
                    "model": request.model_name,
                }
@@ -415,7 +444,6 @@ async def test_model(
                "provider": {
                    "id": provider.id,
                    "name": provider.name,
-                    "display_name": provider.display_name,
                },
                "model": request.model_name,
                "endpoint": {
@@ -433,7 +461,6 @@ async def test_model(
            "provider": {
                "id": provider.id,
                "name": provider.name,
-                "display_name": provider.display_name,
            },
            "model": request.model_name,
            "endpoint": {
--- a/src/api/admin/provider_strategy.py
+++ b/src/api/admin/provider_strategy.py
@@ -78,7 +78,7 @@ async def get_provider_stats(
    """
    获取提供商统计数据

-    获取指定提供商的计费信息、RPM 使用情况和使用统计数据。
+    获取指定提供商的计费信息和使用统计数据。

    **路径参数**:
    - `provider_id`: 提供商 ID
@@ -96,10 +96,6 @@ async def get_provider_stats(
      - `monthly_used_usd`: 月度已使用
      - `quota_remaining_usd`: 剩余配额
      - `quota_expires_at`: 配额过期时间
-    - `rpm_info`: RPM 信息
-      - `rpm_limit`: RPM 限制
-      - `rpm_used`: 已使用 RPM
-      - `rpm_reset_at`: RPM 重置时间
    - `usage_stats`: 使用统计
      - `total_requests`: 总请求数
      - `successful_requests`: 成功请求数
@@ -165,7 +161,6 @@ class AdminProviderBillingAdapter(AdminApiAdapter):
        provider.billing_type = config.billing_type
        provider.monthly_quota_usd = config.monthly_quota_usd
        provider.quota_reset_day = config.quota_reset_day
-        provider.rpm_limit = config.rpm_limit
        provider.provider_priority = config.provider_priority

        from dateutil import parser
@@ -262,13 +257,6 @@ class AdminProviderStatsAdapter(AdminApiAdapter):
                        provider.quota_expires_at.isoformat() if provider.quota_expires_at else None
                    ),
                },
-                "rpm_info": {
-                    "rpm_limit": provider.rpm_limit,
-                    "rpm_used": provider.rpm_used,
-                    "rpm_reset_at": (
-                        provider.rpm_reset_at.isoformat() if provider.rpm_reset_at else None
-                    ),
-                },
                "usage_stats": {
                    "total_requests": total_requests,
                    "successful_requests": total_success,
@@ -296,8 +284,6 @@ class AdminProviderResetQuotaAdapter(AdminApiAdapter):

        old_used = provider.monthly_used_usd
        provider.monthly_used_usd = 0.0
-        provider.rpm_used = 0
-        provider.rpm_reset_at = None
        db.commit()

        logger.info(f"Manually reset quota for provider {provider.name}")
--- a/src/api/admin/providers/models.py
+++ b/src/api/admin/providers/models.py
@@ -338,27 +338,29 @@ async def import_models_from_upstream(
    """
    从上游提供商导入模型

-    从上游提供商导入模型列表。如果全局模型不存在，将自动创建。
+    从上游提供商导入模型列表。导入的模型作为独立的 ProviderModel 存储，
+    不会自动创建 GlobalModel。后续需要手动关联 GlobalModel 才能参与路由。

    **流程说明**:
-    1. 根据 model_ids 检查全局模型是否存在（按 name 匹配）
-    2. 如不存在，自动创建新的 GlobalModel（使用默认免费配置）
-    3. 创建 Model 关联到当前 Provider
-    4. 如模型已关联，则记录到成功列表中
+    1. 检查模型是否已存在于当前 Provider（按 provider_model_name 匹配）
+    2. 创建新的 ProviderModel（global_model_id = NULL）
+    3. 支持设置价格覆盖（tiered_pricing, price_per_request）

    **路径参数**:
    - `provider_id`: 提供商 ID

    **请求体字段**:
    - `model_ids`: 模型 ID 数组（必填，每个 ID 长度 1-100 字符）
+    - `tiered_pricing`: 可选的阶梯计费配置（应用于所有导入的模型）
+    - `price_per_request`: 可选的按次计费价格（应用于所有导入的模型）

    **返回字段**:
    - `success`: 成功导入的模型数组，每项包含：
      - `model_id`: 模型 ID
-      - `global_model_id`: 全局模型 ID
-      - `global_model_name`: 全局模型名称
      - `provider_model_id`: 提供商模型 ID
-      - `created_global_model`: 是否新创建了全局模型
+      - `global_model_id`: 全局模型 ID（如果已关联）
+      - `global_model_name`: 全局模型名称（如果已关联）
+      - `created_global_model`: 是否新创建了全局模型（始终为 false）
    - `errors`: 失败的模型数组，每项包含：
      - `model_id`: 模型 ID
      - `error`: 错误信息
@@ -638,7 +640,7 @@ class AdminBatchAssignModelsToProviderAdapter(AdminApiAdapter):

@dataclass
 class AdminImportFromUpstreamAdapter(AdminApiAdapter):
-    """从上游提供商导入模型"""
+    """从上游提供商导入模型（不创建 GlobalModel，作为独立 ProviderModel）"""

    provider_id: str
    payload: ImportFromUpstreamRequest
@@ -652,16 +654,13 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
        success: list[ImportFromUpstreamSuccessItem] = []
        errors: list[ImportFromUpstreamErrorItem] = []

-        # 默认阶梯计费配置（免费）
-        default_tiered_pricing = {
-            "tiers": [
-                {
-                    "up_to": None,
-                    "input_price_per_1m": 0.0,
-                    "output_price_per_1m": 0.0,
-                }
-            ]
-        }
+        # 获取价格覆盖配置
+        tiered_pricing = None
+        price_per_request = None
+        if hasattr(self.payload, 'tiered_pricing') and self.payload.tiered_pricing:
+            tiered_pricing = self.payload.tiered_pricing
+        if hasattr(self.payload, 'price_per_request') and self.payload.price_per_request is not None:
+            price_per_request = self.payload.price_per_request

        for model_id in self.payload.model_ids:
            # 输入验证：检查 model_id 长度
@@ -678,56 +677,37 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
                # 使用 savepoint 确保单个模型导入的原子性
                savepoint = db.begin_nested()
                try:
-                    # 1. 检查是否已存在同名的 GlobalModel
-                    global_model = (
-                        db.query(GlobalModel).filter(GlobalModel.name == model_id).first()
-                    )
-                    created_global_model = False
-
-                    if not global_model:
-                        # 2. 创建新的 GlobalModel
-                        global_model = GlobalModel(
-                            name=model_id,
-                            display_name=model_id,
-                            default_tiered_pricing=default_tiered_pricing,
-                            is_active=True,
-                        )
-                        db.add(global_model)
-                        db.flush()
-                        created_global_model = True
-                        logger.info(
-                            f"Created new GlobalModel: {model_id} during upstream import"
-                        )
-
-                    # 3. 检查是否已存在关联
+                    # 1. 检查是否已存在同名的 ProviderModel
                    existing = (
                        db.query(Model)
                        .filter(
                            Model.provider_id == self.provider_id,
-                            Model.global_model_id == global_model.id,
+                            Model.provider_model_name == model_id,
                        )
                        .first()
                    )
                    if existing:
-                        # 已存在关联，提交 savepoint 并记录成功
+                        # 已存在，提交 savepoint 并记录成功
                        savepoint.commit()
                        success.append(
                            ImportFromUpstreamSuccessItem(
                                model_id=model_id,
-                                global_model_id=global_model.id,
-                                global_model_name=global_model.name,
+                                global_model_id=existing.global_model_id or "",
+                                global_model_name=existing.global_model.name if existing.global_model else "",
                                provider_model_id=existing.id,
-                                created_global_model=created_global_model,
+                                created_global_model=False,
                            )
                        )
                        continue

-                    # 4. 创建新的 Model 记录
+                    # 2. 创建新的 Model 记录（不关联 GlobalModel）
                    new_model = Model(
                        provider_id=self.provider_id,
-                        global_model_id=global_model.id,
-                        provider_model_name=global_model.name,
+                        global_model_id=None,  # 独立模型，不关联 GlobalModel
+                        provider_model_name=model_id,
                        is_active=True,
+                        tiered_pricing=tiered_pricing,
+                        price_per_request=price_per_request,
                    )
                    db.add(new_model)
                    db.flush()
@@ -737,12 +717,15 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
                    success.append(
                        ImportFromUpstreamSuccessItem(
                            model_id=model_id,
-                            global_model_id=global_model.id,
-                            global_model_name=global_model.name,
+                            global_model_id="",  # 未关联
+                            global_model_name="",  # 未关联
                            provider_model_id=new_model.id,
-                            created_global_model=created_global_model,
+                            created_global_model=False,
                        )
                    )
+                    logger.info(
+                        f"Created independent ProviderModel: {model_id} for provider {provider.name}"
+                    )
                except Exception as e:
                    # 回滚到 savepoint
                    savepoint.rollback()
@@ -753,11 +736,9 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):

        db.commit()
        logger.info(
-            f"Imported {len(success)} models from upstream to provider {provider.name} by {context.user.username}"
+            f"Imported {len(success)} independent models to provider {provider.name} by {context.user.username}"
        )

-        # 清除 /v1/models 列表缓存
-        if success:
-            await invalidate_models_list_cache()
+        # 不需要清除 /v1/models 缓存，因为独立模型不参与路由

        return ImportFromUpstreamResponse(success=success, errors=errors)
--- a/src/api/admin/providers/routes.py
+++ b/src/api/admin/providers/routes.py
@@ -41,8 +41,7 @@ async def list_providers(

    **返回字段**:
    - `id`: 提供商 ID
-    - `name`: 提供商名称（唯一标识）
-    - `display_name`: 显示名称
+    - `name`: 提供商名称（唯一）
    - `api_format`: API 格式（如 claude、openai、gemini 等）
    - `base_url`: API 基础 URL
    - `api_key`: API 密钥（脱敏显示）
@@ -63,8 +62,7 @@ async def create_provider(request: Request, db: Session = Depends(get_db)):
    创建一个新的 AI 模型提供商配置。

    **请求体字段**:
-    - `name`: 提供商名称（必填，唯一，用于系统标识）
-    - `display_name`: 显示名称（必填）
+    - `name`: 提供商名称（必填，唯一）
    - `description`: 描述信息（可选）
    - `website`: 官网地址（可选）
    - `billing_type`: 计费类型（可选，pay_as_you_go/subscription/prepaid，默认 pay_as_you_go）
@@ -72,16 +70,17 @@ async def create_provider(request: Request, db: Session = Depends(get_db)):
    - `quota_reset_day`: 配额重置日期（1-31）（可选）
    - `quota_last_reset_at`: 上次配额重置时间（可选）
    - `quota_expires_at`: 配额过期时间（可选）
-    - `rpm_limit`: 每分钟请求数限制（可选）
    - `provider_priority`: 提供商优先级（数字越小优先级越高，默认 100）
    - `is_active`: 是否启用（默认 true）
    - `concurrent_limit`: 并发限制（可选）
+    - `timeout`: 请求超时（秒，可选）
+    - `max_retries`: 最大重试次数（可选）
+    - `proxy`: 代理配置（可选）
    - `config`: 额外配置信息（JSON，可选）

    **返回字段**:
    - `id`: 新创建的提供商 ID
    - `name`: 提供商名称
-    - `display_name`: 显示名称
    - `message`: 成功提示信息
    """
    adapter = AdminCreateProviderAdapter()
@@ -100,7 +99,6 @@ async def update_provider(provider_id: str, request: Request, db: Session = Depe

    **请求体字段**（所有字段可选）:
    - `name`: 提供商名称
-    - `display_name`: 显示名称
    - `description`: 描述信息
    - `website`: 官网地址
    - `billing_type`: 计费类型（pay_as_you_go/subscription/prepaid）
@@ -108,10 +106,12 @@ async def update_provider(provider_id: str, request: Request, db: Session = Depe
    - `quota_reset_day`: 配额重置日期（1-31）
    - `quota_last_reset_at`: 上次配额重置时间
    - `quota_expires_at`: 配额过期时间
-    - `rpm_limit`: 每分钟请求数限制
    - `provider_priority`: 提供商优先级
    - `is_active`: 是否启用
    - `concurrent_limit`: 并发限制
+    - `timeout`: 请求超时（秒）
+    - `max_retries`: 最大重试次数
+    - `proxy`: 代理配置
    - `config`: 额外配置信息（JSON）

    **返回字段**:
@@ -165,7 +165,6 @@ class AdminListProvidersAdapter(AdminApiAdapter):
                {
                    "id": provider.id,
                    "name": provider.name,
-                    "display_name": provider.display_name,
                    "api_format": api_format.value if api_format else None,
                    "base_url": base_url,
                    "api_key": "***" if api_key else None,
@@ -217,7 +216,6 @@ class AdminCreateProviderAdapter(AdminApiAdapter):
            # 创建 Provider 对象
            provider = Provider(
                name=validated_data.name,
-                display_name=validated_data.display_name,
                description=validated_data.description,
                website=validated_data.website,
                billing_type=billing_type,
@@ -225,10 +223,12 @@ class AdminCreateProviderAdapter(AdminApiAdapter):
                quota_reset_day=validated_data.quota_reset_day,
                quota_last_reset_at=validated_data.quota_last_reset_at,
                quota_expires_at=validated_data.quota_expires_at,
-                rpm_limit=validated_data.rpm_limit,
                provider_priority=validated_data.provider_priority,
                is_active=validated_data.is_active,
                concurrent_limit=validated_data.concurrent_limit,
+                timeout=validated_data.timeout,
+                max_retries=validated_data.max_retries,
+                proxy=validated_data.proxy.model_dump() if validated_data.proxy else None,
                config=validated_data.config,
            )

@@ -248,7 +248,6 @@ class AdminCreateProviderAdapter(AdminApiAdapter):
            return {
                "id": provider.id,
                "name": provider.name,
-                "display_name": provider.display_name,
                "message": "提供商创建成功",
            }
        except InvalidRequestException:
@@ -291,6 +290,9 @@ class AdminUpdateProviderAdapter(AdminApiAdapter):
                if field == "billing_type" and value is not None:
                    # billing_type 需要转换为枚举
                    setattr(provider, field, ProviderBillingType(value))
+                elif field == "proxy" and value is not None:
+                    # proxy 需要转换为 dict（如果是 Pydantic 模型）
+                    setattr(provider, field, value if isinstance(value, dict) else value.model_dump())
                else:
                    setattr(provider, field, value)

--- a/src/api/admin/providers/summary.py
+++ b/src/api/admin/providers/summary.py
@@ -48,7 +48,6 @@ async def get_providers_summary(
    **返回字段**（数组，每项包含）:
    - `id`: 提供商 ID
    - `name`: 提供商名称
-    - `display_name`: 显示名称
    - `description`: 描述信息
    - `website`: 官网地址
    - `provider_priority`: 优先级
@@ -59,9 +58,9 @@ async def get_providers_summary(
    - `quota_reset_day`: 配额重置日期
    - `quota_last_reset_at`: 上次配额重置时间
    - `quota_expires_at`: 配额过期时间
-    - `rpm_limit`: RPM 限制
-    - `rpm_used`: 已使用 RPM
-    - `rpm_reset_at`: RPM 重置时间
+    - `timeout`: 默认请求超时（秒）
+    - `max_retries`: 默认最大重试次数
+    - `proxy`: 默认代理配置
    - `total_endpoints`: 端点总数
    - `active_endpoints`: 活跃端点数
    - `total_keys`: 密钥总数
@@ -96,7 +95,6 @@ async def get_provider_summary(
    **返回字段**:
    - `id`: 提供商 ID
    - `name`: 提供商名称
-    - `display_name`: 显示名称
    - `description`: 描述信息
    - `website`: 官网地址
    - `provider_priority`: 优先级
@@ -107,9 +105,9 @@ async def get_provider_summary(
    - `quota_reset_day`: 配额重置日期
    - `quota_last_reset_at`: 上次配额重置时间
    - `quota_expires_at`: 配额过期时间
-    - `rpm_limit`: RPM 限制
-    - `rpm_used`: 已使用 RPM
-    - `rpm_reset_at`: RPM 重置时间
+    - `timeout`: 默认请求超时（秒）
+    - `max_retries`: 默认最大重试次数
+    - `proxy`: 默认代理配置
    - `total_endpoints`: 端点总数
    - `active_endpoints`: 活跃端点数
    - `total_keys`: 密钥总数
@@ -185,13 +183,13 @@ async def update_provider_settings(
    """
    更新提供商基础配置

-    更新提供商的基础配置信息，如显示名称、描述、优先级等。只需传入需要更新的字段。
+    更新提供商的基础配置信息，如名称、描述、优先级等。只需传入需要更新的字段。

    **路径参数**:
    - `provider_id`: 提供商 ID

    **请求体字段**（所有字段可选）:
-    - `display_name`: 显示名称
+    - `name`: 提供商名称
    - `description`: 描述信息
    - `website`: 官网地址
    - `provider_priority`: 优先级
@@ -199,9 +197,10 @@ async def update_provider_settings(
    - `billing_type`: 计费类型
    - `monthly_quota_usd`: 月度配额（美元）
    - `quota_reset_day`: 配额重置日期
-    - `quota_last_reset_at`: 上次配额重置时间
    - `quota_expires_at`: 配额过期时间
-    - `rpm_limit`: RPM 限制
+    - `timeout`: 默认请求超时（秒）
+    - `max_retries`: 默认最大重试次数
+    - `proxy`: 默认代理配置

    **返回字段**: 返回更新后的提供商摘要信息（与 GET /summary 接口返回格式相同）
    """
@@ -215,18 +214,18 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp

    total_endpoints = len(endpoints)
    active_endpoints = sum(1 for e in endpoints if e.is_active)
-    endpoint_ids = [e.id for e in endpoints]

    # Key 统计（合并为单个查询）
-    total_keys = 0
-    active_keys = 0
-    if endpoint_ids:
-        key_stats = db.query(
+    key_stats = (
+        db.query(
            func.count(ProviderAPIKey.id).label("total"),
            func.sum(case((ProviderAPIKey.is_active == True, 1), else_=0)).label("active"),
-        ).filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids)).first()
-        total_keys = key_stats.total or 0
-        active_keys = int(key_stats.active or 0)
+        )
+        .filter(ProviderAPIKey.provider_id == provider.id)
+        .first()
+    )
+    total_keys = key_stats.total or 0
+    active_keys = int(key_stats.active or 0)

    # Model 统计（合并为单个查询）
    model_stats = db.query(
@@ -238,25 +237,34 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp

    api_formats = [e.api_format for e in endpoints]

-    # 优化: 一次性加载所有 endpoint 的 keys，避免 N+1 查询
-    all_keys = []
-    if endpoint_ids:
-        all_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids)).all()
-        )
+    # 优化: 一次性加载 Provider 的 keys，避免 N+1 查询
+    all_keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.provider_id == provider.id).all()

-    # 按 endpoint_id 分组 keys
-    keys_by_endpoint: dict[str, list[ProviderAPIKey]] = {}
+    # 按 api_formats 分组 keys（通过 api_formats 关联）
+    format_to_endpoint_id: dict[str, str] = {e.api_format: e.id for e in endpoints}
+    keys_by_endpoint: dict[str, list[ProviderAPIKey]] = {e.id: [] for e in endpoints}
    for key in all_keys:
-        if key.endpoint_id not in keys_by_endpoint:
-            keys_by_endpoint[key.endpoint_id] = []
-        keys_by_endpoint[key.endpoint_id].append(key)
+        formats = key.api_formats or []
+        for fmt in formats:
+            endpoint_id = format_to_endpoint_id.get(fmt)
+            if endpoint_id:
+                keys_by_endpoint[endpoint_id].append(key)

    endpoint_health_map: dict[str, float] = {}
    for endpoint in endpoints:
        keys = keys_by_endpoint.get(endpoint.id, [])
        if keys:
-            health_scores = [k.health_score for k in keys if k.health_score is not None]
+            # 从 health_by_format 获取对应格式的健康度
+            api_fmt = endpoint.api_format
+            health_scores = []
+            for k in keys:
+                health_by_format = k.health_by_format or {}
+                if api_fmt in health_by_format:
+                    score = health_by_format[api_fmt].get("health_score")
+                    if score is not None:
+                        health_scores.append(float(score))
+                else:
+                    health_scores.append(1.0)  # 默认健康度
            avg_health = sum(health_scores) / len(health_scores) if health_scores else 1.0
            endpoint_health_map[endpoint.id] = avg_health
        else:
@@ -284,7 +292,6 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
    return ProviderWithEndpointsSummary(
        id=provider.id,
        name=provider.name,
-        display_name=provider.display_name,
        description=provider.description,
        website=provider.website,
        provider_priority=provider.provider_priority,
@@ -295,9 +302,9 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
        quota_reset_day=provider.quota_reset_day,
        quota_last_reset_at=provider.quota_last_reset_at,
        quota_expires_at=provider.quota_expires_at,
-        rpm_limit=provider.rpm_limit,
-        rpm_used=provider.rpm_used,
-        rpm_reset_at=provider.rpm_reset_at,
+        timeout=provider.timeout,
+        max_retries=provider.max_retries,
+        proxy=provider.proxy,
        total_endpoints=total_endpoints,
        active_endpoints=active_endpoints,
        total_keys=total_keys,
@@ -341,7 +348,7 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
        if not endpoint_ids:
            response = ProviderEndpointHealthMonitorResponse(
                provider_id=provider.id,
-                provider_name=provider.display_name or provider.name,
+                provider_name=provider.name,
                generated_at=now,
                endpoints=[],
            )
@@ -416,7 +423,7 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):

        response = ProviderEndpointHealthMonitorResponse(
            provider_id=provider.id,
-            provider_name=provider.display_name or provider.name,
+            provider_name=provider.name,
            generated_at=now,
            endpoints=endpoint_monitors,
        )
--- a/src/api/admin/system.py
+++ b/src/api/admin/system.py
@@ -730,36 +730,6 @@ class AdminExportConfigAdapter(AdminApiAdapter):
            )
            endpoints_data = []
            for ep in endpoints:
-                # 导出 Endpoint Keys
-                keys = (
-                    db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == ep.id).all()
-                )
-                keys_data = []
-                for key in keys:
-                    # 解密 API Key
-                    try:
-                        decrypted_key = crypto_service.decrypt(key.api_key)
-                    except Exception:
-                        decrypted_key = ""
-
-                    keys_data.append(
-                        {
-                            "api_key": decrypted_key,
-                            "name": key.name,
-                            "note": key.note,
-                            "rate_multiplier": key.rate_multiplier,
-                            "internal_priority": key.internal_priority,
-                            "global_priority": key.global_priority,
-                            "max_concurrent": key.max_concurrent,
-                            "rate_limit": key.rate_limit,
-                            "daily_limit": key.daily_limit,
-                            "monthly_limit": key.monthly_limit,
-                            "allowed_models": key.allowed_models,
-                            "capabilities": key.capabilities,
-                            "is_active": key.is_active,
-                        }
-                    )
-
                endpoints_data.append(
                    {
                        "api_format": ep.api_format,
@@ -767,12 +737,44 @@ class AdminExportConfigAdapter(AdminApiAdapter):
                        "headers": ep.headers,
                        "timeout": ep.timeout,
                        "max_retries": ep.max_retries,
-                        "max_concurrent": ep.max_concurrent,
-                        "rate_limit": ep.rate_limit,
                        "is_active": ep.is_active,
                        "custom_path": ep.custom_path,
                        "config": ep.config,
-                        "keys": keys_data,
+                        "proxy": ep.proxy,
+                    }
+                )
+
+            # 导出 Provider Keys（按 provider_id 归属，包含 api_formats）
+            keys = (
+                db.query(ProviderAPIKey)
+                .filter(ProviderAPIKey.provider_id == provider.id)
+                .order_by(ProviderAPIKey.internal_priority.asc(), ProviderAPIKey.created_at.asc())
+                .all()
+            )
+            keys_data = []
+            for key in keys:
+                # 解密 API Key
+                try:
+                    decrypted_key = crypto_service.decrypt(key.api_key)
+                except Exception:
+                    decrypted_key = ""
+
+                keys_data.append(
+                    {
+                        "api_key": decrypted_key,
+                        "name": key.name,
+                        "note": key.note,
+                        "api_formats": key.api_formats or [],
+                        "rate_multiplier": key.rate_multiplier,
+                        "rate_multipliers": key.rate_multipliers,
+                        "internal_priority": key.internal_priority,
+                        "global_priority": key.global_priority,
+                        "rpm_limit": key.rpm_limit,
+                        "allowed_models": key.allowed_models,
+                        "capabilities": key.capabilities,
+                        "cache_ttl_minutes": key.cache_ttl_minutes,
+                        "max_probe_interval_minutes": key.max_probe_interval_minutes,
+                        "is_active": key.is_active,
                    }
                )

@@ -804,24 +806,26 @@ class AdminExportConfigAdapter(AdminApiAdapter):
            providers_data.append(
                {
                    "name": provider.name,
-                    "display_name": provider.display_name,
                    "description": provider.description,
                    "website": provider.website,
                    "billing_type": provider.billing_type.value if provider.billing_type else None,
                    "monthly_quota_usd": provider.monthly_quota_usd,
                    "quota_reset_day": provider.quota_reset_day,
-                    "rpm_limit": provider.rpm_limit,
                    "provider_priority": provider.provider_priority,
                    "is_active": provider.is_active,
                    "concurrent_limit": provider.concurrent_limit,
+                    "timeout": provider.timeout,
+                    "max_retries": provider.max_retries,
+                    "proxy": provider.proxy,
                    "config": provider.config,
                    "endpoints": endpoints_data,
+                    "api_keys": keys_data,
                    "models": models_data,
                }
            )

        return {
-            "version": "1.0",
+            "version": "2.0",
            "exported_at": datetime.now(timezone.utc).isoformat(),
            "global_models": global_models_data,
            "providers": providers_data,
@@ -850,7 +854,7 @@ class AdminImportConfigAdapter(AdminApiAdapter):

        # 验证配置版本
        version = payload.get("version")
-        if version != "1.0":
+        if version != "2.0":
            raise InvalidRequestException(f"不支持的配置版本: {version}")

        # 获取导入选项
@@ -939,8 +943,8 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                        )
                    elif merge_mode == "overwrite":
                        # 更新现有记录
-                        existing_provider.display_name = prov_data.get(
-                            "display_name", existing_provider.display_name
+                        existing_provider.name = prov_data.get(
+                            "name", existing_provider.name
                        )
                        existing_provider.description = prov_data.get("description")
                        existing_provider.website = prov_data.get("website")
@@ -954,7 +958,6 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                        existing_provider.quota_reset_day = prov_data.get(
                            "quota_reset_day", 30
                        )
-                        existing_provider.rpm_limit = prov_data.get("rpm_limit")
                        existing_provider.provider_priority = prov_data.get(
                            "provider_priority", 100
                        )
@@ -962,6 +965,11 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                        existing_provider.concurrent_limit = prov_data.get(
                            "concurrent_limit"
                        )
+                        existing_provider.timeout = prov_data.get("timeout", existing_provider.timeout)
+                        existing_provider.max_retries = prov_data.get(
+                            "max_retries", existing_provider.max_retries
+                        )
+                        existing_provider.proxy = prov_data.get("proxy", existing_provider.proxy)
                        existing_provider.config = prov_data.get("config")
                        existing_provider.updated_at = datetime.now(timezone.utc)
                        stats["providers"]["updated"] += 1
@@ -974,16 +982,17 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                    new_provider = Provider(
                        id=str(uuid.uuid4()),
                        name=prov_data["name"],
-                        display_name=prov_data.get("display_name", prov_data["name"]),
                        description=prov_data.get("description"),
                        website=prov_data.get("website"),
                        billing_type=billing_type,
                        monthly_quota_usd=prov_data.get("monthly_quota_usd"),
                        quota_reset_day=prov_data.get("quota_reset_day", 30),
-                        rpm_limit=prov_data.get("rpm_limit"),
                        provider_priority=prov_data.get("provider_priority", 100),
                        is_active=prov_data.get("is_active", True),
                        concurrent_limit=prov_data.get("concurrent_limit"),
+                        timeout=prov_data.get("timeout"),
+                        max_retries=prov_data.get("max_retries"),
+                        proxy=prov_data.get("proxy"),
                        config=prov_data.get("config"),
                    )
                    db.add(new_provider)
@@ -1003,7 +1012,6 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                    )

                    if existing_ep:
-                        endpoint_id = existing_ep.id
                        if merge_mode == "skip":
                            stats["endpoints"]["skipped"] += 1
                        elif merge_mode == "error":
@@ -1017,11 +1025,10 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                            existing_ep.headers = ep_data.get("headers")
                            existing_ep.timeout = ep_data.get("timeout", 300)
                            existing_ep.max_retries = ep_data.get("max_retries", 2)
-                            existing_ep.max_concurrent = ep_data.get("max_concurrent")
-                            existing_ep.rate_limit = ep_data.get("rate_limit")
                            existing_ep.is_active = ep_data.get("is_active", True)
                            existing_ep.custom_path = ep_data.get("custom_path")
                            existing_ep.config = ep_data.get("config")
+                            existing_ep.proxy = ep_data.get("proxy")
                            existing_ep.updated_at = datetime.now(timezone.utc)
                            stats["endpoints"]["updated"] += 1
                    else:
@@ -1033,68 +1040,106 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                            headers=ep_data.get("headers"),
                            timeout=ep_data.get("timeout", 300),
                            max_retries=ep_data.get("max_retries", 2),
-                            max_concurrent=ep_data.get("max_concurrent"),
-                            rate_limit=ep_data.get("rate_limit"),
                            is_active=ep_data.get("is_active", True),
                            custom_path=ep_data.get("custom_path"),
                            config=ep_data.get("config"),
+                            proxy=ep_data.get("proxy"),
                        )
                        db.add(new_ep)
                        db.flush()
-                        endpoint_id = new_ep.id
                        stats["endpoints"]["created"] += 1

-                    # 导入 Keys
-                    # 获取当前 endpoint 下所有已有的 keys，用于去重
-                    existing_keys = (
-                        db.query(ProviderAPIKey)
-                        .filter(ProviderAPIKey.endpoint_id == endpoint_id)
-                        .all()
-                    )
-                    # 解密已有 keys 用于比对
-                    existing_key_values = set()
-                    for ek in existing_keys:
-                        try:
-                            decrypted = crypto_service.decrypt(ek.api_key)
-                            existing_key_values.add(decrypted)
-                        except Exception:
-                            pass
+                # 导入 Provider Keys（按 provider_id 归属）
+                endpoint_format_rows = (
+                    db.query(ProviderEndpoint.api_format)
+                    .filter(ProviderEndpoint.provider_id == provider_id)
+                    .all()
+                )
+                endpoint_formats: set[str] = set()
+                for (api_format,) in endpoint_format_rows:
+                    fmt = api_format.value if hasattr(api_format, "value") else str(api_format)
+                    endpoint_formats.add(fmt.strip().upper())
+                existing_keys = (
+                    db.query(ProviderAPIKey)
+                    .filter(ProviderAPIKey.provider_id == provider_id)
+                    .all()
+                )
+                existing_key_values = set()
+                for ek in existing_keys:
+                    try:
+                        decrypted = crypto_service.decrypt(ek.api_key)
+                        existing_key_values.add(decrypted)
+                    except Exception:
+                        pass

-                    for key_data in ep_data.get("keys", []):
-                        if not key_data.get("api_key"):
-                            stats["errors"].append(
-                                f"跳过空 API Key (Endpoint: {ep_data['api_format']})"
-                            )
-                            continue
-
-                        # 检查是否已存在相同的 Key（通过明文比对）
-                        if key_data["api_key"] in existing_key_values:
-                            stats["keys"]["skipped"] += 1
-                            continue
-
-                        encrypted_key = crypto_service.encrypt(key_data["api_key"])
-
-                        new_key = ProviderAPIKey(
-                            id=str(uuid.uuid4()),
-                            endpoint_id=endpoint_id,
-                            api_key=encrypted_key,
-                            name=key_data.get("name"),
-                            note=key_data.get("note"),
-                            rate_multiplier=key_data.get("rate_multiplier", 1.0),
-                            internal_priority=key_data.get("internal_priority", 100),
-                            global_priority=key_data.get("global_priority"),
-                            max_concurrent=key_data.get("max_concurrent"),
-                            rate_limit=key_data.get("rate_limit"),
-                            daily_limit=key_data.get("daily_limit"),
-                            monthly_limit=key_data.get("monthly_limit"),
-                            allowed_models=key_data.get("allowed_models"),
-                            capabilities=key_data.get("capabilities"),
-                            is_active=key_data.get("is_active", True),
+                for key_data in prov_data.get("api_keys", []):
+                    if not key_data.get("api_key"):
+                        stats["errors"].append(
+                            f"跳过空 API Key (Provider: {prov_data['name']})"
                        )
-                        db.add(new_key)
-                        # 添加到已有集合，防止同一批导入中重复
-                        existing_key_values.add(key_data["api_key"])
-                        stats["keys"]["created"] += 1
+                        continue
+
+                    plaintext_key = key_data["api_key"]
+                    if plaintext_key in existing_key_values:
+                        stats["keys"]["skipped"] += 1
+                        continue
+
+                    raw_formats = key_data.get("api_formats") or []
+                    if not isinstance(raw_formats, list) or len(raw_formats) == 0:
+                        stats["errors"].append(
+                            f"跳过无 api_formats 的 Key (Provider: {prov_data['name']})"
+                        )
+                        continue
+
+                    normalized_formats: list[str] = []
+                    seen: set[str] = set()
+                    missing_formats: list[str] = []
+                    for fmt in raw_formats:
+                        if not isinstance(fmt, str):
+                            continue
+                        fmt_upper = fmt.strip().upper()
+                        if not fmt_upper or fmt_upper in seen:
+                            continue
+                        seen.add(fmt_upper)
+                        if endpoint_formats and fmt_upper not in endpoint_formats:
+                            missing_formats.append(fmt_upper)
+                            continue
+                        normalized_formats.append(fmt_upper)
+
+                    if missing_formats:
+                        stats["errors"].append(
+                            f"Key (Provider: {prov_data['name']}) 的 api_formats 未配置对应 Endpoint，已跳过: {missing_formats}"
+                        )
+
+                    if len(normalized_formats) == 0:
+                        stats["keys"]["skipped"] += 1
+                        continue
+
+                    encrypted_key = crypto_service.encrypt(plaintext_key)
+
+                    new_key = ProviderAPIKey(
+                        id=str(uuid.uuid4()),
+                        provider_id=provider_id,
+                        api_formats=normalized_formats,
+                        api_key=encrypted_key,
+                        name=key_data.get("name") or "Imported Key",
+                        note=key_data.get("note"),
+                        rate_multiplier=key_data.get("rate_multiplier", 1.0),
+                        rate_multipliers=key_data.get("rate_multipliers"),
+                        internal_priority=key_data.get("internal_priority", 50),
+                        global_priority=key_data.get("global_priority"),
+                        rpm_limit=key_data.get("rpm_limit"),
+                        allowed_models=key_data.get("allowed_models"),
+                        capabilities=key_data.get("capabilities"),
+                        cache_ttl_minutes=key_data.get("cache_ttl_minutes", 5),
+                        max_probe_interval_minutes=key_data.get("max_probe_interval_minutes", 32),
+                        is_active=key_data.get("is_active", True),
+                        health_by_format={},
+                        circuit_breaker_by_format={},
+                    )
+                    db.add(new_key)
+                    existing_key_values.add(plaintext_key)
+                    stats["keys"]["created"] += 1

                # 导入 Models
                for model_data in prov_data.get("models", []):
--- a/src/api/admin/usage/routes.py
+++ b/src/api/admin/usage/routes.py
@@ -247,7 +247,8 @@ async def get_usage_detail(
    - `request_headers`: 请求头
    - `request_body`: 请求体
    - `provider_request_headers`: 提供商请求头
-    - `response_headers`: 响应头
+    - `response_headers`: 提供商响应头
+    - `client_response_headers`: 返回给客户端的响应头
    - `response_body`: 响应体
    - `metadata`: 提供商响应元数据
    - `tiered_pricing`: 阶梯计费信息（如适用）
@@ -916,6 +917,7 @@ class AdminUsageDetailAdapter(AdminApiAdapter):
            "request_body": usage_record.get_request_body(),
            "provider_request_headers": usage_record.provider_request_headers,
            "response_headers": usage_record.response_headers,
+            "client_response_headers": usage_record.client_response_headers,
            "response_body": usage_record.get_response_body(),
            "metadata": usage_record.request_metadata,
            "tiered_pricing": tiered_pricing_info,
--- a/src/api/base/models_service.py
+++ b/src/api/base/models_service.py
@@ -202,20 +202,59 @@ def get_available_provider_ids(db: Session, api_formats: list[str]) -> set[str]:
    条件:
    - 端点 api_format 匹配
    - 端点是活跃的
-    - 端点下有活跃的 Key
+    - Provider 下有活跃的 Key 且支持该 api_format（Key 直属 Provider，通过 api_formats 过滤）
    """
-    rows = (
-        db.query(ProviderEndpoint.provider_id)
-        .join(ProviderAPIKey, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
+    target_formats = {f.upper() for f in api_formats}
+
+    # 1) 先找出有活跃端点的 Provider（记录每个 Provider 支持的格式集合）
+    endpoint_rows = (
+        db.query(ProviderEndpoint.provider_id, ProviderEndpoint.api_format)
        .filter(
-            ProviderEndpoint.api_format.in_(api_formats),
+            ProviderEndpoint.api_format.in_(list(target_formats)),
            ProviderEndpoint.is_active.is_(True),
-            ProviderAPIKey.is_active.is_(True),
        )
-        .distinct()
        .all()
    )
-    return {row[0] for row in rows}
+
+    if not endpoint_rows:
+        return set()
+
+    provider_to_formats: dict[str, set[str]] = {}
+    for provider_id, fmt in endpoint_rows:
+        if not provider_id or not fmt:
+            continue
+        provider_to_formats.setdefault(provider_id, set()).add(str(fmt).upper())
+
+    provider_ids_with_endpoints = set(provider_to_formats.keys())
+    if not provider_ids_with_endpoints:
+        return set()
+
+    # 2) 再检查这些 Provider 是否至少有一个活跃 Key 支持对应格式
+    key_rows = (
+        db.query(ProviderAPIKey.provider_id, ProviderAPIKey.api_formats)
+        .filter(
+            ProviderAPIKey.provider_id.in_(provider_ids_with_endpoints),
+            ProviderAPIKey.is_active.is_(True),
+        )
+        .all()
+    )
+
+    available_provider_ids: set[str] = set()
+    for provider_id, key_formats in key_rows:
+        if not provider_id:
+            continue
+        endpoint_formats = provider_to_formats.get(provider_id)
+        if not endpoint_formats:
+            continue
+
+        formats_list = key_formats if isinstance(key_formats, list) else []
+        key_formats_upper = {str(f).upper() for f in formats_list}
+
+        # 只有同时满足：请求格式 ∩ Provider 端点格式 ∩ Key 支持格式 非空，才算可用
+        if key_formats_upper & endpoint_formats & target_formats:
+            available_provider_ids.add(provider_id)
+
+    return available_provider_ids


 def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) -> set[str]:
@@ -228,36 +267,64 @@ def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) ->
    3. **该端点的 Provider 关联了该模型**
    4. Key 的 allowed_models 允许该模型（null = 允许该 Provider 关联的所有模型）
    """
-    # 查询所有匹配格式的活跃端点及其活跃 Key，同时获取 endpoint_id
-    endpoint_keys = (
-        db.query(
-            ProviderEndpoint.id.label("endpoint_id"),
-            ProviderEndpoint.provider_id,
-            ProviderAPIKey.allowed_models,
-        )
-        .join(ProviderAPIKey, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
+    target_formats = {f.upper() for f in api_formats}
+
+    # 1) 找出有活跃端点的 Provider（记录每个 Provider 支持的格式集合）
+    endpoint_rows = (
+        db.query(ProviderEndpoint.provider_id, ProviderEndpoint.api_format)
        .filter(
-            ProviderEndpoint.api_format.in_(api_formats),
+            ProviderEndpoint.api_format.in_(list(target_formats)),
            ProviderEndpoint.is_active.is_(True),
+        )
+        .all()
+    )
+
+    if not endpoint_rows:
+        return set()
+
+    provider_to_formats: dict[str, set[str]] = {}
+    for provider_id, fmt in endpoint_rows:
+        if not provider_id or not fmt:
+            continue
+        provider_to_formats.setdefault(provider_id, set()).add(str(fmt).upper())
+
+    provider_ids_with_endpoints = set(provider_to_formats.keys())
+    if not provider_ids_with_endpoints:
+        return set()
+
+    # 2) 收集每个 Provider 下「支持对应格式」的活跃 Key 的 allowed_models
+    # Key 直属 Provider，通过 key.api_formats 与 Provider 端点格式交集筛选
+    key_rows = (
+        db.query(ProviderAPIKey.provider_id, ProviderAPIKey.allowed_models, ProviderAPIKey.api_formats)
+        .filter(
+            ProviderAPIKey.provider_id.in_(provider_ids_with_endpoints),
            ProviderAPIKey.is_active.is_(True),
        )
        .all()
    )

-    if not endpoint_keys:
+    # provider_id -> list[(allowed_models, usable_formats)]
+    provider_key_rules: dict[str, list[tuple[object, set[str]]]] = {}
+    for provider_id, allowed_models, key_formats in key_rows:
+        if not provider_id:
+            continue
+
+        endpoint_formats = provider_to_formats.get(provider_id)
+        if not endpoint_formats:
+            continue
+
+        formats_list = key_formats if isinstance(key_formats, list) else []
+        key_formats_upper = {str(f).upper() for f in formats_list}
+        usable_formats = key_formats_upper & endpoint_formats & target_formats
+        if not usable_formats:
+            continue
+
+        provider_key_rules.setdefault(provider_id, []).append((allowed_models, usable_formats))
+
+    provider_ids_with_format = set(provider_key_rules.keys())
+    if not provider_ids_with_format:
        return set()

-    # 收集每个 (provider_id, endpoint_id) 对应的 allowed_models
-    # 使用 provider_id 作为 key，因为模型是关联到 Provider 的
-    provider_allowed_models: dict[str, list[Optional[list[str]]]] = {}
-    provider_ids_with_format: set[str] = set()
-
-    for endpoint_id, provider_id, allowed_models in endpoint_keys:
-        provider_ids_with_format.add(provider_id)
-        if provider_id not in provider_allowed_models:
-            provider_allowed_models[provider_id] = []
-        provider_allowed_models[provider_id].append(allowed_models)
-
    # 只查询那些有匹配格式端点的 Provider 下的模型
    models = (
        db.query(Model)
@@ -285,22 +352,30 @@ def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) ->
        if model_provider_id not in provider_ids_with_format:
            continue

-        # 检查该 provider 下是否有 Key 允许这个模型
-        allowed_lists = provider_allowed_models.get(model_provider_id, [])
-        for allowed_models in allowed_lists:
+        # 检查该 provider 下是否有 Key 允许这个模型（支持 list/dict 两种 allowed_models）
+        from src.core.model_permissions import check_model_allowed
+
+        rules = provider_key_rules.get(model_provider_id, [])
+        for allowed_models, usable_formats in rules:
+            # None = 不限制
            if allowed_models is None:
-                # null = 允许该 Provider 关联的所有模型（已通过上面的查询限制）
-                available_model_ids.add(model_id)
-                break
-            elif model_id in allowed_models:
-                # 明确在允许列表中
-                available_model_ids.add(model_id)
-                break
-            elif global_model and model.provider_model_name in allowed_models:
-                # 也检查 provider_model_name
                available_model_ids.add(model_id)
                break

+            # 对于支持多个格式的 Key：任意一个可用格式允许即可
+            for fmt in usable_formats:
+                if check_model_allowed(
+                    model_name=model_id,
+                    allowed_models=allowed_models,  # type: ignore[arg-type]
+                    api_format=fmt,
+                    resolved_model_name=(model.provider_model_name if global_model else None),
+                ):
+                    available_model_ids.add(model_id)
+                    break
+            else:
+                continue
+            break
+
    return available_model_ids


--- a/src/api/dashboard/routes.py
+++ b/src/api/dashboard/routes.py
@@ -155,7 +155,7 @@ async def get_daily_stats(
 class DashboardAdapter(ApiAdapter):
    """需要登录的仪表盘适配器基类。"""

-    mode = ApiMode.ADMIN
+    mode = ApiMode.USER  # 普通用户也可访问仪表盘

    def authorize(self, context):  # type: ignore[override]
        if not context.user:
--- a/src/api/handlers/base/base_handler.py
+++ b/src/api/handlers/base/base_handler.py
@@ -98,6 +98,7 @@ class MessageTelemetry:
        request_headers: Dict[str, Any],
        response_body: Any,
        response_headers: Dict[str, Any],
+        client_response_headers: Optional[Dict[str, Any]] = None,
        cache_creation_tokens: int = 0,
        cache_read_tokens: int = 0,
        is_stream: bool = False,
@@ -143,6 +144,7 @@ class MessageTelemetry:
            request_body=request_body,
            provider_request_headers=provider_request_headers or {},
            response_headers=response_headers,
+            client_response_headers=client_response_headers,
            response_body=response_body,
            request_id=self.request_id,
            # Provider 侧追踪信息（用于记录真实成本）
@@ -192,6 +194,8 @@ class MessageTelemetry:
        cache_creation_tokens: int = 0,
        cache_read_tokens: int = 0,
        response_body: Optional[Dict[str, Any]] = None,
+        response_headers: Optional[Dict[str, Any]] = None,
+        client_response_headers: Optional[Dict[str, Any]] = None,
        # 模型映射信息
        target_model: Optional[str] = None,
    ) -> None:
@@ -207,6 +211,8 @@ class MessageTelemetry:
            cache_creation_tokens: 缓存创建 tokens
            cache_read_tokens: 缓存读取 tokens
            response_body: 响应体（如果有部分响应）
+            response_headers: 响应头（Provider 返回的原始响应头）
+            client_response_headers: 返回给客户端的响应头
            target_model: 映射后的目标模型名（如果发生了映射）
        """
        provider_name = provider or "unknown"
@@ -232,7 +238,8 @@ class MessageTelemetry:
            request_headers=request_headers,
            request_body=request_body,
            provider_request_headers=provider_request_headers or {},
-            response_headers={},
+            response_headers=response_headers or {},
+            client_response_headers=client_response_headers,
            response_body=response_body or {"error": error_message},
            request_id=self.request_id,
            # 模型映射信息
--- a/src/api/handlers/base/chat_adapter_base.py
+++ b/src/api/handlers/base/chat_adapter_base.py
@@ -351,9 +351,9 @@ class ChatAdapterBase(ApiAdapter):
        # 确定错误消息
        if isinstance(e, ProviderAuthException):
            error_message = (
-                f"提供商认证失败: {str(e)}"
+                "上游服务认证失败"
                if result.metadata.provider != "unknown"
-                else "服务端错误: 无可用提供商"
+                else "服务暂时不可用"
            )
            result.error_message = error_message

--- a/src/api/handlers/base/chat_handler_base.py
+++ b/src/api/handlers/base/chat_handler_base.py
@@ -37,7 +37,7 @@ from src.api.handlers.base.stream_processor import StreamProcessor
 from src.api.handlers.base.stream_telemetry import StreamTelemetryRecorder
 from src.api.handlers.base.utils import build_sse_headers
 from src.config.settings import config
-from src.core.error_utils import extract_error_message
+from src.core.error_utils import extract_client_error_message
 from src.core.exceptions import (
    EmbeddedErrorException,
    ProviderAuthException,
@@ -382,10 +382,17 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                http_request.is_disconnected,
            )

+            # 透传提供商的响应头给客户端
+            # 同时添加必要的 SSE 头以确保流式传输正常工作
+            client_headers = dict(ctx.response_headers) if ctx.response_headers else {}
+            # 添加/覆盖 SSE 必需的头
+            client_headers.update(build_sse_headers())
+            client_headers["content-type"] = "text/event-stream"
+
            return StreamingResponse(
                monitored_stream,
                media_type="text/event-stream",
-                headers=build_sse_headers(),
+                headers=client_headers,
                background=background_tasks,
            )

@@ -463,7 +470,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):

        # 配置 HTTP 超时
        # 注意：read timeout 用于检测连接断开，不是整体请求超时
-        # 整体请求超时由 asyncio.wait_for 控制，使用 endpoint.timeout
+        # 整体请求超时由 asyncio.wait_for 控制，使用 provider.timeout
        timeout_config = httpx.Timeout(
            connect=config.http_connect_timeout,
            read=config.http_read_timeout,  # 使用全局配置，用于检测连接断开
@@ -471,14 +478,14 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
            pool=config.http_pool_timeout,
        )

-        # endpoint.timeout 作为整体请求超时（建立连接 + 获取首字节）
-        request_timeout = float(endpoint.timeout or 300)
+        # provider.timeout 作为整体请求超时（建立连接 + 获取首字节）
+        request_timeout = float(provider.timeout or 300)

-        # 创建 HTTP 客户端（支持代理配置）
+        # 创建 HTTP 客户端（支持代理配置，从 Provider 读取）
        from src.clients.http_client import HTTPClientPool

        http_client = HTTPClientPool.create_client_with_proxy(
-            proxy_config=endpoint.proxy,
+            proxy_config=provider.proxy,
            timeout=timeout_config,
        )

@@ -514,7 +521,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):

        try:
            # 使用 asyncio.wait_for 包裹整个"建立连接 + 获取首字节"阶段
-            # endpoint.timeout 控制整体超时，避免上游长时间无响应
+            # provider.timeout 控制整体超时，避免上游长时间无响应
            await asyncio.wait_for(_connect_and_prefetch(), timeout=request_timeout)

        except asyncio.TimeoutError:
@@ -590,17 +597,22 @@ class ChatHandlerBase(BaseMessageHandler, ABC):

        actual_request_body = ctx.provider_request_body or original_request_body

+        # 失败时返回给客户端的是 JSON 错误响应
+        client_response_headers = {"content-type": "application/json"}
+
        await self.telemetry.record_failure(
            provider=ctx.provider_name or "unknown",
            model=ctx.model,
            response_time_ms=response_time_ms,
            status_code=status_code,
-            error_message=extract_error_message(error),
+            error_message=extract_client_error_message(error),
            request_headers=original_headers,
            request_body=actual_request_body,
            is_stream=True,
            api_format=ctx.api_format,
            provider_request_headers=ctx.provider_request_headers,
+            response_headers=ctx.response_headers,
+            client_response_headers=client_response_headers,
            target_model=ctx.mapped_model,
        )

@@ -691,64 +703,98 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                f"  [{self.request_id}] 请求体stream字段: {provider_payload.get('stream', 'N/A')}"
            )

-            # 创建 HTTP 客户端（支持代理配置）
-            # endpoint.timeout 作为整体请求超时
+            # 获取复用的 HTTP 客户端（支持代理配置，从 Provider 读取）
+            # 注意：使用 get_proxy_client 复用连接池，不再每次创建新客户端
            from src.clients.http_client import HTTPClientPool

-            request_timeout = float(endpoint.timeout or 300)
-            http_client = HTTPClientPool.create_client_with_proxy(
-                proxy_config=endpoint.proxy,
+            request_timeout = float(provider.timeout or 300)
+            http_client = await HTTPClientPool.get_proxy_client(
+                proxy_config=provider.proxy,
+            )
+
+            # 注意：不使用 async with，因为复用的客户端不应该被关闭
+            # 超时通过 timeout 参数控制
+            resp = await http_client.post(
+                url,
+                json=provider_payload,
+                headers=provider_hdrs,
                timeout=httpx.Timeout(request_timeout),
            )
-            async with http_client:
-                resp = await http_client.post(url, json=provider_payload, headers=provider_hdrs)

-                status_code = resp.status_code
-                response_headers = dict(resp.headers)
+            status_code = resp.status_code
+            response_headers = dict(resp.headers)

-                if resp.status_code == 401:
-                    raise ProviderAuthException(f"提供商认证失败: {provider.name}")
-                elif resp.status_code == 429:
-                    raise ProviderRateLimitException(
-                        f"提供商速率限制: {provider.name}",
-                        provider_name=str(provider.name),
-                        response_headers=response_headers,
+            if resp.status_code == 401:
+                raise ProviderAuthException(str(provider.name))
+            elif resp.status_code == 429:
+                raise ProviderRateLimitException(
+                    "请求过于频繁，请稍后重试",
+                    provider_name=str(provider.name),
+                    response_headers=response_headers,
+                )
+            elif resp.status_code >= 500:
+                # 记录响应体以便调试
+                error_body = ""
+                try:
+                    error_body = resp.text[:1000]
+                    logger.error(
+                        f"  [{self.request_id}] 上游返回5xx错误: status={resp.status_code}, body={error_body[:500]}"
                    )
-                elif resp.status_code >= 500:
-                    # 记录响应体以便调试
-                    error_body = ""
-                    try:
-                        error_body = resp.text[:1000]
-                        logger.error(
-                            f"  [{self.request_id}] 上游返回5xx错误: status={resp.status_code}, body={error_body[:500]}"
-                        )
-                    except Exception:
-                        pass
-                    raise ProviderNotAvailableException(
-                        f"提供商服务不可用: {provider.name}",
-                        provider_name=str(provider.name),
-                        upstream_status=resp.status_code,
-                        upstream_response=error_body,
-                    )
-                elif resp.status_code != 200:
-                    # 记录非200响应以便调试
-                    error_body = ""
-                    try:
-                        error_body = resp.text[:1000]
-                        logger.warning(
-                            f"  [{self.request_id}] 上游返回非200: status={resp.status_code}, body={error_body[:500]}"
-                        )
-                    except Exception:
-                        pass
-                    raise ProviderNotAvailableException(
-                        f"提供商返回错误: {provider.name}, 状态: {resp.status_code}",
-                        provider_name=str(provider.name),
-                        upstream_status=resp.status_code,
-                        upstream_response=error_body,
+                except Exception:
+                    pass
+                raise ProviderNotAvailableException(
+                    f"上游服务暂时不可用 (HTTP {resp.status_code})",
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=error_body,
+                )
+            elif resp.status_code != 200:
+                # 记录非200响应以便调试
+                error_body = ""
+                try:
+                    error_body = resp.text[:1000]
+                    logger.warning(
+                        f"  [{self.request_id}] 上游返回非200: status={resp.status_code}, body={error_body[:500]}"
                    )
+                except Exception:
+                    pass
+                raise ProviderNotAvailableException(
+                    f"上游服务返回错误 (HTTP {resp.status_code})",
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=error_body,
+                )

+            # 安全解析 JSON 响应，处理可能的编码错误
+            try:
                response_json = resp.json()
-                return response_json if isinstance(response_json, dict) else {}
+            except (UnicodeDecodeError, json.JSONDecodeError) as e:
+                # 获取原始响应内容用于调试（存入 upstream_response）
+                raw_content = ""
+                try:
+                    raw_content = resp.text[:500] if resp.text else "(empty)"
+                except Exception:
+                    try:
+                        raw_content = repr(resp.content[:500]) if resp.content else "(empty)"
+                    except Exception:
+                        raw_content = "(unable to read)"
+                logger.error(
+                    f"[{self.request_id}] 无法解析响应 JSON: {e}, 原始内容: {raw_content}"
+                )
+                # 判断错误类型，生成友好的客户端错误消息（不暴露提供商信息）
+                if raw_content == "(empty)" or not raw_content.strip():
+                    client_message = "上游服务返回了空响应"
+                elif raw_content.strip().startswith(("<", "<!doctype", "<!DOCTYPE")):
+                    client_message = "上游服务返回了非预期的响应格式"
+                else:
+                    client_message = "上游服务返回了无效的响应"
+                raise ProviderNotAvailableException(
+                    client_message,
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=raw_content,
+                )
+            return response_json if isinstance(response_json, dict) else {}

        try:
            # 解析能力需求
@@ -792,6 +838,11 @@ class ChatHandlerBase(BaseMessageHandler, ABC):

            actual_request_body = provider_request_body or original_request_body

+            # 非流式成功时，返回给客户端的是提供商响应头（透传）
+            # JSONResponse 会自动设置 content-type，但我们记录实际返回的完整头
+            client_response_headers = dict(response_headers) if response_headers else {}
+            client_response_headers["content-type"] = "application/json"
+
            total_cost = await self.telemetry.record_success(
                provider=provider_name,
                model=model,
@@ -802,6 +853,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                request_headers=original_headers,
                request_body=actual_request_body,
                response_headers=response_headers,
+                client_response_headers=client_response_headers,
                response_body=response_json,
                cache_creation_tokens=cache_creation_tokens,
                cache_read_tokens=cached_tokens,
@@ -823,7 +875,12 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                f"in:{input_tokens or 0} out:{output_tokens or 0}"
            )

-            return JSONResponse(status_code=status_code, content=response_json)
+            # 透传提供商的响应头
+            return JSONResponse(
+                status_code=status_code,
+                content=response_json,
+                headers=response_headers if response_headers else None,
+            )

        except Exception as e:
            response_time_ms = self.elapsed_ms()
@@ -838,17 +895,27 @@ class ChatHandlerBase(BaseMessageHandler, ABC):

            actual_request_body = provider_request_body or original_request_body

+            # 尝试从异常中提取响应头
+            error_response_headers: Dict[str, str] = {}
+            if isinstance(e, ProviderRateLimitException) and e.response_headers:
+                error_response_headers = e.response_headers
+            elif isinstance(e, httpx.HTTPStatusError) and hasattr(e, "response"):
+                error_response_headers = dict(e.response.headers)
+
            await self.telemetry.record_failure(
                provider=provider_name or "unknown",
                model=model,
                response_time_ms=response_time_ms,
                status_code=status_code,
-                error_message=extract_error_message(e),
+                error_message=extract_client_error_message(e),
                request_headers=original_headers,
                request_body=actual_request_body,
                is_stream=False,
                api_format=api_format,
                provider_request_headers=provider_request_headers,
+                response_headers=error_response_headers,
+                # 非流式失败返回给客户端的是 JSON 错误响应
+                client_response_headers={"content-type": "application/json"},
                # 模型映射信息
                target_model=mapped_model_result,
            )
--- a/src/api/handlers/base/cli_adapter_base.py
+++ b/src/api/handlers/base/cli_adapter_base.py
@@ -306,9 +306,9 @@ class CliAdapterBase(ApiAdapter):
        # 确定错误消息
        if isinstance(e, ProviderAuthException):
            error_message = (
-                f"提供商认证失败: {str(e)}"
+                "上游服务认证失败"
                if result.metadata.provider != "unknown"
-                else "服务端错误: 无可用提供商"
+                else "服务暂时不可用"
            )
            result.error_message = error_message

--- a/src/api/handlers/base/cli_handler_base.py
+++ b/src/api/handlers/base/cli_handler_base.py
@@ -47,7 +47,7 @@ from src.api.handlers.base.utils import (
 )
 from src.config.constants import StreamDefaults
 from src.config.settings import config
-from src.core.error_utils import extract_error_message
+from src.core.error_utils import extract_client_error_message
 from src.core.exceptions import (
    EmbeddedErrorException,
    ProviderAuthException,
@@ -376,10 +376,18 @@ class CliMessageHandlerBase(BaseMessageHandler):
            # 创建监控流
            monitored_stream = self._create_monitored_stream(ctx, stream_generator)

+            # 透传提供商的响应头给客户端
+            # 同时添加必要的 SSE 头以确保流式传输正常工作
+            client_headers = dict(ctx.response_headers) if ctx.response_headers else {}
+            # 添加/覆盖 SSE 必需的头
+            client_headers.update(build_sse_headers())
+            client_headers["content-type"] = "text/event-stream"
+            ctx.client_response_headers = client_headers
+
            return StreamingResponse(
                monitored_stream,
                media_type="text/event-stream",
-                headers=build_sse_headers(),
+                headers=client_headers,
                background=background_tasks,
            )

@@ -475,8 +483,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
            pool=config.http_pool_timeout,
        )

-        # endpoint.timeout 作为整体请求超时（建立连接 + 获取首字节）
-        request_timeout = float(endpoint.timeout or 300)
+        # provider.timeout 作为整体请求超时（建立连接 + 获取首字节）
+        request_timeout = float(provider.timeout or 300)

        logger.debug(
            f"  └─ [{self.request_id}] 发送流式请求: "
@@ -486,11 +494,11 @@ class CliMessageHandlerBase(BaseMessageHandler):
            f"timeout={request_timeout}s"
        )

-        # 创建 HTTP 客户端（支持代理配置）
+        # 创建 HTTP 客户端（支持代理配置，从 Provider 读取）
        from src.clients.http_client import HTTPClientPool

        http_client = HTTPClientPool.create_client_with_proxy(
-            proxy_config=endpoint.proxy,
+            proxy_config=provider.proxy,
            timeout=timeout_config,
        )

@@ -524,7 +532,7 @@ class CliMessageHandlerBase(BaseMessageHandler):

        try:
            # 使用 asyncio.wait_for 包裹整个"建立连接 + 获取首字节"阶段
-            # endpoint.timeout 控制整体超时，避免上游长时间无响应
+            # provider.timeout 控制整体超时，避免上游长时间无响应
            await asyncio.wait_for(_connect_and_prefetch(), timeout=request_timeout)

        except asyncio.TimeoutError:
@@ -636,12 +644,16 @@ class CliMessageHandlerBase(BaseMessageHandler):
                    if ctx.chunk_count > self.EMPTY_CHUNK_THRESHOLD and ctx.data_count == 0:
                        elapsed = time.time() - last_data_time
                        if elapsed > self.DATA_TIMEOUT:
-                            logger.warning(f"提供商 '{ctx.provider_name}' 流超时且无数据")
+                            logger.warning(f"Provider '{ctx.provider_name}' 流超时且无数据")
+                            # 设置错误状态用于后续记录
+                            ctx.status_code = 504
+                            ctx.error_message = "流式响应超时，未收到有效数据"
+                            ctx.upstream_response = f"流超时: Provider={ctx.provider_name}, elapsed={elapsed:.1f}s, chunk_count={ctx.chunk_count}, data_count=0"
                            error_event = {
                                "type": "error",
                                "error": {
                                    "type": "empty_stream_timeout",
-                                    "message": f"提供商 '{ctx.provider_name}' 流超时且未返回有效数据",
+                                    "message": ctx.error_message,
                                },
                            }
                            self._mark_first_output(ctx, output_state)
@@ -682,12 +694,16 @@ class CliMessageHandlerBase(BaseMessageHandler):
            if ctx.data_count == 0:
                # 流已开始，无法抛出异常进行故障转移
                # 发送错误事件并记录日志
-                logger.warning(f"提供商 '{ctx.provider_name}' 返回空流式响应")
+                logger.warning(f"Provider '{ctx.provider_name}' 返回空流式响应")
+                # 设置错误状态用于后续记录
+                ctx.status_code = 503
+                ctx.error_message = "上游服务返回了空的流式响应"
+                ctx.upstream_response = f"空流式响应: Provider={ctx.provider_name}, chunk_count={ctx.chunk_count}, data_count=0"
                error_event = {
                    "type": "error",
                    "error": {
                        "type": "empty_response",
-                        "message": f"提供商 '{ctx.provider_name}' 返回了空的流式响应",
+                        "message": ctx.error_message,
                    },
                }
                yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
@@ -699,12 +715,16 @@ class CliMessageHandlerBase(BaseMessageHandler):
        except httpx.StreamClosed:
            if ctx.data_count == 0:
                # 流已开始，发送错误事件而不是抛出异常
-                logger.warning(f"提供商 '{ctx.provider_name}' 流连接关闭且无数据")
+                logger.warning(f"Provider '{ctx.provider_name}' 流连接关闭且无数据")
+                # 设置错误状态用于后续记录
+                ctx.status_code = 503
+                ctx.error_message = "上游服务连接关闭且未返回数据"
+                ctx.upstream_response = f"流连接关闭: Provider={ctx.provider_name}, chunk_count={ctx.chunk_count}, data_count=0"
                error_event = {
                    "type": "error",
                    "error": {
                        "type": "stream_closed",
-                        "message": f"提供商 '{ctx.provider_name}' 连接关闭且未返回数据",
+                        "message": ctx.error_message,
                    },
                }
                yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
@@ -824,8 +844,10 @@ class CliMessageHandlerBase(BaseMessageHandler):
                            f"base_url={endpoint.base_url}"
                        )
                        raise ProviderNotAvailableException(
-                            f"提供商 '{provider.name}' 返回了 HTML 页面而非 API 响应，"
-                            f"请检查 endpoint 的 base_url 配置是否正确"
+                            "上游服务返回了非预期的响应格式",
+                            provider_name=str(provider.name),
+                            upstream_status=200,
+                            upstream_response=normalized_line[:500] if normalized_line else "(empty)",
                        )

                    if not normalized_line or normalized_line.startswith(":"):
@@ -1024,12 +1046,16 @@ class CliMessageHandlerBase(BaseMessageHandler):
                    if ctx.chunk_count > self.EMPTY_CHUNK_THRESHOLD and ctx.data_count == 0:
                        elapsed = time.time() - last_data_time
                        if elapsed > self.DATA_TIMEOUT:
-                            logger.warning(f"提供商 '{ctx.provider_name}' 流超时且无数据")
+                            logger.warning(f"Provider '{ctx.provider_name}' 流超时且无数据")
+                            # 设置错误状态用于后续记录
+                            ctx.status_code = 504
+                            ctx.error_message = "流式响应超时，未收到有效数据"
+                            ctx.upstream_response = f"流超时: Provider={ctx.provider_name}, elapsed={elapsed:.1f}s, chunk_count={ctx.chunk_count}, data_count=0"
                            error_event = {
                                "type": "error",
                                "error": {
                                    "type": "empty_stream_timeout",
-                                    "message": f"提供商 '{ctx.provider_name}' 流超时且未返回有效数据",
+                                    "message": ctx.error_message,
                                },
                            }
                            self._mark_first_output(ctx, output_state)
@@ -1071,14 +1097,18 @@ class CliMessageHandlerBase(BaseMessageHandler):
            if ctx.data_count == 0:
                # 空流通常意味着配置错误（如 base_url 指向了网页而非 API）
                logger.error(
-                    f"提供商 '{ctx.provider_name}' 返回空流式响应 (收到 {ctx.chunk_count} 个非数据行), "
+                    f"Provider '{ctx.provider_name}' 返回空流式响应 (收到 {ctx.chunk_count} 个非数据行), "
                    f"可能是 endpoint base_url 配置错误"
                )
+                # 设置错误状态用于后续记录
+                ctx.status_code = 503
+                ctx.error_message = "上游服务返回了空的流式响应"
+                ctx.upstream_response = f"空流式响应: Provider={ctx.provider_name}, chunk_count={ctx.chunk_count}, data_count=0, 可能是 base_url 配置错误"
                error_event = {
                    "type": "error",
                    "error": {
                        "type": "empty_response",
-                        "message": f"提供商 '{ctx.provider_name}' 返回了空的流式响应 (收到 {ctx.chunk_count} 行非 SSE 数据)，请检查 endpoint 的 base_url 配置是否指向了正确的 API 地址",
+                        "message": ctx.error_message,
                    },
                }
                yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
@@ -1089,12 +1119,16 @@ class CliMessageHandlerBase(BaseMessageHandler):
            raise
        except httpx.StreamClosed:
            if ctx.data_count == 0:
-                logger.warning(f"提供商 '{ctx.provider_name}' 流连接关闭且无数据")
+                logger.warning(f"Provider '{ctx.provider_name}' 流连接关闭且无数据")
+                # 设置错误状态用于后续记录
+                ctx.status_code = 503
+                ctx.error_message = "上游服务连接关闭且未返回数据"
+                ctx.upstream_response = f"流连接关闭: Provider={ctx.provider_name}, chunk_count={ctx.chunk_count}, data_count=0"
                error_event = {
                    "type": "error",
                    "error": {
                        "type": "stream_closed",
-                        "message": f"提供商 '{ctx.provider_name}' 连接关闭且未返回数据",
+                        "message": ctx.error_message,
                    },
                }
                yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
@@ -1289,6 +1323,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
                if ctx.status_code and ctx.status_code >= 400:
                    # 记录失败的 Usage，但使用已收到的预估 token 信息（来自 message_start）
                    # 这样即使请求中断，也能记录预估成本
+                    # 失败时返回给客户端的是 JSON 错误响应，如果没有设置则使用默认值
+                    client_response_headers = ctx.client_response_headers or {"content-type": "application/json"}
                    await bg_telemetry.record_failure(
                        provider=ctx.provider_name or "unknown",
                        model=ctx.model,
@@ -1306,6 +1342,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
                        cache_creation_tokens=ctx.cache_creation_tokens,
                        cache_read_tokens=ctx.cached_tokens,
                        response_body=response_body,
+                        response_headers=ctx.response_headers,
+                        client_response_headers=client_response_headers,
                        # 模型映射信息
                        target_model=ctx.mapped_model,
                    )
@@ -1319,6 +1357,14 @@ class CliMessageHandlerBase(BaseMessageHandler):
                    # 在记录统计前，允许子类从 parsed_chunks 中提取额外的元数据
                    self._finalize_stream_metadata(ctx)

+                    # 流式成功时，返回给客户端的是提供商响应头 + SSE 必需头
+                    client_response_headers = dict(ctx.response_headers) if ctx.response_headers else {}
+                    client_response_headers.update({
+                        "Cache-Control": "no-cache, no-transform",
+                        "X-Accel-Buffering": "no",
+                        "content-type": "text/event-stream",
+                    })
+
                    total_cost = await bg_telemetry.record_success(
                        provider=ctx.provider_name,
                        model=ctx.model,
@@ -1330,6 +1376,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                        request_headers=original_headers,
                        request_body=actual_request_body,
                        response_headers=ctx.response_headers,
+                        client_response_headers=client_response_headers,
                        response_body=response_body,
                        cache_creation_tokens=ctx.cache_creation_tokens,
                        cache_read_tokens=ctx.cached_tokens,
@@ -1367,13 +1414,15 @@ class CliMessageHandlerBase(BaseMessageHandler):
                    # 499 = 客户端断开连接，应标记为失败
                    # 503 = 服务不可用（如流中断），应标记为失败
                    if ctx.status_code and ctx.status_code >= 400:
+                        # 请求链路追踪使用 upstream_response（原始响应），回退到 error_message（友好消息）
+                        trace_error_message = ctx.upstream_response or ctx.error_message or f"HTTP {ctx.status_code}"
                        RequestCandidateService.mark_candidate_failed(
                            db=bg_db,
                            candidate_id=ctx.attempt_id,
                            error_type=(
                                "client_disconnected" if ctx.status_code == 499 else "stream_error"
                            ),
-                            error_message=ctx.error_message or f"HTTP {ctx.status_code}",
+                            error_message=trace_error_message,
                            status_code=ctx.status_code,
                            latency_ms=response_time_ms,
                            extra_data={
@@ -1426,17 +1475,22 @@ class CliMessageHandlerBase(BaseMessageHandler):
        # 使用实际发送给 Provider 的请求体（如果有），否则用原始请求体
        actual_request_body = ctx.provider_request_body or original_request_body

+        # 失败时返回给客户端的是 JSON 错误响应
+        client_response_headers = {"content-type": "application/json"}
+
        await self.telemetry.record_failure(
            provider=ctx.provider_name or "unknown",
            model=ctx.model,
            response_time_ms=response_time_ms,
            status_code=status_code,
-            error_message=extract_error_message(error),
+            error_message=extract_client_error_message(error),
            request_headers=original_headers,
            request_body=actual_request_body,
            is_stream=True,
            api_format=ctx.api_format,
            provider_request_headers=ctx.provider_request_headers,
+            response_headers=ctx.response_headers,
+            client_response_headers=client_response_headers,
            # 模型映射信息
            target_model=ctx.mapped_model,
        )
@@ -1534,72 +1588,99 @@ class CliMessageHandlerBase(BaseMessageHandler):
                f"原始模型={model}, 映射后={mapped_model or '无映射'}, URL模型={url_model}"
            )

-            # 创建 HTTP 客户端（支持代理配置）
-            # endpoint.timeout 作为整体请求超时
+            # 获取复用的 HTTP 客户端（支持代理配置，从 Provider 读取）
+            # 注意：使用 get_proxy_client 复用连接池，不再每次创建新客户端
            from src.clients.http_client import HTTPClientPool

-            request_timeout = float(endpoint.timeout or 300)
-            http_client = HTTPClientPool.create_client_with_proxy(
-                proxy_config=endpoint.proxy,
+            request_timeout = float(provider.timeout or 300)
+            http_client = await HTTPClientPool.get_proxy_client(
+                proxy_config=provider.proxy,
+            )
+
+            # 注意：不使用 async with，因为复用的客户端不应该被关闭
+            # 超时通过 timeout 参数控制
+            resp = await http_client.post(
+                url,
+                json=provider_payload,
+                headers=provider_headers,
                timeout=httpx.Timeout(request_timeout),
            )
-            async with http_client:
-                resp = await http_client.post(url, json=provider_payload, headers=provider_headers)

-                status_code = resp.status_code
-                response_headers = dict(resp.headers)
+            status_code = resp.status_code
+            response_headers = dict(resp.headers)

-                if resp.status_code == 401:
-                    raise ProviderAuthException(f"提供商认证失败: {provider.name}")
-                elif resp.status_code == 429:
-                    raise ProviderRateLimitException(
-                        f"提供商速率限制: {provider.name}",
-                        provider_name=str(provider.name),
-                        response_headers=response_headers,
-                        retry_after=int(resp.headers.get("retry-after", 0)) or None,
-                    )
-                elif resp.status_code >= 500:
-                    error_text = resp.text
-                    raise ProviderNotAvailableException(
-                        f"提供商服务不可用: {provider.name}, 状态: {resp.status_code}",
-                        provider_name=str(provider.name),
-                        upstream_status=resp.status_code,
-                        upstream_response=error_text,
-                    )
-                elif 300 <= resp.status_code < 400:
-                    redirect_url = resp.headers.get("location", "unknown")
-                    raise ProviderNotAvailableException(
-                        f"提供商配置错误: {provider.name}, 返回重定向 {resp.status_code} -> {redirect_url}"
-                    )
-                elif resp.status_code != 200:
-                    error_text = resp.text
-                    raise ProviderNotAvailableException(
-                        f"提供商返回错误: {provider.name}, 状态: {resp.status_code}",
-                        provider_name=str(provider.name),
-                        upstream_status=resp.status_code,
-                        upstream_response=error_text,
-                    )
+            if resp.status_code == 401:
+                raise ProviderAuthException(str(provider.name))
+            elif resp.status_code == 429:
+                raise ProviderRateLimitException(
+                    "请求过于频繁，请稍后重试",
+                    provider_name=str(provider.name),
+                    response_headers=response_headers,
+                    retry_after=int(resp.headers.get("retry-after", 0)) or None,
+                )
+            elif resp.status_code >= 500:
+                error_text = resp.text
+                raise ProviderNotAvailableException(
+                    f"上游服务暂时不可用 (HTTP {resp.status_code})",
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=error_text,
+                )
+            elif 300 <= resp.status_code < 400:
+                redirect_url = resp.headers.get("location", "unknown")
+                raise ProviderNotAvailableException(
+                    "上游服务返回重定向响应",
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=f"重定向 {resp.status_code} -> {redirect_url}",
+                )
+            elif resp.status_code != 200:
+                error_text = resp.text
+                raise ProviderNotAvailableException(
+                    f"上游服务返回错误 (HTTP {resp.status_code})",
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=error_text,
+                )

-                # 安全解析 JSON 响应，处理可能的编码错误
+            # 安全解析 JSON 响应，处理可能的编码错误
+            try:
+                response_json = resp.json()
+            except (UnicodeDecodeError, json.JSONDecodeError) as e:
+                # 获取原始响应内容用于调试（存入 upstream_response）
+                content_type = resp.headers.get("content-type", "unknown")
+                content_encoding = resp.headers.get("content-encoding", "none")
+                raw_content = ""
                try:
-                    response_json = resp.json()
-                except (UnicodeDecodeError, json.JSONDecodeError) as e:
-                    # 记录原始响应信息用于调试
-                    content_type = resp.headers.get("content-type", "unknown")
-                    content_encoding = resp.headers.get("content-encoding", "none")
-                    logger.error(
-                        f"[{self.request_id}] 无法解析响应 JSON: {e}, "
-                        f"Content-Type: {content_type}, Content-Encoding: {content_encoding}, "
-                        f"响应长度: {len(resp.content)} bytes"
-                    )
-                    raise ProviderNotAvailableException(
-                        f"提供商返回无效响应: {provider.name}, 无法解析 JSON: {str(e)[:100]}"
-                    )
+                    raw_content = resp.text[:500] if resp.text else "(empty)"
+                except Exception:
+                    try:
+                        raw_content = repr(resp.content[:500]) if resp.content else "(empty)"
+                    except Exception:
+                        raw_content = "(unable to read)"
+                logger.error(
+                    f"[{self.request_id}] 无法解析响应 JSON: {e}, "
+                    f"Content-Type: {content_type}, Content-Encoding: {content_encoding}, "
+                    f"响应长度: {len(resp.content)} bytes, 原始内容: {raw_content}"
+                )
+                # 判断错误类型，生成友好的客户端错误消息（不暴露提供商信息）
+                if raw_content == "(empty)" or not raw_content.strip():
+                    client_message = "上游服务返回了空响应"
+                elif raw_content.strip().startswith(("<", "<!doctype", "<!DOCTYPE")):
+                    client_message = "上游服务返回了非预期的响应格式"
+                else:
+                    client_message = "上游服务返回了无效的响应"
+                raise ProviderNotAvailableException(
+                    client_message,
+                    provider_name=str(provider.name),
+                    upstream_status=resp.status_code,
+                    upstream_response=raw_content,
+                )

-                # 提取 Provider 响应元数据（子类可覆盖）
-                response_metadata_result = self._extract_response_metadata(response_json)
+            # 提取 Provider 响应元数据（子类可覆盖）
+            response_metadata_result = self._extract_response_metadata(response_json)

-                return response_json if isinstance(response_json, dict) else {}
+            return response_json if isinstance(response_json, dict) else {}

        try:
            # 解析能力需求
@@ -1663,6 +1744,10 @@ class CliMessageHandlerBase(BaseMessageHandler):
            # 使用实际发送给 Provider 的请求体（如果有），否则用原始请求体
            actual_request_body = provider_request_body or original_request_body

+            # 非流式成功时，返回给客户端的是提供商响应头（透传）
+            client_response_headers = dict(response_headers) if response_headers else {}
+            client_response_headers["content-type"] = "application/json"
+
            total_cost = await self.telemetry.record_success(
                provider=provider_name,
                model=model,
@@ -1673,6 +1758,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                request_headers=original_headers,
                request_body=actual_request_body,
                response_headers=response_headers,
+                client_response_headers=client_response_headers,
                response_body=response_json,
                cache_creation_tokens=cache_creation_tokens,
                cache_read_tokens=cached_tokens,
@@ -1691,7 +1777,12 @@ class CliMessageHandlerBase(BaseMessageHandler):

            logger.info(f"{self.FORMAT_ID} 非流式响应处理完成")

-            return JSONResponse(status_code=status_code, content=response_json)
+            # 透传提供商的响应头
+            return JSONResponse(
+                status_code=status_code,
+                content=response_json,
+                headers=response_headers if response_headers else None,
+            )

        except Exception as e:
            response_time_ms = int((time.time() - sync_start_time) * 1000)
@@ -1707,17 +1798,27 @@ class CliMessageHandlerBase(BaseMessageHandler):
            # 使用实际发送给 Provider 的请求体（如果有），否则用原始请求体
            actual_request_body = provider_request_body or original_request_body

+            # 尝试从异常中提取响应头
+            error_response_headers: Dict[str, str] = {}
+            if isinstance(e, ProviderRateLimitException) and e.response_headers:
+                error_response_headers = e.response_headers
+            elif isinstance(e, httpx.HTTPStatusError) and hasattr(e, "response"):
+                error_response_headers = dict(e.response.headers)
+
            await self.telemetry.record_failure(
                provider=provider_name or "unknown",
                model=model,
                response_time_ms=response_time_ms,
                status_code=status_code,
-                error_message=extract_error_message(e),
+                error_message=extract_client_error_message(e),
                request_headers=original_headers,
                request_body=actual_request_body,
                is_stream=False,
                api_format=api_format,
                provider_request_headers=provider_request_headers,
+                response_headers=error_response_headers,
+                # 非流式失败返回给客户端的是 JSON 错误响应
+                client_response_headers={"content-type": "application/json"},
                # 模型映射信息
                target_model=mapped_model_result,
            )
--- a/src/api/handlers/base/endpoint_checker.py
+++ b/src/api/handlers/base/endpoint_checker.py
@@ -74,7 +74,6 @@ def build_safe_headers(
    return headers


-# 保持向后兼容的run_endpoint_check函数（使用新架构）
 async def run_endpoint_check(
    *,
    client: httpx.AsyncClient,  # 保持兼容性，但内部不使用
@@ -176,10 +175,16 @@ async def _calculate_and_record_usage(
        logger.warning(f"Provider API Key not found for usage calculation: {api_key_id}")
        return {"error": "Provider API Key not found"}

-    # 获取Provider Endpoint信息
+    # 获取Provider Endpoint信息（通过 api_format 查找）
    provider_endpoint = None
-    if provider_api_key.endpoint_id:
-        provider_endpoint = db.query(ProviderEndpoint).filter(ProviderEndpoint.id == provider_api_key.endpoint_id).first()
+    if api_format and provider_api_key.provider_id:
+        from src.models.database import Provider
+        provider = db.query(Provider).filter(Provider.id == provider_api_key.provider_id).first()
+        if provider:
+            for ep in provider.endpoints:
+                if ep.api_format == api_format and ep.is_active:
+                    provider_endpoint = ep
+                    break

    # 获取用户的API Key（用于记录关联，即使实际使用的是Provider API Key）
    user_api_key = None
--- a/src/api/handlers/base/stream_context.py
+++ b/src/api/handlers/base/stream_context.py
@@ -61,11 +61,13 @@ class StreamContext:

    # 响应状态
    status_code: int = 200
-    error_message: Optional[str] = None
+    error_message: Optional[str] = None  # 客户端友好的错误消息
+    upstream_response: Optional[str] = None  # 原始 Provider 响应（用于请求链路追踪）
    has_completion: bool = False

    # 请求/响应数据
-    response_headers: Dict[str, str] = field(default_factory=dict)
+    response_headers: Dict[str, str] = field(default_factory=dict)  # 提供商响应头
+    client_response_headers: Dict[str, str] = field(default_factory=dict)  # 返回给客户端的响应头
    provider_request_headers: Dict[str, str] = field(default_factory=dict)
    provider_request_body: Optional[Dict[str, Any]] = None

@@ -97,9 +99,11 @@ class StreamContext:
        self.cached_tokens = 0
        self.cache_creation_tokens = 0
        self.error_message = None
+        self.upstream_response = None
        self.status_code = 200
        self.first_byte_time_ms = None
        self.response_headers = {}
+        self.client_response_headers = {}
        self.provider_request_headers = {}
        self.provider_request_body = None
        self.response_id = None
@@ -174,10 +178,24 @@ class StreamContext:
        ):
            self.cache_creation_tokens = cache_creation_tokens

-    def mark_failed(self, status_code: int, error_message: str) -> None:
-        """标记请求失败"""
+    def mark_failed(
+        self,
+        status_code: int,
+        error_message: str,
+        upstream_response: Optional[str] = None,
+    ) -> None:
+        """
+        标记请求失败
+
+        Args:
+            status_code: HTTP 状态码
+            error_message: 客户端友好的错误消息
+            upstream_response: 原始 Provider 响应（用于请求链路追踪）
+        """
        self.status_code = status_code
        self.error_message = error_message
+        if upstream_response:
+            self.upstream_response = upstream_response

    def record_first_byte_time(self, start_time: float) -> None:
        """
--- a/src/api/handlers/base/stream_processor.py
+++ b/src/api/handlers/base/stream_processor.py
@@ -251,8 +251,10 @@ class StreamProcessor:
                            f"base_url={endpoint.base_url}"
                        )
                        raise ProviderNotAvailableException(
-                            f"提供商 '{provider.name}' 返回了 HTML 页面而非 API 响应，"
-                            f"请检查 endpoint 的 base_url 配置是否正确"
+                            "上游服务返回了非预期的响应格式",
+                            provider_name=str(provider.name),
+                            upstream_status=200,
+                            upstream_response=line[:500] if line else "(empty)",
                        )

                    # 跳过空行和注释行
--- a/src/api/handlers/base/stream_telemetry.py
+++ b/src/api/handlers/base/stream_telemetry.py
@@ -154,6 +154,14 @@ class StreamTelemetryRecorder:
        response_time_ms: int,
    ) -> None:
        """记录成功的请求"""
+        # 流式成功时，返回给客户端的是提供商响应头 + SSE 必需头
+        client_response_headers = dict(ctx.response_headers) if ctx.response_headers else {}
+        client_response_headers.update({
+            "Cache-Control": "no-cache, no-transform",
+            "X-Accel-Buffering": "no",
+            "content-type": "text/event-stream",
+        })
+
        await telemetry.record_success(
            provider=ctx.provider_name or "unknown",
            model=ctx.model,
@@ -165,6 +173,7 @@ class StreamTelemetryRecorder:
            request_headers=original_headers,
            request_body=actual_request_body,
            response_headers=ctx.response_headers,
+            client_response_headers=client_response_headers,
            response_body=response_body,
            cache_creation_tokens=ctx.cache_creation_tokens,
            cache_read_tokens=ctx.cached_tokens,
@@ -190,6 +199,9 @@ class StreamTelemetryRecorder:
        response_time_ms: int,
    ) -> None:
        """记录失败的请求"""
+        # 失败时返回给客户端的是 JSON 错误响应，如果没有设置则使用默认值
+        client_response_headers = ctx.client_response_headers or {"content-type": "application/json"}
+
        await telemetry.record_failure(
            provider=ctx.provider_name or "unknown",
            model=ctx.model,
@@ -206,6 +218,8 @@ class StreamTelemetryRecorder:
            cache_creation_tokens=ctx.cache_creation_tokens,
            cache_read_tokens=ctx.cached_tokens,
            response_body=response_body,
+            response_headers=ctx.response_headers,
+            client_response_headers=client_response_headers,
            target_model=ctx.mapped_model,
        )

@@ -239,11 +253,13 @@ class StreamTelemetryRecorder:
            )
        else:
            error_type = "client_disconnected" if ctx.status_code == 499 else "stream_error"
+            # 请求链路追踪使用 upstream_response（原始响应），回退到 error_message（友好消息）
+            trace_error_message = ctx.upstream_response or ctx.error_message or f"HTTP {ctx.status_code}"
            RequestCandidateService.mark_candidate_failed(
                db=db,
                candidate_id=ctx.attempt_id,
                error_type=error_type,
-                error_message=ctx.error_message or f"HTTP {ctx.status_code}",
+                error_message=trace_error_message,
                status_code=ctx.status_code,
                latency_ms=response_time_ms,
                extra_data={
--- a/src/api/handlers/base/utils.py
+++ b/src/api/handlers/base/utils.py
@@ -26,8 +26,10 @@ def extract_cache_creation_tokens(usage: Dict[str, Any]) -> int:
    3. **旧格式（优先级第三）**：
       usage.cache_creation_input_tokens

-    优先使用嵌套格式，如果嵌套格式字段存在但值为 0，则智能 fallback 到旧格式。
-    扁平格式和嵌套格式互斥，按顺序检查。
+    说明：
+    - 只要检测到新格式字段（嵌套/扁平），即视为权威来源：哪怕值为 0 也不回退到旧字段。
+    - 仅当新格式字段完全不存在时，才回退到旧字段。
+    - 扁平格式和嵌套格式互斥，按顺序检查。

    Args:
        usage: API 响应中的 usage 字典
@@ -37,27 +39,20 @@ def extract_cache_creation_tokens(usage: Dict[str, Any]) -> int:
    """
    # 1. 检查嵌套格式（最新格式）
    cache_creation = usage.get("cache_creation")
-    if isinstance(cache_creation, dict):
+    has_nested_format = isinstance(cache_creation, dict) and (
+        "ephemeral_5m_input_tokens" in cache_creation
+        or "ephemeral_1h_input_tokens" in cache_creation
+    )
+
+    if has_nested_format:
        cache_5m = int(cache_creation.get("ephemeral_5m_input_tokens", 0))
        cache_1h = int(cache_creation.get("ephemeral_1h_input_tokens", 0))
        total = cache_5m + cache_1h

-        if total > 0:
-            logger.debug(
-                f"Using nested cache_creation: 5m={cache_5m}, 1h={cache_1h}, total={total}"
-            )
-            return total
-
-        # 嵌套格式存在但为 0，fallback 到旧格式
-        old_format = int(usage.get("cache_creation_input_tokens", 0))
-        if old_format > 0:
-            logger.debug(
-                f"Nested cache_creation is 0, using old format: {old_format}"
-            )
-            return old_format
-
-        # 都是 0，返回 0
-        return 0
+        logger.debug(
+            f"Using nested cache_creation: 5m={cache_5m}, 1h={cache_1h}, total={total}"
+        )
+        return total

    # 2. 检查扁平新格式
    has_flat_format = (
@@ -70,22 +65,10 @@ def extract_cache_creation_tokens(usage: Dict[str, Any]) -> int:
        cache_1h = int(usage.get("claude_cache_creation_1_h_tokens", 0))
        total = cache_5m + cache_1h

-        if total > 0:
-            logger.debug(
-                f"Using flat new format: 5m={cache_5m}, 1h={cache_1h}, total={total}"
-            )
-            return total
-
-        # 扁平格式存在但为 0，fallback 到旧格式
-        old_format = int(usage.get("cache_creation_input_tokens", 0))
-        if old_format > 0:
-            logger.debug(
-                f"Flat cache_creation is 0, using old format: {old_format}"
-            )
-            return old_format
-
-        # 都是 0，返回 0
-        return 0
+        logger.debug(
+            f"Using flat new format: 5m={cache_5m}, 1h={cache_1h}, total={total}"
+        )
+        return total

    # 3. 回退到旧格式
    old_format = int(usage.get("cache_creation_input_tokens", 0))
@@ -173,8 +156,10 @@ def check_prefetched_response_error(
                f"base_url={base_url}"
            )
            raise ProviderNotAvailableException(
-                f"提供商 '{provider_name}' 返回了 HTML 页面而非 API 响应，"
-                f"请检查 endpoint 的 base_url 配置是否正确"
+                "上游服务返回了非预期的响应格式",
+                provider_name=provider_name,
+                upstream_status=200,
+                upstream_response=stripped.decode("utf-8", errors="replace")[:500],
            )

        # 纯 JSON（可能无换行/多行 JSON）
--- a/src/api/handlers/claude_cli/handler.py
+++ b/src/api/handlers/claude_cli/handler.py
@@ -2,7 +2,6 @@
 Claude CLI Message Handler - 基于通用 CLI Handler 基类的简化实现

 继承 CliMessageHandlerBase，只需覆盖格式特定的配置和事件处理逻辑。
-验证新架构的有效性：代码量从数百行减少到 ~80 行。
 """

 from typing import Any, Dict, Optional
--- a/src/api/public/catalog.py
+++ b/src/api/public/catalog.py
@@ -102,7 +102,6 @@ async def get_public_models(
    - id: 模型唯一标识符
    - provider_id: 所属提供商 ID
    - provider_name: 提供商名称
-    - provider_display_name: 提供商显示名称
    - name: 模型统一名称（优先使用 GlobalModel 名称）
    - display_name: 模型显示名称
    - description: 模型描述信息
@@ -300,10 +299,20 @@ class PublicProvidersAdapter(PublicApiAdapter):
        providers = query.offset(self.skip).limit(self.limit).all()
        result = []
        for provider in providers:
-            models_count = db.query(Model).filter(Model.provider_id == provider.id).count()
+            models_count = (
+                db.query(Model)
+                .filter(Model.provider_id == provider.id, Model.global_model_id.isnot(None))
+                .count()
+            )
            active_models_count = (
                db.query(Model)
-                .filter(and_(Model.provider_id == provider.id, Model.is_active.is_(True)))
+                .filter(
+                    and_(
+                        Model.provider_id == provider.id,
+                        Model.is_active.is_(True),
+                        Model.global_model_id.isnot(None),
+                    )
+                )
                .count()
            )
            endpoints_count = len(provider.endpoints) if provider.endpoints else 0
@@ -313,7 +322,6 @@ class PublicProvidersAdapter(PublicApiAdapter):
            provider_data = PublicProviderResponse(
                id=provider.id,
                name=provider.name,
-                display_name=provider.display_name,
                description=provider.description,
                is_active=provider.is_active,
                provider_priority=provider.provider_priority,
@@ -342,7 +350,13 @@ class PublicModelsAdapter(PublicApiAdapter):
            db.query(Model, Provider)
            .options(joinedload(Model.global_model))
            .join(Provider)
-            .filter(and_(Model.is_active.is_(True), Provider.is_active.is_(True)))
+            .filter(
+                and_(
+                    Model.is_active.is_(True),
+                    Provider.is_active.is_(True),
+                    Model.global_model_id.isnot(None),
+                )
+            )
        )
        if self.provider_id is not None:
            query = query.filter(Model.provider_id == self.provider_id)
@@ -357,7 +371,6 @@ class PublicModelsAdapter(PublicApiAdapter):
                id=model.id,
                provider_id=model.provider_id,
                provider_name=provider.name,
-                provider_display_name=provider.display_name,
                name=unified_name,
                display_name=display_name,
                description=global_model.config.get("description") if global_model and global_model.config else None,
@@ -386,7 +399,13 @@ class PublicStatsAdapter(PublicApiAdapter):
        active_models = (
            db.query(Model)
            .join(Provider)
-            .filter(and_(Model.is_active.is_(True), Provider.is_active.is_(True)))
+            .filter(
+                and_(
+                    Model.is_active.is_(True),
+                    Provider.is_active.is_(True),
+                    Model.global_model_id.isnot(None),
+                )
+            )
            .count()
        )
        formats = (
@@ -418,7 +437,13 @@ class PublicSearchModelsAdapter(PublicApiAdapter):
            .options(joinedload(Model.global_model))
            .join(Provider)
            .outerjoin(GlobalModel, Model.global_model_id == GlobalModel.id)
-            .filter(and_(Model.is_active.is_(True), Provider.is_active.is_(True)))
+            .filter(
+                and_(
+                    Model.is_active.is_(True),
+                    Provider.is_active.is_(True),
+                    Model.global_model_id.isnot(None),
+                )
+            )
        )
        search_filter = (
            Model.provider_model_name.ilike(f"%{self.query}%")
@@ -439,7 +464,6 @@ class PublicSearchModelsAdapter(PublicApiAdapter):
                id=model.id,
                provider_id=model.provider_id,
                provider_name=provider.name,
-                provider_display_name=provider.display_name,
                name=unified_name,
                display_name=display_name,
                description=global_model.config.get("description") if global_model and global_model.config else None,
--- a/src/api/public/system_catalog.py
+++ b/src/api/public/system_catalog.py
@@ -43,7 +43,6 @@ def _serialize_provider(
    provider_data: Dict[str, Any] = {
        "id": provider.id,
        "name": provider.name,
-        "display_name": provider.display_name,
        "is_active": provider.is_active,
        "provider_priority": provider.provider_priority,
    }
--- a/src/api/user_me/routes.py
+++ b/src/api/user_me/routes.py
@@ -1023,7 +1023,6 @@ class ListAvailableProvidersAdapter(AuthenticatedApiAdapter):
                {
                    "id": provider.id,
                    "name": provider.name,
-                    "display_name": provider.display_name,
                    "description": provider.description,
                    "provider_priority": provider.provider_priority,
                    "endpoints": endpoints_data,
--- a/src/clients/http_client.py
+++ b/src/clients/http_client.py
@@ -1,10 +1,18 @@
 """
 全局HTTP客户端池管理
 避免每次请求都创建新的AsyncClient,提高性能
+
+性能优化说明：
+1. 默认客户端：无代理场景，全局复用单一客户端
+2. 代理客户端缓存：相同代理配置复用同一客户端，避免重复创建
+3. 连接池复用：Keep-alive 连接减少 TCP 握手开销
 """

+import asyncio
+import hashlib
+import time
 from contextlib import asynccontextmanager
-from typing import Any, Dict, Optional
+from typing import Any, Dict, Optional, Tuple
 from urllib.parse import quote, urlparse

 import httpx
@@ -12,6 +20,32 @@ import httpx
 from src.config import config
 from src.core.logger import logger

+# 模块级锁，避免类属性延迟初始化的竞态条件
+_proxy_clients_lock = asyncio.Lock()
+_default_client_lock = asyncio.Lock()
+
+
+def _compute_proxy_cache_key(proxy_config: Optional[Dict[str, Any]]) -> str:
+    """
+    计算代理配置的缓存键
+
+    Args:
+        proxy_config: 代理配置字典
+
+    Returns:
+        缓存键字符串，无代理时返回 "__no_proxy__"
+    """
+    if not proxy_config:
+        return "__no_proxy__"
+
+    # 构建代理 URL 作为缓存键的基础
+    proxy_url = build_proxy_url(proxy_config)
+    if not proxy_url:
+        return "__no_proxy__"
+
+    # 使用 MD5 哈希来避免过长的键名
+    return f"proxy:{hashlib.md5(proxy_url.encode()).hexdigest()[:16]}"
+

 def build_proxy_url(proxy_config: Dict[str, Any]) -> Optional[str]:
    """
@@ -61,11 +95,20 @@ class HTTPClientPool:
    全局HTTP客户端池单例

    管理可重用的httpx.AsyncClient实例,避免频繁创建/销毁连接
+
+    性能优化：
+    1. 默认客户端：无代理场景复用
+    2. 代理客户端缓存：相同代理配置复用同一客户端
+    3. LRU 淘汰：代理客户端超过上限时淘汰最久未使用的
    """

    _instance: Optional["HTTPClientPool"] = None
    _default_client: Optional[httpx.AsyncClient] = None
    _clients: Dict[str, httpx.AsyncClient] = {}
+    # 代理客户端缓存：{cache_key: (client, last_used_time)}
+    _proxy_clients: Dict[str, Tuple[httpx.AsyncClient, float]] = {}
+    # 代理客户端缓存上限（避免内存泄漏）
+    _max_proxy_clients: int = 50

    def __new__(cls):
        if cls._instance is None:
@@ -73,12 +116,50 @@ class HTTPClientPool:
        return cls._instance

    @classmethod
-    def get_default_client(cls) -> httpx.AsyncClient:
+    async def get_default_client_async(cls) -> httpx.AsyncClient:
        """
-        获取默认的HTTP客户端
+        获取默认的HTTP客户端（异步线程安全版本）

        用于大多数HTTP请求,具有合理的默认配置
        """
+        if cls._default_client is not None:
+            return cls._default_client
+
+        async with _default_client_lock:
+            # 双重检查，避免重复创建
+            if cls._default_client is None:
+                cls._default_client = httpx.AsyncClient(
+                    http2=False,  # 暂时禁用HTTP/2以提高兼容性
+                    verify=True,  # 启用SSL验证
+                    timeout=httpx.Timeout(
+                        connect=config.http_connect_timeout,
+                        read=config.http_read_timeout,
+                        write=config.http_write_timeout,
+                        pool=config.http_pool_timeout,
+                    ),
+                    limits=httpx.Limits(
+                        max_connections=config.http_max_connections,
+                        max_keepalive_connections=config.http_keepalive_connections,
+                        keepalive_expiry=config.http_keepalive_expiry,
+                    ),
+                    follow_redirects=True,  # 跟随重定向
+                )
+                logger.info(
+                    f"全局HTTP客户端池已初始化: "
+                    f"max_connections={config.http_max_connections}, "
+                    f"keepalive={config.http_keepalive_connections}, "
+                    f"keepalive_expiry={config.http_keepalive_expiry}s"
+                )
+        return cls._default_client
+
+    @classmethod
+    def get_default_client(cls) -> httpx.AsyncClient:
+        """
+        获取默认的HTTP客户端（同步版本，向后兼容）
+
+        ⚠️ 注意：此方法在高并发首次调用时可能存在竞态条件，
+        推荐使用 get_default_client_async() 异步版本。
+        """
        if cls._default_client is None:
            cls._default_client = httpx.AsyncClient(
                http2=False,  # 暂时禁用HTTP/2以提高兼容性
@@ -135,6 +216,101 @@ class HTTPClientPool:

        return cls._clients[name]

+    @classmethod
+    def _get_proxy_clients_lock(cls) -> asyncio.Lock:
+        """获取代理客户端缓存锁（模块级单例，避免竞态条件）"""
+        return _proxy_clients_lock
+
+    @classmethod
+    async def _evict_lru_proxy_client(cls) -> None:
+        """淘汰最久未使用的代理客户端"""
+        if len(cls._proxy_clients) < cls._max_proxy_clients:
+            return
+
+        # 找到最久未使用的客户端
+        oldest_key = min(cls._proxy_clients.keys(), key=lambda k: cls._proxy_clients[k][1])
+        old_client, _ = cls._proxy_clients.pop(oldest_key)
+
+        # 异步关闭旧客户端
+        try:
+            await old_client.aclose()
+            logger.debug(f"淘汰代理客户端: {oldest_key}")
+        except Exception as e:
+            logger.warning(f"关闭代理客户端失败: {e}")
+
+    @classmethod
+    async def get_proxy_client(
+        cls,
+        proxy_config: Optional[Dict[str, Any]] = None,
+    ) -> httpx.AsyncClient:
+        """
+        获取代理客户端（带缓存复用）
+
+        相同代理配置会复用同一个客户端，大幅减少连接建立开销。
+        注意：返回的客户端使用默认超时配置，如需自定义超时请在请求时传递 timeout 参数。
+
+        Args:
+            proxy_config: 代理配置字典，包含 url, username, password
+
+        Returns:
+            可复用的 httpx.AsyncClient 实例
+        """
+        cache_key = _compute_proxy_cache_key(proxy_config)
+
+        # 无代理时返回默认客户端
+        if cache_key == "__no_proxy__":
+            return await cls.get_default_client_async()
+
+        lock = cls._get_proxy_clients_lock()
+        async with lock:
+            # 检查缓存
+            if cache_key in cls._proxy_clients:
+                client, _ = cls._proxy_clients[cache_key]
+                # 健康检查：如果客户端已关闭，移除并重新创建
+                if client.is_closed:
+                    del cls._proxy_clients[cache_key]
+                    logger.debug(f"代理客户端已关闭，将重新创建: {cache_key}")
+                else:
+                    # 更新最后使用时间
+                    cls._proxy_clients[cache_key] = (client, time.time())
+                    return client
+
+            # 淘汰旧客户端（如果超过上限）
+            await cls._evict_lru_proxy_client()
+
+            # 创建新客户端（使用默认超时，请求时可覆盖）
+            client_config: Dict[str, Any] = {
+                "http2": False,
+                "verify": True,
+                "follow_redirects": True,
+                "limits": httpx.Limits(
+                    max_connections=config.http_max_connections,
+                    max_keepalive_connections=config.http_keepalive_connections,
+                    keepalive_expiry=config.http_keepalive_expiry,
+                ),
+                "timeout": httpx.Timeout(
+                    connect=config.http_connect_timeout,
+                    read=config.http_read_timeout,
+                    write=config.http_write_timeout,
+                    pool=config.http_pool_timeout,
+                ),
+            }
+
+            # 添加代理配置
+            proxy_url = build_proxy_url(proxy_config) if proxy_config else None
+            if proxy_url:
+                client_config["proxy"] = proxy_url
+
+            client = httpx.AsyncClient(**client_config)
+            cls._proxy_clients[cache_key] = (client, time.time())
+
+            logger.debug(
+                f"创建代理客户端(缓存): {proxy_config.get('url', 'unknown') if proxy_config else 'none'}, "
+                f"缓存数量: {len(cls._proxy_clients)}"
+            )
+
+            return client
+
    @classmethod
    async def close_all(cls):
        """关闭所有HTTP客户端"""
@@ -148,6 +324,16 @@ class HTTPClientPool:
            logger.debug(f"命名HTTP客户端已关闭: {name}")

        cls._clients.clear()
+
+        # 关闭代理客户端缓存
+        for cache_key, (client, _) in cls._proxy_clients.items():
+            try:
+                await client.aclose()
+                logger.debug(f"代理客户端已关闭: {cache_key}")
+            except Exception as e:
+                logger.warning(f"关闭代理客户端失败: {e}")
+
+        cls._proxy_clients.clear()
        logger.info("所有HTTP客户端已关闭")

    @classmethod
@@ -190,13 +376,15 @@ class HTTPClientPool:
        """
        创建带代理配置的HTTP客户端

+        ⚠️ 性能警告：此方法每次都创建新客户端，推荐使用 get_proxy_client() 复用连接。
+
        Args:
            proxy_config: 代理配置字典，包含 url, username, password
            timeout: 超时配置
            **kwargs: 其他 httpx.AsyncClient 配置参数

        Returns:
-            配置好的 httpx.AsyncClient 实例
+            配置好的 httpx.AsyncClient 实例（调用者需要负责关闭）
        """
        client_config: Dict[str, Any] = {
            "http2": False,
@@ -218,11 +406,21 @@ class HTTPClientPool:
        proxy_url = build_proxy_url(proxy_config) if proxy_config else None
        if proxy_url:
            client_config["proxy"] = proxy_url
-            logger.debug(f"创建带代理的HTTP客户端: {proxy_config.get('url', 'unknown')}")
+            logger.debug(f"创建带代理的HTTP客户端(一次性): {proxy_config.get('url', 'unknown')}")

        client_config.update(kwargs)
        return httpx.AsyncClient(**client_config)

+    @classmethod
+    def get_pool_stats(cls) -> Dict[str, Any]:
+        """获取连接池统计信息"""
+        return {
+            "default_client_active": cls._default_client is not None,
+            "named_clients_count": len(cls._clients),
+            "proxy_clients_count": len(cls._proxy_clients),
+            "max_proxy_clients": cls._max_proxy_clients,
+        }
+

 # 便捷访问函数
 def get_http_client() -> httpx.AsyncClient:
--- a/src/config/constants.py
+++ b/src/config/constants.py
@@ -52,19 +52,33 @@ class StreamDefaults:
    MAX_PREFETCH_BYTES = 64 * 1024  # 64KB


-class ConcurrencyDefaults:
-    """并发控制默认值
+class RPMDefaults:
+    """RPM（每分钟请求数）限制默认值

    算法说明：边界记忆 + 渐进探测
-    - 触发 429 时记录边界（last_concurrent_peak），新限制 = 边界 - 1
+    - 触发 429 时记录边界（last_rpm_peak），新限制 = 边界 - 1
    - 扩容时不超过边界，除非是探测性扩容（长时间无 429）
    - 这样可以快速收敛到真实限制附近，避免过度保守
+
+    初始值 50 RPM：
+    - 系统会根据实际使用自动调整
    """

-    # 自适应并发初始限制（宽松起步，遇到 429 再降低）
-    INITIAL_LIMIT = 50
+    # 自适应 RPM 初始限制
+    INITIAL_LIMIT = 50  # 每分钟 50 次请求

-    # 429错误后的冷却时间（分钟）- 在此期间不会增加并发限制
+    # === 内存模式 RPM 计数器配置 ===
+    # 内存模式下的最大条目限制（防止内存泄漏）
+    # 每个条目约占 100 字节，10000 条目 = ~1MB
+    # 计算依据：1000 Key × 5 API 格式 × 2 (buffer) = 10000
+    # 可通过环境变量 RPM_MAX_MEMORY_ENTRIES 覆盖
+    MAX_MEMORY_RPM_ENTRIES = 10000
+
+    # 内存使用告警阈值（达到此比例时记录警告日志）
+    # 可通过环境变量 RPM_MEMORY_WARNING_THRESHOLD 覆盖
+    MEMORY_WARNING_THRESHOLD = 0.6  # 60%
+
+    # 429错误后的冷却时间（分钟）- 在此期间不会增加 RPM 限制
    COOLDOWN_AFTER_429_MINUTES = 5

    # 探测间隔上限（分钟）- 用于长期探测策略
@@ -86,30 +100,30 @@ class ConcurrencyDefaults:
    # 最小采样数 - 窗口内至少需要这么多采样才能做出扩容决策
    MIN_SAMPLES_FOR_DECISION = 5

-    # 扩容步长 - 每次扩容增加的并发数
-    INCREASE_STEP = 2
+    # 扩容步长 - 每次扩容增加的 RPM
+    INCREASE_STEP = 5  # 每次增加 5 RPM

-    # 最大并发限制上限
-    MAX_CONCURRENT_LIMIT = 200
+    # 最大 RPM 限制上限（不设上限，让系统自适应学习）
+    MAX_RPM_LIMIT = 10000

-    # 最小并发限制下限
-    # 设置为 3 而不是 1，因为预留机制（10%预留给缓存用户）会导致
-    # 当 learned_max_concurrent=1 时新用户实际可用槽位为 0，永远无法命中
-    # 注意：当 limit < 10 时，预留机制实际不生效（预留槽位 = 0），这是可接受的
-    MIN_CONCURRENT_LIMIT = 3
+    # 最小 RPM 限制下限
+    MIN_RPM_LIMIT = 5
+
+    # 缓存用户预留比例（默认 10%，新用户可用 90%）
+    # 已被动态预留机制 (AdaptiveReservationDefaults) 替代，保留用于向后兼容
+    CACHE_RESERVATION_RATIO = 0.1

    # === 探测性扩容参数 ===
    # 探测性扩容间隔（分钟）- 长时间无 429 且有流量时尝试扩容
-    # 探测性扩容可以突破已知边界，尝试更高的并发
+    # 探测性扩容可以突破已知边界，尝试更高的 RPM
    PROBE_INCREASE_INTERVAL_MINUTES = 30

    # 探测性扩容最小请求数 - 在探测间隔内至少需要这么多请求
    PROBE_INCREASE_MIN_REQUESTS = 10

-    # === 缓存用户预留比例 ===
-    # 缓存用户槽位预留比例（新用户可用 1 - 此值）
-    # 0.1 表示缓存用户预留 10%，新用户可用 90%
-    CACHE_RESERVATION_RATIO = 0.1
+
+# 向后兼容别名
+ConcurrencyDefaults = RPMDefaults


 class CircuitBreakerDefaults:
@@ -193,10 +207,19 @@ class AdaptiveReservationDefaults:


 class TimeoutDefaults:
-    """超时配置默认值（秒）"""
+    """超时配置默认值（秒）

-    # HTTP 请求默认超时
-    HTTP_REQUEST = 300  # 5分钟
+    超时配置说明：
+    - 全局默认值和 Provider 默认值统一为 120 秒
+    - 120 秒是 LLM API 的合理默认值：
+      * 大多数请求在 30 秒内完成
+      * 复杂推理（如 Claude extended thinking）可能需要 60-90 秒
+      * 120 秒足够覆盖大部分场景，同时避免线程池被长时间占用
+    - 如需更长超时，可在 Provider 级别单独配置
+    """
+
+    # HTTP 请求默认超时（与 Provider 默认值保持一致）
+    HTTP_REQUEST = 120  # 2分钟

    # 数据库连接池获取超时
    DB_POOL = 30
--- a/src/core/cache_service.py
+++ b/src/core/cache_service.py
@@ -36,6 +36,12 @@ class CacheService:
                try:
                    return json.loads(value)
                except (json.JSONDecodeError, TypeError):
+                    # 非 JSON 值：统一返回字符串，避免上层出现 bytes/str 混用
+                    if isinstance(value, (bytes, bytearray)):
+                        try:
+                            return value.decode("utf-8")
+                        except Exception:
+                            return value
                    return value

            return None
@@ -98,6 +104,48 @@ class CacheService:
            logger.warning(f"缓存删除失败: {key} - {e}")
            return False

+    @staticmethod
+    async def delete_pattern(pattern: str, batch_size: int = 100) -> int:
+        """
+        删除匹配模式的所有缓存
+
+        使用 SCAN 遍历并分批删除，避免阻塞 Redis
+
+        Args:
+            pattern: 缓存键模式（支持 * 通配符）
+            batch_size: 每批删除的最大键数量
+
+        Returns:
+            删除的键数量
+        """
+        try:
+            redis = await get_redis_client(require_redis=False)
+            if not redis:
+                return 0
+
+            # 使用 SCAN 遍历匹配的键
+            deleted_count = 0
+            cursor = 0
+            while True:
+                cursor, keys = await redis.scan(cursor, match=pattern, count=batch_size)
+                if keys:
+                    # 分批删除，避免单次删除过多键导致 Redis 阻塞
+                    for i in range(0, len(keys), batch_size):
+                        batch = keys[i : i + batch_size]
+                        await redis.delete(*batch)
+                        deleted_count += len(batch)
+                if cursor == 0:
+                    break
+
+            if deleted_count > 0:
+                logger.debug(f"缓存模式删除成功: {pattern}, 删除 {deleted_count} 个键")
+
+            return deleted_count
+
+        except Exception as e:
+            logger.warning(f"缓存模式删除失败: {pattern} - {e}")
+            return 0
+
    @staticmethod
    async def exists(key: str) -> bool:
        """
--- a/src/core/error_utils.py
+++ b/src/core/error_utils.py
@@ -7,16 +7,19 @@ from typing import Optional

 def extract_error_message(error: Exception, status_code: Optional[int] = None) -> str:
    """
-    从异常中提取错误消息，优先使用上游响应内容
+    从异常中提取错误消息，优先使用上游原始响应（用于链路追踪/调试）
+
+    此函数用于 RequestCandidate 表的 error_message 字段，
+    用于请求链路追踪中显示原始 Provider 响应。

    Args:
        error: 异常对象
        status_code: 可选的 HTTP 状态码，用于构建更详细的错误消息

    Returns:
-        错误消息字符串
+        错误消息字符串（原始 Provider 响应）
    """
-    # 优先使用 upstream_response 属性（包含上游 Provider 的原始错误）
+    # 优先使用 upstream_response 属性（包含上游 Provider 的原始错误，用于调试）
    upstream_response = getattr(error, "upstream_response", None)
    if upstream_response and isinstance(upstream_response, str) and upstream_response.strip():
        return str(upstream_response)
@@ -26,3 +29,25 @@ def extract_error_message(error: Exception, status_code: Optional[int] = None) -
    if status_code is not None:
        return f"HTTP {status_code}: {error_str}"
    return error_str
+
+
+def extract_client_error_message(error: Exception) -> str:
+    """
+    从异常中提取客户端友好的错误消息（用于返回给客户端/Usage 记录）
+
+    此函数用于 Usage 表的 error_message 字段，
+    用于显示给最终用户的友好错误消息。
+
+    Args:
+        error: 异常对象
+
+    Returns:
+        友好的错误消息字符串
+    """
+    # 优先使用 message 属性（已经是友好处理过的消息）
+    message = getattr(error, "message", None)
+    if message and isinstance(message, str) and message.strip():
+        return message
+
+    # 回退到异常的字符串表示
+    return str(error) or repr(error)
--- a/src/core/exceptions.py
+++ b/src/core/exceptions.py
@@ -205,7 +205,7 @@ class ProviderTimeoutException(ProviderException):

    def __init__(self, provider_name: str, timeout: int, request_metadata: Optional[Any] = None):
        super().__init__(
-            message=f"提供商 '{provider_name}' 请求超时（{timeout}秒）",
+            message=f"请求超时（{timeout}秒）",
            provider_name=provider_name,
            request_metadata=request_metadata,
            timeout=timeout,
@@ -217,7 +217,7 @@ class ProviderAuthException(ProviderException):

    def __init__(self, provider_name: str, request_metadata: Optional[Any] = None):
        super().__init__(
-            message=f"提供商 '{provider_name}' 认证失败，请检查API密钥",
+            message="上游服务认证失败",
            provider_name=provider_name,
            request_metadata=request_metadata,
        )
@@ -292,9 +292,8 @@ class ModelNotSupportedException(ProxyException):
    """模型不支持"""

    def __init__(self, model: str, provider_name: Optional[str] = None):
+        # 客户端消息不暴露提供商信息
        message = f"模型 '{model}' 不受支持"
-        if provider_name:
-            message = f"提供商 '{provider_name}' 不支持模型 '{model}'"
        super().__init__(
            status_code=status.HTTP_400_BAD_REQUEST,
            error_type="model_not_supported",
@@ -307,10 +306,8 @@ class StreamingNotSupportedException(ProxyException):
    """流式请求不支持"""

    def __init__(self, model: str, provider_name: Optional[str] = None):
-        if provider_name:
-            message = f"模型 '{model}' 在提供商 '{provider_name}' 上不支持流式请求"
-        else:
-            message = f"模型 '{model}' 不支持流式请求"
+        # 客户端消息不暴露提供商信息
+        message = f"模型 '{model}' 不支持流式请求"
        super().__init__(
            status_code=status.HTTP_400_BAD_REQUEST,
            error_type="streaming_not_supported",
@@ -389,7 +386,7 @@ class JSONParseException(ProviderException):
            details["response_content"] = response_content

        super().__init__(
-            message=f"提供商 '{provider_name}' 返回了无效的JSON响应",
+            message="上游服务返回了无效的响应",
            provider_name=provider_name,
            request_metadata=request_metadata,
            **details,
@@ -406,7 +403,7 @@ class EmptyStreamException(ProviderException):
        request_metadata: Optional[Any] = None,
    ):
        super().__init__(
-            message=f"提供商 '{provider_name}' 返回了空的流式响应（status=200 但无数据）",
+            message="上游服务返回了空的流式响应",
            provider_name=provider_name,
            request_metadata=request_metadata,
            chunk_count=chunk_count,
@@ -428,11 +425,10 @@ class EmbeddedErrorException(ProviderException):
        error_status: Optional[str] = None,
        request_metadata: Optional[Any] = None,
    ):
-        message = f"提供商 '{provider_name}' 返回了嵌套错误"
+        # 客户端消息不暴露提供商信息
+        message = "上游服务返回了错误"
        if error_code:
            message += f" (code={error_code})"
-        if error_message:
-            message += f": {error_message}"

        super().__init__(
            message=message,
@@ -549,12 +545,14 @@ class ErrorResponse:
        if isinstance(e, ProxyException):
            details = e.details.copy() if e.details else {}
            status_code = e.status_code
-            message = e.message
-            # 如果是 ProviderNotAvailableException 且有上游错误，直接透传上游信息
-            if isinstance(e, ProviderNotAvailableException) and e.upstream_response:
+            message = e.message  # 使用友好的错误消息
+            # 如果是 ProviderNotAvailableException 且有上游错误信息
+            if isinstance(e, ProviderNotAvailableException):
                if e.upstream_status:
                    status_code = e.upstream_status
-                message = e.upstream_response
+                # upstream_response 存入 details 供请求链路追踪使用，不作为客户端消息
+                if e.upstream_response:
+                    details["upstream_response"] = e.upstream_response
            return ErrorResponse.create(
                error_type=e.error_type,
                message=message,
--- a/src/core/model_permissions.py
+++ b/src/core/model_permissions.py
@@ -0,0 +1,286 @@
+"""
+模型权限工具
+
+支持两种 allowed_models 格式：
+1. 简单模式（列表）: ["claude-sonnet-4", "gpt-4o"]
+2. 按格式模式（字典）: {"OPENAI": ["gpt-4o"], "CLAUDE": ["claude-sonnet-4"]}
+
+使用 None/null 表示不限制（允许所有模型）
+"""
+
+from typing import Any, Dict, List, Optional, Set, Union
+
+# 类型别名
+AllowedModels = Optional[Union[List[str], Dict[str, List[str]]]]
+
+
+def normalize_allowed_models(
+    allowed_models: AllowedModels,
+    api_format: Optional[str] = None,
+) -> Optional[Set[str]]:
+    """
+    将 allowed_models 规范化为模型名称集合
+
+    Args:
+        allowed_models: 允许的模型配置（列表或字典）
+        api_format: 当前请求的 API 格式（用于字典模式）
+
+    Returns:
+        - None: 不限制（允许所有模型）
+        - Set[str]: 允许的模型名称集合（可能为空集，表示拒绝所有）
+    """
+    if allowed_models is None:
+        return None
+
+    # 简单模式：直接是列表
+    if isinstance(allowed_models, list):
+        return set(allowed_models)
+
+    # 按格式模式：字典
+    if isinstance(allowed_models, dict):
+        if api_format is None:
+            # 没有指定格式，合并所有格式的模型
+            all_models: Set[str] = set()
+            for models in allowed_models.values():
+                if isinstance(models, list):
+                    all_models.update(models)
+            return all_models if all_models else None
+
+        # 查找指定格式的模型列表
+        api_format_upper = api_format.upper()
+        models = allowed_models.get(api_format_upper)
+        if models is None:
+            # 该格式未配置，检查是否有通配符 "*"
+            models = allowed_models.get("*")
+
+        if models is None:
+            # 字典模式下未配置的格式 = 不限制该格式
+            return None
+
+        return set(models) if isinstance(models, list) else None
+
+    # 未知类型，视为不限制
+    return None
+
+
+def check_model_allowed(
+    model_name: str,
+    allowed_models: AllowedModels,
+    api_format: Optional[str] = None,
+    resolved_model_name: Optional[str] = None,
+) -> bool:
+    """
+    检查模型是否被允许
+
+    Args:
+        model_name: 请求的模型名称
+        allowed_models: 允许的模型配置
+        api_format: 当前请求的 API 格式
+        resolved_model_name: 解析后的 GlobalModel.name（可选）
+
+    Returns:
+        True: 允许使用该模型
+        False: 不允许使用该模型
+    """
+    allowed_set = normalize_allowed_models(allowed_models, api_format)
+
+    if allowed_set is None:
+        # 不限制
+        return True
+
+    if len(allowed_set) == 0:
+        # 空集合 = 拒绝所有
+        return False
+
+    # 检查请求的模型名或解析后的名称是否在白名单中
+    if model_name in allowed_set:
+        return True
+
+    if resolved_model_name and resolved_model_name in allowed_set:
+        return True
+
+    return False
+
+
+def merge_allowed_models(
+    allowed_models_1: AllowedModels,
+    allowed_models_2: AllowedModels,
+) -> AllowedModels:
+    """
+    合并两个 allowed_models 配置，取交集
+
+    规则：
+    - 如果任一为 None，返回另一个
+    - 如果都有值，取交集
+    - 如果都是列表，取列表交集
+    - 如果有字典，按 API 格式分别取交集（保持字典语义，不丢失格式区分信息）
+
+    Args:
+        allowed_models_1: 第一个配置
+        allowed_models_2: 第二个配置
+
+    Returns:
+        合并后的配置
+    """
+    if allowed_models_1 is None:
+        return allowed_models_2
+    if allowed_models_2 is None:
+        return allowed_models_1
+
+    # 两个都是简单列表：直接取交集（返回确定性顺序）
+    if isinstance(allowed_models_1, list) and isinstance(allowed_models_2, list):
+        intersection = set(allowed_models_1) & set(allowed_models_2)
+        return sorted(intersection) if intersection else []
+
+    # 任一为字典模式：按 API 格式分别取交集，避免把 dict 合并成 list 导致权限过宽
+    from src.core.enums import APIFormat
+
+    def merge_sets(a: Optional[Set[str]], b: Optional[Set[str]]) -> Optional[Set[str]]:
+        # None 表示不限制：交集规则下等价于“只受另一方限制”
+        if a is None:
+            return b
+        if b is None:
+            return a
+        return a & b
+
+    known_formats = [fmt.value for fmt in APIFormat]
+
+    per_format: Dict[str, Optional[Set[str]]] = {}
+    for fmt in known_formats:
+        s1 = normalize_allowed_models(allowed_models_1, api_format=fmt)
+        s2 = normalize_allowed_models(allowed_models_2, api_format=fmt)
+        per_format[fmt] = merge_sets(s1, s2)
+
+    # 计算默认（未知格式）的交集，用 "*" 作为默认值以覆盖未枚举的格式
+    default_s1 = normalize_allowed_models(allowed_models_1, api_format="__DEFAULT__")
+    default_s2 = normalize_allowed_models(allowed_models_2, api_format="__DEFAULT__")
+    default_set = merge_sets(default_s1, default_s2)
+
+    # 如果 default_set 非 None 且不存在“某些格式不限制”的情况，可用 "*" 作为默认规则并按需覆盖
+    can_use_wildcard = default_set is not None and all(v is not None for v in per_format.values())
+
+    merged_dict: Dict[str, List[str]] = {}
+
+    if can_use_wildcard and default_set is not None:
+        merged_dict["*"] = sorted(default_set)
+        for fmt, s in per_format.items():
+            # can_use_wildcard 保证 s 非 None
+            if s is not None and s != default_set:
+                merged_dict[fmt] = sorted(s)
+    else:
+        for fmt, s in per_format.items():
+            if s is None:
+                continue
+            merged_dict[fmt] = sorted(s)
+
+    if not merged_dict:
+        # 全部不限制
+        return None
+
+    return merged_dict
+
+
+def get_allowed_models_preview(
+    allowed_models: AllowedModels,
+    max_items: int = 3,
+) -> str:
+    """
+    获取 allowed_models 的预览字符串（用于日志和错误消息）
+
+    Args:
+        allowed_models: 允许的模型配置
+        max_items: 最多显示的模型数
+
+    Returns:
+        预览字符串，如 "gpt-4o, claude-sonnet-4, ..."
+    """
+    if allowed_models is None:
+        return "(不限制)"
+
+    all_models: Set[str] = set()
+
+    if isinstance(allowed_models, list):
+        all_models = set(allowed_models)
+    elif isinstance(allowed_models, dict):
+        for models in allowed_models.values():
+            if isinstance(models, list):
+                all_models.update(models)
+
+    if not all_models:
+        return "(无)"
+
+    sorted_models = sorted(all_models)
+    preview = ", ".join(sorted_models[:max_items])
+    if len(sorted_models) > max_items:
+        preview += f", ...共{len(sorted_models)}个"
+
+    return preview
+
+
+def is_format_mode(allowed_models: AllowedModels) -> bool:
+    """
+    判断 allowed_models 是否为按格式模式
+
+    Args:
+        allowed_models: 允许的模型配置
+
+    Returns:
+        True: 按格式模式（字典）
+        False: 简单模式（列表或 None）
+    """
+    return isinstance(allowed_models, dict)
+
+
+def convert_to_format_mode(
+    allowed_models: AllowedModels,
+    api_formats: Optional[List[str]] = None,
+) -> Dict[str, List[str]]:
+    """
+    将 allowed_models 转换为按格式模式
+
+    Args:
+        allowed_models: 原始配置
+        api_formats: 要应用的 API 格式列表
+
+    Returns:
+        按格式模式的配置
+    """
+    if allowed_models is None:
+        return {}
+
+    if isinstance(allowed_models, dict):
+        return allowed_models
+
+    # 简单列表模式 -> 按格式模式
+    if isinstance(allowed_models, list):
+        if not api_formats:
+            return {"*": allowed_models}
+        return {fmt.upper(): list(allowed_models) for fmt in api_formats}
+
+    return {}
+
+
+def convert_to_simple_mode(allowed_models: AllowedModels) -> Optional[List[str]]:
+    """
+    将 allowed_models 转换为简单列表模式
+
+    Args:
+        allowed_models: 原始配置
+
+    Returns:
+        简单列表或 None
+    """
+    if allowed_models is None:
+        return None
+
+    if isinstance(allowed_models, list):
+        return allowed_models
+
+    if isinstance(allowed_models, dict):
+        all_models: Set[str] = set()
+        for models in allowed_models.values():
+            if isinstance(models, list):
+                all_models.update(models)
+        return sorted(all_models) if all_models else None
+
+    return None
--- a/src/database/database.py
+++ b/src/database/database.py
@@ -369,7 +369,6 @@ def init_db():
    _ensure_engine()

    # 数据库表结构由 Alembic 迁移管理
-    # 首次部署或更新后请运行: ./migrate.sh

    db = _SessionLocal()
    try:
--- a/src/models/admin_requests.py
+++ b/src/models/admin_requests.py
@@ -52,15 +52,31 @@ class ProxyConfig(BaseModel):
 class CreateProviderRequest(BaseModel):
    """创建 Provider 请求"""

-    name: str = Field(
-        ...,
-        min_length=1,
-        max_length=100,
-        description="Provider 名称（英文字母、数字、下划线、连字符）",
-    )
-    display_name: str = Field(..., min_length=1, max_length=100, description="显示名称")
+    name: str = Field(..., min_length=1, max_length=100, description="提供商名称（唯一）")
    description: Optional[str] = Field(None, max_length=1000, description="描述")
    website: Optional[str] = Field(None, max_length=500, description="官网地址")
+
+    @field_validator("name")
+    @classmethod
+    def validate_name(cls, v: str) -> str:
+        """验证名称格式，防止注入攻击"""
+        v = v.strip()
+
+        # 只允许安全的字符：字母、数字、下划线、连字符、空格、中文
+        if not re.match(r"^[\w\s\u4e00-\u9fff-]+$", v):
+            raise ValueError("名称只能包含字母、数字、下划线、连字符、空格和中文")
+
+        # 检查 SQL 注入关键字（不区分大小写）
+        sql_keywords = [
+            "SELECT", "INSERT", "UPDATE", "DELETE", "DROP", "CREATE",
+            "ALTER", "TRUNCATE", "UNION", "EXEC", "EXECUTE", "--", "/*", "*/"
+        ]
+        v_upper = v.upper()
+        for keyword in sql_keywords:
+            if keyword in v_upper:
+                raise ValueError(f"名称包含非法关键字: {keyword}")
+
+        return v
    billing_type: Optional[str] = Field(
        ProviderBillingType.PAY_AS_YOU_GO.value, description="计费类型"
    )
@@ -68,47 +84,16 @@ class CreateProviderRequest(BaseModel):
    quota_reset_day: Optional[int] = Field(30, ge=1, le=365, description="配额重置周期（天数）")
    quota_last_reset_at: Optional[datetime] = Field(None, description="当前周期开始时间")
    quota_expires_at: Optional[datetime] = Field(None, description="配额过期时间")
-    rpm_limit: Optional[int] = Field(None, ge=0, description="RPM 限制")
    provider_priority: Optional[int] = Field(100, ge=0, le=1000, description="提供商优先级（数字越小越优先）")
    is_active: Optional[bool] = Field(True, description="是否启用")
    concurrent_limit: Optional[int] = Field(None, ge=0, description="并发限制")
+    # 请求配置（从 Endpoint 迁移）
+    timeout: Optional[int] = Field(300, ge=1, le=600, description="请求超时（秒）")
+    max_retries: Optional[int] = Field(2, ge=0, le=10, description="最大重试次数")
+    proxy: Optional[ProxyConfig] = Field(None, description="代理配置")
    config: Optional[Dict[str, Any]] = Field(None, description="其他配置")

-    @field_validator("name")
-    @classmethod
-    def validate_name(cls, v: str) -> str:
-        """验证名称格式"""
-        if not re.match(r"^[a-zA-Z0-9_-]+$", v):
-            raise ValueError("名称只能包含英文字母、数字、下划线和连字符")
-
-        # SQL 注入防护：检查危险关键字
-        dangerous_keywords = [
-            "SELECT",
-            "INSERT",
-            "UPDATE",
-            "DELETE",
-            "DROP",
-            "CREATE",
-            "ALTER",
-            "EXEC",
-            "UNION",
-            "OR",
-            "AND",
-            "--",
-            ";",
-            "'",
-            '"',
-            "<",
-            ">",
-        ]
-        upper_name = v.upper()
-        for keyword in dangerous_keywords:
-            if keyword in upper_name:
-                raise ValueError(f"名称包含禁止的字符或关键字: {keyword}")
-
-        return v
-
-    @field_validator("display_name", "description")
+    @field_validator("name", "description")
    @classmethod
    def sanitize_text(cls, v: Optional[str]) -> Optional[str]:
        """清理文本输入，防止 XSS"""
@@ -162,7 +147,7 @@ class CreateProviderRequest(BaseModel):
 class UpdateProviderRequest(BaseModel):
    """更新 Provider 请求"""

-    display_name: Optional[str] = Field(None, min_length=1, max_length=100)
+    name: Optional[str] = Field(None, min_length=1, max_length=100)
    description: Optional[str] = Field(None, max_length=1000)
    website: Optional[str] = Field(None, max_length=500)
    billing_type: Optional[str] = None
@@ -170,14 +155,17 @@ class UpdateProviderRequest(BaseModel):
    quota_reset_day: Optional[int] = Field(None, ge=1, le=365)
    quota_last_reset_at: Optional[datetime] = None
    quota_expires_at: Optional[datetime] = None
-    rpm_limit: Optional[int] = Field(None, ge=0)
    provider_priority: Optional[int] = Field(None, ge=0, le=1000)
    is_active: Optional[bool] = None
    concurrent_limit: Optional[int] = Field(None, ge=0)
+    # 请求配置（从 Endpoint 迁移）
+    timeout: Optional[int] = Field(None, ge=1, le=600, description="请求超时（秒）")
+    max_retries: Optional[int] = Field(None, ge=0, le=10, description="最大重试次数")
+    proxy: Optional[ProxyConfig] = Field(None, description="代理配置")
    config: Optional[Dict[str, Any]] = None

    # 复用相同的验证器
-    _sanitize_text = field_validator("display_name", "description")(
+    _sanitize_text = field_validator("name", "description")(
        CreateProviderRequest.sanitize_text.__func__
    )
    _validate_website = field_validator("website")(CreateProviderRequest.validate_website.__func__)
@@ -196,7 +184,6 @@ class CreateEndpointRequest(BaseModel):
    custom_path: Optional[str] = Field(None, max_length=200, description="自定义路径")
    priority: Optional[int] = Field(100, ge=0, le=1000, description="优先级")
    is_active: Optional[bool] = Field(True, description="是否启用")
-    rpm_limit: Optional[int] = Field(None, ge=0, description="RPM 限制")
    concurrent_limit: Optional[int] = Field(None, ge=0, description="并发限制")
    config: Optional[Dict[str, Any]] = Field(None, description="其他配置")
    proxy: Optional[ProxyConfig] = Field(None, description="代理配置")
@@ -252,7 +239,6 @@ class UpdateEndpointRequest(BaseModel):
    custom_path: Optional[str] = Field(None, max_length=200)
    priority: Optional[int] = Field(None, ge=0, le=1000)
    is_active: Optional[bool] = None
-    rpm_limit: Optional[int] = Field(None, ge=0)
    concurrent_limit: Optional[int] = Field(None, ge=0)
    config: Optional[Dict[str, Any]] = None
    proxy: Optional[ProxyConfig] = Field(None, description="代理配置")
@@ -277,8 +263,7 @@ class CreateAPIKeyRequest(BaseModel):
    api_key: str = Field(..., min_length=1, max_length=500, description="API Key")
    priority: Optional[int] = Field(100, ge=0, le=1000, description="优先级")
    is_active: Optional[bool] = Field(True, description="是否启用")
-    max_rpm: Optional[int] = Field(None, ge=0, description="最大 RPM")
-    max_concurrent: Optional[int] = Field(None, ge=0, description="最大并发")
+    rpm_limit: Optional[int] = Field(None, ge=0, description="RPM 限制（NULL=自适应）")
    notes: Optional[str] = Field(None, max_length=500, description="备注")

    @field_validator("api_key")
--- a/src/models/api.py
+++ b/src/models/api.py
@@ -376,14 +376,13 @@ class ApiKeyResponse(BaseModel):
 class ProviderCreate(BaseModel):
    """创建提供商请求

-    新架构说明：
+    架构说明：
    - Provider 仅包含提供商的元数据和计费配置
    - API格式、URL、认证等配置应在 ProviderEndpoint 中设置
    - API密钥应在 ProviderAPIKey 中设置
    """

-    name: str = Field(..., min_length=1, max_length=100, description="提供商唯一标识")
-    display_name: str = Field(..., min_length=1, max_length=100, description="显示名称")
+    name: str = Field(..., min_length=1, max_length=100, description="提供商名称（唯一）")
    description: Optional[str] = Field(None, description="提供商描述")
    website: Optional[str] = Field(None, max_length=500, description="主站网站")

@@ -397,7 +396,7 @@ class ProviderCreate(BaseModel):
 class ProviderUpdate(BaseModel):
    """更新提供商请求"""

-    display_name: Optional[str] = Field(None, min_length=1, max_length=100)
+    name: Optional[str] = Field(None, min_length=1, max_length=100)
    description: Optional[str] = None
    website: Optional[str] = Field(None, max_length=500)
    api_format: Optional[str] = None
@@ -418,7 +417,6 @@ class ProviderResponse(BaseModel):

    id: str
    name: str
-    display_name: str
    description: Optional[str]
    website: Optional[str]
    api_format: str
@@ -609,7 +607,6 @@ class PublicProviderResponse(BaseModel):

    id: str
    name: str
-    display_name: str
    description: Optional[str]
    website: Optional[str]
    is_active: bool
@@ -627,7 +624,6 @@ class PublicModelResponse(BaseModel):
    id: str
    provider_id: str
    provider_name: str
-    provider_display_name: str
    name: str
    display_name: str
    description: Optional[str] = None
--- a/src/models/api_key.py
+++ b/src/models/api_key.py
@@ -13,9 +13,7 @@ class ProviderAPIKeyBase(BaseModel):

    name: Optional[str] = Field(None, description="密钥名称/备注")
    api_key: str = Field(..., description="API密钥")
-    rate_limit: Optional[int] = Field(None, description="速率限制（每分钟请求数）")
-    daily_limit: Optional[int] = Field(None, description="每日请求限制")
-    monthly_limit: Optional[int] = Field(None, description="每月请求限制")
+    rpm_limit: Optional[int] = Field(None, description="RPM限制（每分钟请求数），NULL=自适应模式")
    priority: int = Field(0, description="优先级（越高越优先使用）")
    is_active: bool = Field(True, description="是否启用")
    expires_at: Optional[datetime] = Field(None, description="过期时间")
@@ -32,9 +30,7 @@ class ProviderAPIKeyUpdate(BaseModel):

    name: Optional[str] = None
    api_key: Optional[str] = None
-    rate_limit: Optional[int] = None
-    daily_limit: Optional[int] = None
-    monthly_limit: Optional[int] = None
+    rpm_limit: Optional[int] = None
    priority: Optional[int] = None
    is_active: Optional[bool] = None
    expires_at: Optional[datetime] = None
@@ -67,5 +63,3 @@ class ProviderAPIKeyStats(BaseModel):
    last_used_at: Optional[datetime]
    is_active: bool
    is_expired: bool
-    remaining_daily: Optional[int] = Field(None, description="今日剩余请求数")
-    remaining_monthly: Optional[int] = Field(None, description="本月剩余请求数")
--- a/src/models/database.py
+++ b/src/models/database.py
@@ -338,7 +338,8 @@ class Usage(Base):
    request_headers = Column(JSON, nullable=True)  # 客户端请求头
    request_body = Column(JSON, nullable=True)  # 请求体（7天内未压缩）
    provider_request_headers = Column(JSON, nullable=True)  # 向提供商发送的请求头
-    response_headers = Column(JSON, nullable=True)  # 响应头
+    response_headers = Column(JSON, nullable=True)  # 提供商响应头
+    client_response_headers = Column(JSON, nullable=True)  # 返回给客户端的响应头
    response_body = Column(JSON, nullable=True)  # 响应体（7天内未压缩）

    # 压缩存储字段（7天后自动压缩到这里）
@@ -513,8 +514,7 @@ class Provider(Base):
    __tablename__ = "providers"

    id = Column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()), index=True)
-    name = Column(String(100), unique=True, nullable=False, index=True)  # 提供商唯一标识
-    display_name = Column(String(100), nullable=False)  # 显示名称
+    name = Column(String(100), unique=True, nullable=False, index=True)  # 提供商名称（唯一）
    description = Column(Text, nullable=True)  # 提供商描述
    website = Column(String(500), nullable=True)  # 主站网站

@@ -537,11 +537,6 @@ class Provider(Base):
    quota_last_reset_at = Column(DateTime(timezone=True), nullable=True)  # 上次额度重置时间
    quota_expires_at = Column(DateTime(timezone=True), nullable=True)  # 月卡过期时间

-    # RPM限制：NULL=无限制，0=禁止请求
-    rpm_limit = Column(Integer, nullable=True)  # 每分钟请求数限制（NULL=无限制，0=禁止请求）
-    rpm_used = Column(Integer, default=0)  # 当前分钟已用请求数
-    rpm_reset_at = Column(DateTime(timezone=True), nullable=True)  # RPM重置时间
-
    # 提供商优先级 (数字越小越优先，用于提供商优先模式下的 Provider 排序)
    # 0-10: 急需消耗(如即将过期的月卡)
    # 11-50: 优先消耗(月卡)
@@ -555,6 +550,15 @@ class Provider(Base):
    # 限制
    concurrent_limit = Column(Integer, nullable=True)  # 并发请求限制

+    # 请求配置（从 Endpoint 迁移，作为全局默认值）
+    # 超时 300 秒对于 LLM API 是合理的默认值：
+    # - 大多数请求在 30 秒内完成
+    # - 复杂推理（如 Claude thinking）可能需要 60-120 秒
+    # - 300 秒足够覆盖极端场景（如超长上下文、复杂工具调用）
+    timeout = Column(Integer, default=300, nullable=True)  # 请求超时（秒）
+    max_retries = Column(Integer, default=2, nullable=True)  # 最大重试次数
+    proxy = Column(JSONB, nullable=True)  # 代理配置: {url, username, password, enabled}
+
    # 配置
    config = Column(JSON, nullable=True)  # 额外配置（如Azure deployment name等）

@@ -574,6 +578,9 @@ class Provider(Base):
    endpoints = relationship(
        "ProviderEndpoint", back_populates="provider", cascade="all, delete-orphan"
    )
+    api_keys = relationship(
+        "ProviderAPIKey", back_populates="provider", cascade="all, delete-orphan"
+    )
    api_key_mappings = relationship(
        "ApiKeyProviderMapping", back_populates="provider", cascade="all, delete-orphan"
    )
@@ -599,12 +606,6 @@ class ProviderEndpoint(Base):
    timeout = Column(Integer, default=300)  # 超时（秒）
    max_retries = Column(Integer, default=2)  # 最大重试次数

-    # 限制
-    max_concurrent = Column(
-        Integer, nullable=True, default=None
-    )  # 该端点的最大并发数（NULL=不限制）
-    rate_limit = Column(Integer, nullable=True)  # 每分钟请求限制
-
    # 状态
    is_active = Column(Boolean, default=True, nullable=False)

@@ -632,9 +633,6 @@ class ProviderEndpoint(Base):

    # 关系
    provider = relationship("Provider", back_populates="endpoints")
-    api_keys = relationship(
-        "ProviderAPIKey", back_populates="endpoint", cascade="all, delete-orphan"
-    )

    # 唯一约束和索引在表定义后
    __table_args__ = (
@@ -734,9 +732,11 @@ class GlobalModel(Base):
 class Model(Base):
    """Provider 模型配置表 - Provider 如何使用某个 GlobalModel

-    设计原则 (方案 A):
-    - 每个 Model 必须关联一个 GlobalModel (global_model_id 不可为空)
-    - Model 表示 Provider 对某个 GlobalModel 的具体实现
+    设计原则:
+    - Model 表示 Provider 对某个模型的具体实现
+    - global_model_id 可为空：
+      - 为空时：模型尚未关联到 GlobalModel，不参与路由
+      - 不为空时：模型已关联 GlobalModel，参与路由
    - provider_model_name 是 Provider 侧的实际模型名称 (可能与 GlobalModel.name 不同)
    - 价格和能力配置可为空，为空时使用 GlobalModel 的默认值
    """
@@ -745,7 +745,8 @@ class Model(Base):

    id = Column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()), index=True)
    provider_id = Column(String(36), ForeignKey("providers.id"), nullable=False)
-    global_model_id = Column(String(36), ForeignKey("global_models.id"), nullable=False, index=True)
+    # 可为空：NULL 表示未关联，不参与路由；非 NULL 表示已关联，参与路由
+    global_model_id = Column(String(36), ForeignKey("global_models.id"), nullable=True, index=True)

    # Provider 映射配置
    provider_model_name = Column(String(200), nullable=False)  # Provider 侧的主模型名称
@@ -983,17 +984,20 @@ class Model(Base):


 class ProviderAPIKey(Base):
-    """Provider API密钥表 - 归属于特定 ProviderEndpoint"""
+    """Provider API密钥表 - 直接归属于 Provider，支持多种 API 格式"""

    __tablename__ = "provider_api_keys"

    id = Column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()), index=True)

-    # 外键关系
-    endpoint_id = Column(
-        String(36), ForeignKey("provider_endpoints.id", ondelete="CASCADE"), nullable=False
+    # 外键关系 - 直接关联 Provider
+    provider_id = Column(
+        String(36), ForeignKey("providers.id", ondelete="CASCADE"), nullable=False, index=True
    )

+    # API 格式支持列表（核心字段）
+    api_formats = Column(JSON, nullable=False, default=list)  # ["CLAUDE", "CLAUDE_CLI"]
+
    # API密钥信息
    api_key = Column(String(500), nullable=False)  # API密钥（加密存储）
    name = Column(String(100), nullable=False)  # 密钥名称（必填，用于识别）
@@ -1002,7 +1006,10 @@ class ProviderAPIKey(Base):
    # 成本计算
    rate_multiplier = Column(
        Float, default=1.0, nullable=False
-    )  # 成本倍率（真实成本 = 表面成本 × 倍率）
+    )  # 默认成本倍率（真实成本 = 表面成本 × 倍率）
+    rate_multipliers = Column(
+        JSON, nullable=True
+    )  # 按 API 格式的成本倍率 {"CLAUDE": 1.0, "OPENAI": 0.8}

    # 优先级配置 (数字越小越优先)
    internal_priority = Column(
@@ -1012,14 +1019,11 @@ class ProviderAPIKey(Base):
        Integer, nullable=True
    )  # 全局 Key 优先级（用于全局 Key 优先模式，跨 Provider 的 Key 排序，NULL=未配置使用默认排序）

-    # 并发限制配置
-    # max_concurrent 决定并发控制模式：
-    #   - NULL: 自适应模式，系统自动学习并调整（使用 learned_max_concurrent）
+    # RPM 限制配置（自适应学习）
+    # rpm_limit 决定 RPM 控制模式：
+    #   - NULL: 自适应模式，系统自动学习并调整（使用 learned_rpm_limit）
    #   - 数字: 固定限制模式，使用用户指定的值
-    max_concurrent = Column(Integer, nullable=True, default=None)
-    rate_limit = Column(Integer, nullable=True)  # 速率限制（每分钟请求数）
-    daily_limit = Column(Integer, nullable=True)  # 每日请求限制
-    monthly_limit = Column(Integer, nullable=True)  # 每月请求限制
+    rpm_limit = Column(Integer, nullable=True, default=None)

    # 模型权限控制
    allowed_models = Column(JSON, nullable=True)  # 允许使用的模型列表（null = 支持所有模型）
@@ -1028,16 +1032,16 @@ class ProviderAPIKey(Base):
    capabilities = Column(JSON, nullable=True)  # Key 拥有的能力
    # 示例: {"cache_1h": true, "context_1m": true}

-    # 自适应并发调整（仅当 max_concurrent = NULL 时生效）
-    learned_max_concurrent = Column(
+    # 自适应 RPM 调整（仅当 rpm_limit = NULL 时生效）
+    learned_rpm_limit = Column(
        Integer, nullable=True
-    )  # 学习到的并发限制（自适应模式下的有效值）
+    )  # 学习到的 RPM 限制（自适应模式下的有效值）
    concurrent_429_count = Column(Integer, default=0, nullable=False)  # 因并发导致的429次数
    rpm_429_count = Column(Integer, default=0, nullable=False)  # 因RPM导致的429次数
    last_429_at = Column(DateTime(timezone=True), nullable=True)  # 最后429时间
    last_429_type = Column(String(50), nullable=True)  # 最后429类型: concurrent/rpm/unknown
-    last_concurrent_peak = Column(Integer, nullable=True)  # 触发429时的并发数
-    adjustment_history = Column(JSON, nullable=True)  # 并发调整历史
+    last_rpm_peak = Column(Integer, nullable=True)  # 触发429时的RPM峰值
+    adjustment_history = Column(JSON, nullable=True)  # RPM调整历史
    # 基于滑动窗口的利用率追踪
    utilization_samples = Column(
        JSON, nullable=True
@@ -1046,12 +1050,9 @@ class ProviderAPIKey(Base):
        DateTime(timezone=True), nullable=True
    )  # 上次探测性扩容时间

-    # 健康度追踪（基于滑动窗口）
-    health_score = Column(Float, default=1.0)  # 0.0-1.0（保留用于展示，实际熔断基于滑动窗口）
-    consecutive_failures = Column(Integer, default=0)
-    last_failure_at = Column(DateTime(timezone=True), nullable=True)  # 最后失败时间
-    # 滑动窗口：记录最近 N 次请求的结果 [{"ts": timestamp, "ok": true/false}, ...]
-    request_results_window = Column(JSON, nullable=True)
+    # 健康度追踪（按 API 格式存储）
+    # 结构: {"CLAUDE": {"health_score": 1.0, "consecutive_failures": 0, "last_failure_at": null, "request_results_window": []}, ...}
+    health_by_format = Column(JSON, nullable=True, default=dict)

    # 缓存与熔断配置
    cache_ttl_minutes = Column(
@@ -1061,14 +1062,9 @@ class ProviderAPIKey(Base):
        Integer, default=32, nullable=False
    )  # 最大探测间隔(分钟)，默认32分钟（硬上限）

-    # 熔断器字段（滑动窗口 + 半开状态模式）
-    circuit_breaker_open = Column(Boolean, default=False, nullable=False)  # 熔断器是否打开
-    circuit_breaker_open_at = Column(DateTime(timezone=True), nullable=True)  # 熔断器打开时间
-    next_probe_at = Column(DateTime(timezone=True), nullable=True)  # 下次探测时间
-    # 半开状态：允许少量请求通过验证服务是否恢复
-    half_open_until = Column(DateTime(timezone=True), nullable=True)  # 半开状态结束时间
-    half_open_successes = Column(Integer, default=0)  # 半开状态下的成功次数
-    half_open_failures = Column(Integer, default=0)  # 半开状态下的失败次数
+    # 熔断器状态（按 API 格式存储）
+    # 结构: {"CLAUDE": {"open": false, "open_at": null, "next_probe_at": null, "half_open_until": null, "half_open_successes": 0, "half_open_failures": 0}, ...}
+    circuit_breaker_by_format = Column(JSON, nullable=True, default=dict)

    # 使用统计
    request_count = Column(Integer, default=0)  # 请求次数
@@ -1095,7 +1091,7 @@ class ProviderAPIKey(Base):
    )

    # 关系
-    endpoint = relationship("ProviderEndpoint", back_populates="api_keys")
+    provider = relationship("Provider", back_populates="api_keys")


 class UserPreference(Base):
--- a/src/models/endpoint_models.py
+++ b/src/models/endpoint_models.py
@@ -4,7 +4,7 @@ ProviderEndpoint 相关的 API 模型定义

 import re
 from datetime import datetime
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, List, Optional, Union

 from pydantic import BaseModel, ConfigDict, Field, field_validator

@@ -26,10 +26,6 @@ class ProviderEndpointCreate(BaseModel):
    timeout: int = Field(default=300, ge=10, le=600, description="超时时间（秒）")
    max_retries: int = Field(default=2, ge=0, le=10, description="最大重试次数")

-    # 限制
-    max_concurrent: Optional[int] = Field(default=None, ge=1, description="最大并发数")
-    rate_limit: Optional[int] = Field(default=None, ge=1, description="速率限制（请求/秒）")
-
    # 额外配置
    config: Optional[Dict[str, Any]] = Field(default=None, description="额外配置（JSON）")

@@ -67,8 +63,6 @@ class ProviderEndpointUpdate(BaseModel):
    headers: Optional[Dict[str, str]] = Field(default=None, description="自定义请求头")
    timeout: Optional[int] = Field(default=None, ge=10, le=600, description="超时时间（秒）")
    max_retries: Optional[int] = Field(default=None, ge=0, le=10, description="最大重试次数")
-    max_concurrent: Optional[int] = Field(default=None, ge=1, description="最大并发数")
-    rate_limit: Optional[int] = Field(default=None, ge=1, description="速率限制")
    is_active: Optional[bool] = Field(default=None, description="是否启用")
    config: Optional[Dict[str, Any]] = Field(default=None, description="额外配置")
    proxy: Optional[ProxyConfig] = Field(default=None, description="代理配置")
@@ -103,10 +97,6 @@ class ProviderEndpointResponse(BaseModel):
    timeout: int
    max_retries: int

-    # 限制
-    max_concurrent: Optional[int] = None
-    rate_limit: Optional[int] = None
-
    # 状态
    is_active: bool

@@ -127,32 +117,37 @@ class ProviderEndpointResponse(BaseModel):
    model_config = ConfigDict(from_attributes=True)


-# ========== ProviderAPIKey 相关（新架构） ==========
+# ========== ProviderAPIKey 相关 ==========


 class EndpointAPIKeyCreate(BaseModel):
-    """为 Endpoint 添加 API Key"""
+    """为 Provider 添加 API Key"""
+
+    provider_id: Optional[str] = Field(default=None, description="Provider ID（从 URL 获取）")
+    api_formats: Optional[List[str]] = Field(
+        default=None, min_length=1, description="支持的 API 格式列表（必填，路由层校验）"
+    )

-    endpoint_id: str = Field(..., description="Endpoint ID")
    api_key: str = Field(..., min_length=3, max_length=500, description="API Key（将自动加密）")
    name: str = Field(..., min_length=1, max_length=100, description="密钥名称（必填，用于识别）")

    # 成本计算
    rate_multiplier: float = Field(
-        default=1.0, ge=0.01, description="成本倍率（真实成本 = 表面成本 × 倍率）"
+        default=1.0, ge=0.01, description="默认成本倍率（真实成本 = 表面成本 × 倍率）"
+    )
+    rate_multipliers: Optional[Dict[str, float]] = Field(
+        default=None, description="按 API 格式的成本倍率，如 {'CLAUDE': 1.0, 'OPENAI': 0.8}"
    )

    # 优先级和限制（数字越小越优先）
-    internal_priority: int = Field(default=50, description="Endpoint 内部优先级（提供商优先模式）")
-    # max_concurrent: NULL=自适应模式（系统自动学习），数字=固定限制模式
-    max_concurrent: Optional[int] = Field(
-        default=None, ge=1, description="最大并发数（NULL=自适应模式）"
+    internal_priority: int = Field(default=50, description="Key 内部优先级（提供商优先模式）")
+    # rpm_limit: NULL=自适应模式（系统自动学习），数字=固定限制模式
+    rpm_limit: Optional[int] = Field(
+        default=None, ge=1, le=10000, description="RPM 限制（NULL=自适应模式）"
    )
-    rate_limit: Optional[int] = Field(default=None, ge=1, description="速率限制")
-    daily_limit: Optional[int] = Field(default=None, ge=1, description="每日限制")
-    monthly_limit: Optional[int] = Field(default=None, ge=1, description="每月限制")
-    allowed_models: Optional[List[str]] = Field(
-        default=None, description="允许使用的模型列表（null = 支持所有模型）"
+    allowed_models: Optional[Union[List[str], Dict[str, List[str]]]] = Field(
+        default=None,
+        description="允许使用的模型列表（null=不限制，列表=简单白名单，字典=按API格式区分）",
    )

    # 能力标签
@@ -171,6 +166,92 @@ class EndpointAPIKeyCreate(BaseModel):
    # 备注
    note: Optional[str] = Field(default=None, max_length=500, description="备注说明（可选）")

+    @field_validator("api_formats")
+    @classmethod
+    def validate_api_formats(cls, v: Optional[List[str]]) -> Optional[List[str]]:
+        """验证 API 格式列表"""
+        if v is None:
+            return v
+
+        from src.core.enums import APIFormat
+
+        allowed = [fmt.value for fmt in APIFormat]
+        validated = []
+        seen = set()
+        for fmt in v:
+            fmt_upper = fmt.upper()
+            if fmt_upper not in allowed:
+                raise ValueError(f"API 格式必须是 {allowed} 之一，当前值: {fmt}")
+            if fmt_upper in seen:
+                continue  # 静默去重
+            seen.add(fmt_upper)
+            validated.append(fmt_upper)
+        return validated
+
+    @field_validator("allowed_models")
+    @classmethod
+    def validate_allowed_models(
+        cls, v: Optional[Union[List[str], Dict[str, List[str]]]]
+    ) -> Optional[Union[List[str], Dict[str, List[str]]]]:
+        """
+        规范化 allowed_models：
+        - 列表模式：去空、去重、保留顺序
+        - 字典模式：key 统一大写（支持 "*"），value 去空、去重、保留顺序
+        """
+        if v is None:
+            return v
+
+        if isinstance(v, list):
+            cleaned: List[str] = []
+            seen: set[str] = set()
+            for item in v:
+                if not isinstance(item, str):
+                    raise ValueError("allowed_models 列表必须为字符串数组")
+                name = item.strip()
+                if not name or name in seen:
+                    continue
+                seen.add(name)
+                cleaned.append(name)
+            return cleaned
+
+        if isinstance(v, dict):
+            from src.core.enums import APIFormat
+
+            allowed_formats = {fmt.value for fmt in APIFormat}
+            normalized: Dict[str, List[str]] = {}
+            for raw_key, models in v.items():
+                if not isinstance(raw_key, str):
+                    raise ValueError("allowed_models 字典的 key 必须为字符串")
+
+                key = raw_key.upper()
+                if key != "*" and key not in allowed_formats:
+                    raise ValueError(
+                        f"allowed_models 字典的 key 必须是 {sorted(allowed_formats)} 或 '*'，当前值: {raw_key}"
+                    )
+
+                if models is None:
+                    # null 表示该格式不限制，跳过（不加入字典）
+                    continue
+                if not isinstance(models, list):
+                    raise ValueError("allowed_models 字典的 value 必须为字符串数组")
+
+                cleaned: List[str] = []
+                seen: set[str] = set()
+                for item in models:
+                    if not isinstance(item, str):
+                        raise ValueError("allowed_models 字典的 value 必须为字符串数组")
+                    name = item.strip()
+                    if not name or name in seen:
+                        continue
+                    seen.add(name)
+                    cleaned.append(name)
+
+                normalized[key] = cleaned
+
+            return normalized
+
+        raise ValueError("allowed_models 必须是列表或字典")
+
    @field_validator("api_key")
    @classmethod
    def validate_api_key(cls, v: str) -> str:
@@ -214,26 +295,35 @@ class EndpointAPIKeyCreate(BaseModel):
 class EndpointAPIKeyUpdate(BaseModel):
    """更新 Endpoint API Key"""

+    api_formats: Optional[List[str]] = Field(
+        default=None, min_length=1, description="支持的 API 格式列表"
+    )
+
    api_key: Optional[str] = Field(
        default=None, min_length=3, max_length=500, description="API Key（将自动加密）"
    )
    name: Optional[str] = Field(default=None, min_length=1, max_length=100, description="密钥名称")
-    rate_multiplier: Optional[float] = Field(default=None, ge=0.01, description="成本倍率")
+    rate_multiplier: Optional[float] = Field(default=None, ge=0.01, description="默认成本倍率")
+    rate_multipliers: Optional[Dict[str, float]] = Field(
+        default=None, description="按 API 格式的成本倍率，如 {'CLAUDE': 1.0, 'OPENAI': 0.8}"
+    )
    internal_priority: Optional[int] = Field(
-        default=None, description="Endpoint 内部优先级（提供商优先模式，数字越小越优先）"
+        default=None, description="Key 内部优先级（提供商优先模式，数字越小越优先）"
    )
    global_priority: Optional[int] = Field(
        default=None, description="全局 Key 优先级（全局 Key 优先模式，数字越小越优先）"
    )
-    # max_concurrent: 使用特殊标记区分"未提供"和"设置为 null（自适应模式）"
+    # rpm_limit: 使用特殊标记区分"未提供"和"设置为 null（自适应模式）"
    # - 不提供字段：不更新
    # - 提供 null：切换为自适应模式
-    # - 提供数字：设置固定并发限制
-    max_concurrent: Optional[int] = Field(default=None, ge=1, description="最大并发数（null=自适应模式）")
-    rate_limit: Optional[int] = Field(default=None, ge=1, description="速率限制")
-    daily_limit: Optional[int] = Field(default=None, ge=1, description="每日限制")
-    monthly_limit: Optional[int] = Field(default=None, ge=1, description="每月限制")
-    allowed_models: Optional[List[str]] = Field(default=None, description="允许使用的模型列表")
+    # - 提供数字：设置固定 RPM 限制
+    rpm_limit: Optional[int] = Field(
+        default=None, ge=1, le=10000, description="RPM 限制（null=自适应模式）"
+    )
+    allowed_models: Optional[Union[List[str], Dict[str, List[str]]]] = Field(
+        default=None,
+        description="允许使用的模型列表（null=不限制，列表=简单白名单，字典=按API格式区分）",
+    )
    capabilities: Optional[Dict[str, bool]] = Field(
        default=None, description="Key 能力标签，如 {'cache_1h': true, 'context_1m': true}"
    )
@@ -246,6 +336,36 @@ class EndpointAPIKeyUpdate(BaseModel):
    is_active: Optional[bool] = Field(default=None, description="是否启用")
    note: Optional[str] = Field(default=None, max_length=500, description="备注说明")

+    @field_validator("api_formats")
+    @classmethod
+    def validate_api_formats(cls, v: Optional[List[str]]) -> Optional[List[str]]:
+        """验证 API 格式列表"""
+        if v is None:
+            return v
+
+        from src.core.enums import APIFormat
+
+        allowed = [fmt.value for fmt in APIFormat]
+        validated = []
+        seen = set()
+        for fmt in v:
+            fmt_upper = fmt.upper()
+            if fmt_upper not in allowed:
+                raise ValueError(f"API 格式必须是 {allowed} 之一，当前值: {fmt}")
+            if fmt_upper in seen:
+                continue  # 静默去重
+            seen.add(fmt_upper)
+            validated.append(fmt_upper)
+        return validated
+
+    @field_validator("allowed_models")
+    @classmethod
+    def validate_allowed_models(
+        cls, v: Optional[Union[List[str], Dict[str, List[str]]]]
+    ) -> Optional[Union[List[str], Dict[str, List[str]]]]:
+        # 与 EndpointAPIKeyCreate 保持一致
+        return EndpointAPIKeyCreate.validate_allowed_models(v)
+
    @field_validator("api_key")
    @classmethod
    def validate_api_key(cls, v: Optional[str]) -> Optional[str]:
@@ -295,7 +415,9 @@ class EndpointAPIKeyResponse(BaseModel):
    """Endpoint API Key 响应"""

    id: str
-    endpoint_id: str
+
+    provider_id: str = Field(..., description="Provider ID")
+    api_formats: List[str] = Field(default=[], description="支持的 API 格式列表")

    # Key 信息（脱敏）
    api_key_masked: str = Field(..., description="脱敏后的 Key")
@@ -303,31 +425,37 @@ class EndpointAPIKeyResponse(BaseModel):
    name: str = Field(..., description="密钥名称")

    # 成本计算
-    rate_multiplier: float = Field(default=1.0, description="成本倍率")
+    rate_multiplier: float = Field(default=1.0, description="默认成本倍率")
+    rate_multipliers: Optional[Dict[str, float]] = Field(
+        default=None, description="按 API 格式的成本倍率，如 {'CLAUDE': 1.0, 'OPENAI': 0.8}"
+    )

    # 优先级和限制
    internal_priority: int = Field(default=50, description="Endpoint 内部优先级")
    global_priority: Optional[int] = Field(default=None, description="全局 Key 优先级")
-    max_concurrent: Optional[int] = None
-    rate_limit: Optional[int] = None
-    daily_limit: Optional[int] = None
-    monthly_limit: Optional[int] = None
-    allowed_models: Optional[List[str]] = None
-    capabilities: Optional[Dict[str, bool]] = Field(
-        default=None, description="Key 能力标签"
-    )
+    rpm_limit: Optional[int] = None
+    allowed_models: Optional[Union[List[str], Dict[str, List[str]]]] = None
+    capabilities: Optional[Dict[str, bool]] = Field(default=None, description="Key 能力标签")

    # 缓存与熔断配置
    cache_ttl_minutes: int = Field(default=5, description="缓存 TTL（分钟），0=禁用")
    max_probe_interval_minutes: int = Field(default=32, description="熔断探测间隔（分钟）")

-    # 健康度
-    health_score: float
-    consecutive_failures: int
+    # 按格式的健康度数据
+    health_by_format: Optional[Dict[str, Any]] = Field(
+        default=None, description="按 API 格式存储的健康度数据"
+    )
+    circuit_breaker_by_format: Optional[Dict[str, Any]] = Field(
+        default=None, description="按 API 格式存储的熔断器状态"
+    )
+
+    # 聚合字段（从 health_by_format 计算，用于列表显示）
+    health_score: float = Field(default=1.0, description="健康度（所有格式中的最低值）")
+    consecutive_failures: int = Field(default=0, description="连续失败次数")
    last_failure_at: Optional[datetime] = None

-    # 熔断器状态（滑动窗口 + 半开模式）
-    circuit_breaker_open: bool = Field(default=False, description="熔断器是否打开")
+    # 聚合熔断器字段
+    circuit_breaker_open: bool = Field(default=False, description="熔断器是否打开（任何格式）")
    circuit_breaker_open_at: Optional[datetime] = Field(default=None, description="熔断器打开时间")
    next_probe_at: Optional[datetime] = Field(default=None, description="下次进入半开状态时间")
    half_open_until: Optional[datetime] = Field(default=None, description="半开状态结束时间")
@@ -345,9 +473,9 @@ class EndpointAPIKeyResponse(BaseModel):
    # 状态
    is_active: bool

-    # 自适应并发信息
-    is_adaptive: bool = Field(default=False, description="是否为自适应模式（max_concurrent=NULL）")
-    learned_max_concurrent: Optional[int] = Field(None, description="学习到的并发限制")
+    # 自适应 RPM 信息
+    is_adaptive: bool = Field(default=False, description="是否为自适应模式（rpm_limit=NULL）")
+    learned_rpm_limit: Optional[int] = Field(None, description="学习到的 RPM 限制")
    effective_limit: Optional[int] = Field(None, description="当前有效限制")
    # 滑动窗口利用率采样
    utilization_samples: Optional[List[dict]] = Field(None, description="利用率采样窗口")
@@ -371,22 +499,42 @@ class EndpointAPIKeyResponse(BaseModel):
 # ========== 健康监控相关 ==========


-class HealthStatusResponse(BaseModel):
-    """健康状态响应（仅 Key 级别）"""
+class FormatHealthData(BaseModel):
+    """单个 API 格式的健康度数据"""

-    # Key 健康状态
+    health_score: float = 1.0
+    error_rate: float = 0.0
+    window_size: int = 0
+    consecutive_failures: int = 0
+    last_failure_at: Optional[str] = None
+    circuit_breaker: Dict[str, Any] = Field(default_factory=dict)
+
+
+class HealthStatusResponse(BaseModel):
+    """健康状态响应（支持按格式查询）"""
+
+    # 基础信息
    key_id: str
-    key_health_score: float
-    key_consecutive_failures: int
-    key_last_failure_at: Optional[datetime] = None
    key_is_active: bool
    key_statistics: Optional[Dict[str, Any]] = None

-    # 熔断器状态（滑动窗口 + 半开模式）
+    # 整体健康度（取所有格式中的最低值）
+    key_health_score: float = 1.0
+    any_circuit_open: bool = False
+
+    # 按格式的健康度数据
+    health_by_format: Optional[Dict[str, FormatHealthData]] = None
+
+    # 单格式查询时的字段
+    api_format: Optional[str] = None
+    key_consecutive_failures: Optional[int] = None
+    key_last_failure_at: Optional[str] = None
+
+    # 单格式查询时的熔断器状态
    circuit_breaker_open: bool = False
-    circuit_breaker_open_at: Optional[datetime] = None
-    next_probe_at: Optional[datetime] = None
-    half_open_until: Optional[datetime] = None
+    circuit_breaker_open_at: Optional[str] = None
+    next_probe_at: Optional[str] = None
+    half_open_until: Optional[str] = None
    half_open_successes: int = 0
    half_open_failures: int = 0

@@ -398,33 +546,22 @@ class HealthSummaryResponse(BaseModel):
    keys: Dict[str, int] = Field(..., description="Key 统计 (total, active, unhealthy)")


-# ========== 并发控制相关 ==========
+# ========== RPM 控制相关 ==========


-class ConcurrencyStatusResponse(BaseModel):
-    """并发状态响应"""
+class KeyRpmStatusResponse(BaseModel):
+    """Key RPM 状态响应"""

-    endpoint_id: Optional[str] = None
-    endpoint_current_concurrency: int = Field(default=0, description="Endpoint 当前并发数")
-    endpoint_max_concurrent: Optional[int] = Field(default=None, description="Endpoint 最大并发数")
-
-    key_id: Optional[str] = None
-    key_current_concurrency: int = Field(default=0, description="Key 当前并发数")
-    key_max_concurrent: Optional[int] = Field(default=None, description="Key 最大并发数")
-
-
-class ResetConcurrencyRequest(BaseModel):
-    """重置并发计数请求"""
-
-    endpoint_id: Optional[str] = Field(default=None, description="Endpoint ID（可选）")
-    key_id: Optional[str] = Field(default=None, description="Key ID（可选）")
+    key_id: str = Field(..., description="Key ID")
+    current_rpm: int = Field(default=0, description="当前 RPM 计数")
+    rpm_limit: Optional[int] = Field(default=None, description="RPM 限制")


 class KeyPriorityItem(BaseModel):
    """单个 Key 优先级项"""

    key_id: str = Field(..., description="Key ID")
-    internal_priority: int = Field(..., ge=0, description="Endpoint 内部优先级（数字越小越优先）")
+    internal_priority: int = Field(..., ge=0, description="Key 内部优先级（数字越小越优先）")


 class BatchUpdateKeyPriorityRequest(BaseModel):
@@ -439,11 +576,9 @@ class BatchUpdateKeyPriorityRequest(BaseModel):
 class ProviderUpdateRequest(BaseModel):
    """Provider 基础配置更新请求"""

-    display_name: Optional[str] = Field(None, min_length=1, max_length=100)
+    name: Optional[str] = Field(None, min_length=1, max_length=100)
    description: Optional[str] = None
    website: Optional[str] = Field(None, max_length=500, description="主站网站")
-    priority: Optional[int] = None
-    weight: Optional[float] = Field(None, gt=0)
    provider_priority: Optional[int] = Field(None, description="提供商优先级(数字越小越优先)")
    is_active: Optional[bool] = None
    billing_type: Optional[str] = Field(
@@ -452,9 +587,10 @@ class ProviderUpdateRequest(BaseModel):
    monthly_quota_usd: Optional[float] = Field(None, ge=0, description="订阅配额（美元）")
    quota_reset_day: Optional[int] = Field(None, ge=1, le=31, description="配额重置日（1-31）")
    quota_expires_at: Optional[datetime] = Field(None, description="配额过期时间")
-    rpm_limit: Optional[int] = Field(
-        None, ge=0, description="每分钟请求数限制（NULL=无限制，0=禁止请求）"
-    )
+    # 请求配置（从 Endpoint 迁移）
+    timeout: Optional[int] = Field(None, ge=1, le=600, description="请求超时（秒）")
+    max_retries: Optional[int] = Field(None, ge=0, le=10, description="最大重试次数")
+    proxy: Optional[Dict[str, Any]] = Field(None, description="代理配置")


 class ProviderWithEndpointsSummary(BaseModel):
@@ -463,7 +599,6 @@ class ProviderWithEndpointsSummary(BaseModel):
    # Provider 基本信息
    id: str
    name: str
-    display_name: str
    description: Optional[str] = None
    website: Optional[str] = None
    provider_priority: int = Field(default=100, description="提供商优先级(数字越小越优先)")
@@ -477,12 +612,10 @@ class ProviderWithEndpointsSummary(BaseModel):
    quota_last_reset_at: Optional[datetime] = Field(default=None, description="当前周期开始时间")
    quota_expires_at: Optional[datetime] = Field(default=None, description="配额过期时间")

-    # RPM 限制
-    rpm_limit: Optional[int] = Field(
-        default=None, description="每分钟请求数限制（NULL=无限制，0=禁止请求）"
-    )
-    rpm_used: Optional[int] = Field(default=None, description="当前分钟已用请求数")
-    rpm_reset_at: Optional[datetime] = Field(default=None, description="RPM 重置时间")
+    # 请求配置（从 Endpoint 迁移）
+    timeout: Optional[int] = Field(default=300, description="请求超时（秒）")
+    max_retries: Optional[int] = Field(default=2, description="最大重试次数")
+    proxy: Optional[Dict[str, Any]] = Field(default=None, description="代理配置")

    # Endpoint 统计
    total_endpoints: int = Field(default=0, description="总 Endpoint 数量")
@@ -617,12 +750,8 @@ class PublicApiFormatHealthMonitor(BaseModel):
        default_factory=list,
        description="Usage 表生成的健康时间线（healthy/warning/unhealthy/unknown）",
    )
-    time_range_start: Optional[datetime] = Field(
-        default=None, description="时间线覆盖区间开始时间"
-    )
-    time_range_end: Optional[datetime] = Field(
-        default=None, description="时间线覆盖区间结束时间"
-    )
+    time_range_start: Optional[datetime] = Field(default=None, description="时间线覆盖区间开始时间")
+    time_range_end: Optional[datetime] = Field(default=None, description="时间线覆盖区间结束时间")


 class PublicApiFormatHealthMonitorResponse(BaseModel):
--- a/src/models/pydantic_models.py
+++ b/src/models/pydantic_models.py
@@ -114,7 +114,6 @@ class ModelCatalogProviderDetail(BaseModel):

    provider_id: str
    provider_name: str
-    provider_display_name: Optional[str]
    model_id: Optional[str]
    target_model: str
    input_price_per_1m: Optional[float]
@@ -312,16 +311,26 @@ class ImportFromUpstreamRequest(BaseModel):
    """从上游提供商导入模型请求"""

    model_ids: List[str] = Field(..., min_length=1, description="上游模型 ID 列表")
+    # 价格覆盖配置（应用于所有导入的模型）
+    tiered_pricing: Optional[Dict] = Field(
+        None,
+        description="阶梯计费配置（可选），格式: {tiers: [{up_to, input_price_per_1m, output_price_per_1m, ...}]}"
+    )
+    price_per_request: Optional[float] = Field(
+        None,
+        ge=0,
+        description="按次计费价格（可选，单位：美元）"
+    )


 class ImportFromUpstreamSuccessItem(BaseModel):
    """导入成功的模型信息"""

    model_id: str = Field(..., description="上游模型 ID")
-    global_model_id: str = Field(..., description="GlobalModel ID")
-    global_model_name: str = Field(..., description="GlobalModel 名称")
    provider_model_id: str = Field(..., description="Provider Model ID")
-    created_global_model: bool = Field(..., description="是否新创建了 GlobalModel")
+    global_model_id: Optional[str] = Field("", description="GlobalModel ID（如果已关联）")
+    global_model_name: Optional[str] = Field("", description="GlobalModel 名称（如果已关联）")
+    created_global_model: bool = Field(False, description="是否新创建了 GlobalModel（始终为 false）")


 class ImportFromUpstreamErrorItem(BaseModel):
--- a/src/services/cache/aware_scheduler.py
+++ b/src/services/cache/aware_scheduler.py
@@ -34,7 +34,7 @@ import hashlib
 import random
 import time
 from dataclasses import dataclass
-from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union

 from sqlalchemy.orm import Session, selectinload

@@ -80,8 +80,6 @@ class ProviderCandidate:

@dataclass
 class ConcurrencySnapshot:
-    endpoint_current: int
-    endpoint_limit: Optional[int]
    key_current: int
    key_limit: Optional[int]
    is_cached_user: bool = False
@@ -91,11 +89,9 @@ class ConcurrencySnapshot:
    reservation_confidence: float = 0.0

    def describe(self) -> str:
-        endpoint_limit_text = str(self.endpoint_limit) if self.endpoint_limit is not None else "inf"
        key_limit_text = str(self.key_limit) if self.key_limit is not None else "inf"
        reservation_text = f"{self.reservation_ratio:.0%}" if self.reservation_ratio > 0 else "N/A"
        return (
-            f"endpoint={self.endpoint_current}/{endpoint_limit_text}, "
            f"key={self.key_current}/{key_limit_text}, "
            f"cached={self.is_cached_user}, "
            f"reserve={reservation_text}({self.reservation_phase})"
@@ -246,9 +242,8 @@ class CacheAwareScheduler:

            if not candidates:
                if provider_offset == 0:
-                    # 没有找到任何候选，提供友好的错误提示
-                    error_msg = f"模型 '{model_name}' 不可用"
-                    raise ProviderNotAvailableException(error_msg)
+                    # 没有找到任何候选，提供友好的错误提示（不暴露内部信息）
+                    raise ProviderNotAvailableException("请求的模型当前不可用")
                break

            self._metrics["total_batches"] += 1
@@ -270,7 +265,6 @@ class CacheAwareScheduler:

                is_cached_user = bool(candidate.is_cached)
                can_use, snapshot = await self._check_concurrent_available(
-                    endpoint,
                    key,
                    is_cached_user=is_cached_user,
                )
@@ -312,47 +306,51 @@ class CacheAwareScheduler:

            provider_offset += provider_batch_size

-        raise ProviderNotAvailableException(f"所有Provider的资源当前不可用 (model={model_name})")
+        raise ProviderNotAvailableException("服务暂时繁忙，请稍后重试")

-    def _get_effective_concurrent_limit(self, key: ProviderAPIKey) -> Optional[int]:
+    def _get_effective_rpm_limit(self, key: ProviderAPIKey) -> Optional[int]:
        """
-        获取有效的并发限制
+        获取有效的 RPM 限制

        新逻辑：
-        - max_concurrent=NULL: 启用自适应，使用 learned_max_concurrent（如无学习记录则为 None）
-        - max_concurrent=数字: 固定限制，直接使用该值
+        - rpm_limit=NULL: 启用自适应，使用 learned_rpm_limit（如无学习记录则使用默认初始值）
+        - rpm_limit=数字: 固定限制，直接使用该值

        Args:
            key: API Key对象

        Returns:
-            有效的并发限制（None 表示不限制）
+            有效的 RPM 限制（None 表示不限制）
        """
-        if key.max_concurrent is None:
+        if key.rpm_limit is None:
            # 自适应模式：使用学习到的值
-            learned = key.learned_max_concurrent
-            return int(learned) if learned is not None else None
+            learned = key.learned_rpm_limit
+            if learned is not None:
+                return int(learned)
+
+            # 未学习到值时，使用默认初始限制，避免无限制打爆上游
+            from src.config.constants import RPMDefaults
+
+            return int(RPMDefaults.INITIAL_LIMIT)
        else:
            # 固定限制模式
-            return int(key.max_concurrent)
+            return int(key.rpm_limit)

    async def _check_concurrent_available(
        self,
-        endpoint: ProviderEndpoint,
        key: ProviderAPIKey,
        is_cached_user: bool = False,
    ) -> Tuple[bool, ConcurrencySnapshot]:
        """
-        检查并发是否可用（使用动态预留机制）
+        检查 RPM 限制是否可用（使用动态预留机制）

        核心逻辑 - 动态缓存预留机制:
-        - 总槽位: 有效并发限制（固定值或学习到的值）
+        - 总槽位: 有效 RPM 限制（固定值或学习到的值）
        - 预留比例: 由 AdaptiveReservationManager 根据置信度和负载动态计算
        - 缓存用户可用: 全部槽位
        - 新用户可用: 总槽位 × (1 - 动态预留比例)

        Args:
-            endpoint: ProviderEndpoint对象
            key: ProviderAPIKey对象
            is_cached_user: 是否是缓存用户

@@ -360,7 +358,7 @@ class CacheAwareScheduler:
            (是否可用, 并发快照)
        """
        # 获取有效的并发限制
-        effective_key_limit = self._get_effective_concurrent_limit(key)
+        effective_key_limit = self._get_effective_rpm_limit(key)

        logger.debug(
            f"            -> 并发检查: _concurrency_manager={self._concurrency_manager is not None}, "
@@ -371,33 +369,23 @@ class CacheAwareScheduler:
            # 并发管理器不可用，直接返回True
            logger.debug(f"            -> 无并发管理器，直接通过")
            snapshot = ConcurrencySnapshot(
-                endpoint_current=0,
-                endpoint_limit=(
-                    int(endpoint.max_concurrent) if endpoint.max_concurrent is not None else None
-                ),
                key_current=0,
                key_limit=effective_key_limit,
                is_cached_user=is_cached_user,
            )
            return True, snapshot

-        # 获取当前并发数
-        endpoint_count, key_count = await self._concurrency_manager.get_current_concurrency(
-            endpoint_id=str(endpoint.id),
+        # 获取当前 RPM 计数
+        key_count = await self._concurrency_manager.get_key_rpm_count(
            key_id=str(key.id),
        )

        can_use = True

-        # 检查Endpoint级别限制
-        if endpoint.max_concurrent is not None:
-            if endpoint_count >= endpoint.max_concurrent:
-                can_use = False
-
        # 计算动态预留比例
        reservation_result = self._reservation_manager.calculate_reservation(
            key=key,
-            current_concurrent=key_count,
+            current_usage=key_count,
            effective_limit=effective_key_limit,
        )

@@ -440,7 +428,8 @@ class CacheAwareScheduler:
                # 使用 max 确保至少有 1 个槽位可用
                import math

-                available_for_new = max(1, math.ceil(effective_key_limit * (1 - reservation_ratio)))
+                # 与 ConcurrencyManager 的 Lua 脚本保持一致：使用 floor 计算新用户可用槽位
+                available_for_new = max(1, math.floor(effective_key_limit * (1 - reservation_ratio)))
                if key_count >= available_for_new:
                    logger.debug(
                        f"Key {key.id[:8]}... 新用户配额已满 "
@@ -460,8 +449,6 @@ class CacheAwareScheduler:
            key_limit_for_snapshot = None

        snapshot = ConcurrencySnapshot(
-            endpoint_current=endpoint_count,
-            endpoint_limit=endpoint.max_concurrent,
            key_current=key_count,
            key_limit=key_limit_for_snapshot,
            is_cached_user=is_cached_user,
@@ -475,7 +462,7 @@ class CacheAwareScheduler:
    def _get_effective_restrictions(
        self,
        user_api_key: Optional[ApiKey],
-    ) -> Dict[str, Optional[set]]:
+    ) -> Dict[str, Any]:
        """
        获取有效的访问限制（合并 ApiKey 和 User 的限制）

@@ -536,7 +523,10 @@ class CacheAwareScheduler:
        )

        # 合并 allowed_models
-        result["allowed_models"] = merge_restrictions(
+        # allowed_models 支持 list/dict 两种结构，不能转成 set 否则会导致权限校验失效
+        from src.core.model_permissions import merge_allowed_models
+
+        result["allowed_models"] = merge_allowed_models(
            user_api_key.allowed_models, user.allowed_models if user else None
        )

@@ -617,22 +607,25 @@ class CacheAwareScheduler:
                )
                return [], global_model_id

-        # 0.2 检查模型是否被允许
-        if allowed_models is not None:
-            if (
-                requested_model_name not in allowed_models
-                and resolved_model_name not in allowed_models
-            ):
-                resolved_note = (
-                    f" (解析为 {resolved_model_name})"
-                    if resolved_model_name != requested_model_name
-                    else ""
-                )
-                logger.debug(
-                    f"用户/API Key 不允许使用模型 {requested_model_name}{resolved_note}, "
-                    f"允许的模型: {allowed_models}"
-                )
-                return [], global_model_id
+        # 0.2 检查模型是否被允许（支持简单列表和按格式字典两种模式）
+        from src.core.model_permissions import check_model_allowed, get_allowed_models_preview
+
+        if not check_model_allowed(
+            model_name=requested_model_name,
+            allowed_models=allowed_models,
+            api_format=target_format.value,
+            resolved_model_name=resolved_model_name,
+        ):
+            resolved_note = (
+                f" (解析为 {resolved_model_name})"
+                if resolved_model_name != requested_model_name
+                else ""
+            )
+            logger.debug(
+                f"用户/API Key 不允许使用模型 {requested_model_name}{resolved_note}, "
+                f"允许的模型: {get_allowed_models_preview(allowed_models)}"
+            )
+            return [], global_model_id

        # 1. 查询 Providers
        providers = self._query_providers(
@@ -724,8 +717,11 @@ class CacheAwareScheduler:
        provider_query = (
            db.query(Provider)
            .options(
-                selectinload(Provider.endpoints).selectinload(ProviderEndpoint.api_keys),
-                # 同时加载 models 和 global_model 关系，以便 get_effective_* 方法能正确继承默认值
+                # 预加载 Provider 级别的 api_keys
+                selectinload(Provider.api_keys),
+                # 预加载 endpoints（用于按 api_format 选择请求配置）
+                selectinload(Provider.endpoints),
+                # 同时加载 models 和 global_model 关系
                selectinload(Provider.models).selectinload(Model.global_model),
            )
            .filter(Provider.is_active == True)
@@ -852,6 +848,7 @@ class CacheAwareScheduler:
    def _check_key_availability(
        self,
        key: ProviderAPIKey,
+        api_format: Optional[str],
        model_name: str,
        capability_requirements: Optional[Dict[str, bool]] = None,
        resolved_model_name: Optional[str] = None,
@@ -871,20 +868,24 @@ class CacheAwareScheduler:
        Returns:
            (is_available, skip_reason)
        """
-        # 检查熔断器状态（使用详细状态方法获取更丰富的跳过原因）
-        is_available, circuit_reason = health_monitor.get_circuit_breaker_status(key)
+        # 检查熔断器状态（使用详细状态方法获取更丰富的跳过原因，按 API 格式）
+        is_available, circuit_reason = health_monitor.get_circuit_breaker_status(
+            key, api_format=api_format
+        )
        if not is_available:
            return False, circuit_reason or "熔断器已打开"

-        # 模型权限检查：使用 allowed_models 白名单
+        # 模型权限检查：使用 allowed_models 白名单（支持简单列表和按格式字典两种模式）
        # None = 允许所有模型，[] = 拒绝所有模型，["a","b"] = 只允许指定模型
-        if key.allowed_models is not None and (
-            model_name not in key.allowed_models
-            and (not resolved_model_name or resolved_model_name not in key.allowed_models)
+        from src.core.model_permissions import check_model_allowed, get_allowed_models_preview
+
+        if not check_model_allowed(
+            model_name=model_name,
+            allowed_models=key.allowed_models,
+            api_format=api_format,
+            resolved_model_name=resolved_model_name,
        ):
-            allowed_preview = ", ".join(key.allowed_models[:3]) if key.allowed_models else "(无)"
-            suffix = "..." if len(key.allowed_models) > 3 else ""
-            return False, f"模型权限不匹配(允许: {allowed_preview}{suffix})"
+            return False, f"模型权限不匹配(允许: {get_allowed_models_preview(key.allowed_models)})"

        # Key 级别的能力匹配检查
        # 注意：模型级别的能力检查已在 _check_model_support 中完成
@@ -914,6 +915,8 @@ class CacheAwareScheduler:
        """
        构建候选列表

+        Key 直属 Provider，通过 api_formats 筛选符合目标格式的 Key。
+
        Args:
            db: 数据库会话
            providers: Provider 列表
@@ -929,10 +932,10 @@ class CacheAwareScheduler:
            候选列表
        """
        candidates: List[ProviderCandidate] = []
+        target_format_str = target_format.value

        for provider in providers:
            # 检查模型支持（同时检查流式支持和模型能力需求）
-            # 模型能力检查在 Provider 级别进行，如果模型不支持所需能力，整个 Provider 被跳过
            supports_model, skip_reason, _model_caps = await self._check_model_support(
                db, provider, model_name, is_stream, capability_requirements
            )
@@ -940,49 +943,63 @@ class CacheAwareScheduler:
                logger.debug(f"Provider {provider.name} 不支持模型 {model_name}: {skip_reason}")
                continue

+            # 查找目标格式对应的 Endpoint（获取请求配置）
+            target_endpoint = None
            for endpoint in provider.endpoints:
-                # endpoint.api_format 是字符串，target_format 是枚举
                endpoint_format_str = (
                    endpoint.api_format
                    if isinstance(endpoint.api_format, str)
                    else endpoint.api_format.value
                )
-                if not endpoint.is_active or endpoint_format_str != target_format.value:
-                    continue
+                if endpoint.is_active and endpoint_format_str == target_format_str:
+                    target_endpoint = endpoint
+                    break

-                # 获取活跃的 Key 并按 internal_priority + 负载均衡排序
-                active_keys = [key for key in endpoint.api_keys if key.is_active]
-                # 检查是否所有 Key 都是 TTL=0（轮换模式）
-                # 如果所有 Key 的 cache_ttl_minutes 都是 0 或 None，则使用随机排序
-                use_random = all(
-                    (key.cache_ttl_minutes or 0) == 0 for key in active_keys
-                ) if active_keys else False
-                if use_random and len(active_keys) > 1:
-                    logger.debug(
-                        f"  Endpoint {endpoint.id[:8]}... 启用 Key 轮换模式 (TTL=0, {len(active_keys)} keys)"
-                    )
-                keys = self._shuffle_keys_by_internal_priority(active_keys, affinity_key, use_random)
+            if not target_endpoint:
+                logger.debug(f"Provider {provider.name} 没有活跃的 {target_format_str} 端点")
+                continue

-                for key in keys:
-                    # Key 级别的能力检查（模型级别的能力检查已在上面完成）
-                    is_available, skip_reason = self._check_key_availability(
-                        key,
-                        model_name,
-                        capability_requirements,
-                        resolved_model_name=resolved_model_name,
-                    )
+            # Key 直属 Provider，通过 api_formats 筛选
+            active_keys = [
+                key for key in provider.api_keys
+                if key.is_active and target_format_str in (key.api_formats or [])
+            ]

-                    candidate = ProviderCandidate(
-                        provider=provider,
-                        endpoint=endpoint,
-                        key=key,
-                        is_skipped=not is_available,
-                        skip_reason=skip_reason,
-                    )
-                    candidates.append(candidate)
+            if not active_keys:
+                logger.debug(f"Provider {provider.name} 没有支持 {target_format_str} 的活跃 Key")
+                continue

-                    if max_candidates and len(candidates) >= max_candidates:
-                        return candidates
+            # 检查是否所有 Key 都是 TTL=0（轮换模式）
+            use_random = all(
+                (key.cache_ttl_minutes or 0) == 0 for key in active_keys
+            ) if active_keys else False
+            if use_random and len(active_keys) > 1:
+                logger.debug(
+                    f"  Provider {provider.name} 启用 Key 轮换模式 (TTL=0, {len(active_keys)} keys)"
+                )
+            keys = self._shuffle_keys_by_internal_priority(active_keys, affinity_key, use_random)
+
+            for key in keys:
+                # Key 级别的能力检查
+                is_available, skip_reason = self._check_key_availability(
+                    key,
+                    target_format_str,
+                    model_name,
+                    capability_requirements,
+                    resolved_model_name=resolved_model_name,
+                )
+
+                candidate = ProviderCandidate(
+                    provider=provider,
+                    endpoint=target_endpoint,
+                    key=key,
+                    is_skipped=not is_available,
+                    skip_reason=skip_reason,
+                )
+                candidates.append(candidate)
+
+                if max_candidates and len(candidates) >= max_candidates:
+                    return candidates

        return candidates

@@ -1187,7 +1204,6 @@ class CacheAwareScheduler:

        from collections import defaultdict

-        # 使用 tuple 作为统一的 key 类型，兼容两种模式
        priority_groups: Dict[tuple, List[ProviderCandidate]] = defaultdict(list)

        # 根据优先级模式选择分组方式
--- a/src/services/cache/provider_cache.py
+++ b/src/services/cache/provider_cache.py
@@ -27,24 +27,29 @@ class ProviderCacheService:

    @staticmethod
    async def get_provider_api_key_rate_multiplier(
-        db: Session, provider_api_key_id: str
+        db: Session, provider_api_key_id: str, api_format: Optional[str] = None
    ) -> Optional[float]:
        """
        获取 ProviderAPIKey 的 rate_multiplier（带缓存）

+        优先返回指定 API 格式的倍率，如果没有则返回默认倍率。
+
        Args:
            db: 数据库会话
            provider_api_key_id: ProviderAPIKey ID
+            api_format: API 格式（可选），如 "CLAUDE"、"OPENAI"

        Returns:
            rate_multiplier 或 None（如果找不到）
        """
-        cache_key = f"provider_api_key:rate_multiplier:{provider_api_key_id}"
+        # 缓存键包含 api_format
+        format_suffix = api_format.upper() if api_format else "default"
+        cache_key = f"provider_api_key:rate_multiplier:{provider_api_key_id}:{format_suffix}"

        # 1. 尝试从缓存获取
        cached_data = await CacheService.get(cache_key)
        if cached_data is not None:
-            logger.debug(f"ProviderAPIKey rate_multiplier 缓存命中: {provider_api_key_id[:8]}...")
+            logger.debug(f"ProviderAPIKey rate_multiplier 缓存命中: {provider_api_key_id[:8]}... format={format_suffix}")
            # 缓存的 "NOT_FOUND" 表示数据库中不存在
            if cached_data == "NOT_FOUND":
                return None
@@ -52,18 +57,24 @@ class ProviderCacheService:

        # 2. 缓存未命中，查询数据库
        provider_key = (
-            db.query(ProviderAPIKey.rate_multiplier)
+            db.query(ProviderAPIKey.rate_multiplier, ProviderAPIKey.rate_multipliers)
            .filter(ProviderAPIKey.id == provider_api_key_id)
            .first()
        )

-        # 3. 写入缓存
+        # 3. 计算倍率并写入缓存
        if provider_key:
+            # 优先使用 rate_multipliers[api_format]，回退到 rate_multiplier
            rate_multiplier = provider_key.rate_multiplier or 1.0
+            if api_format and provider_key.rate_multipliers:
+                format_upper = api_format.upper()
+                if format_upper in provider_key.rate_multipliers:
+                    rate_multiplier = provider_key.rate_multipliers[format_upper]
+
            await CacheService.set(
                cache_key, rate_multiplier, ttl_seconds=ProviderCacheService.CACHE_TTL
            )
-            logger.debug(f"ProviderAPIKey rate_multiplier 已缓存: {provider_api_key_id[:8]}...")
+            logger.debug(f"ProviderAPIKey rate_multiplier 已缓存: {provider_api_key_id[:8]}... format={format_suffix} value={rate_multiplier}")
            return rate_multiplier
        else:
            # 缓存负结果
@@ -125,6 +136,7 @@ class ProviderCacheService:
        db: Session,
        provider_api_key_id: Optional[str],
        provider_id: Optional[str],
+        api_format: Optional[str] = None,
    ) -> Tuple[float, bool]:
        """
        获取费率倍数和是否免费套餐（带缓存）
@@ -135,6 +147,7 @@ class ProviderCacheService:
            db: 数据库会话
            provider_api_key_id: ProviderAPIKey ID（可选）
            provider_id: Provider ID（可选）
+            api_format: API 格式（可选），用于获取按格式配置的倍率

        Returns:
            (rate_multiplier, is_free_tier) 元组
@@ -142,10 +155,10 @@ class ProviderCacheService:
        actual_rate_multiplier = 1.0
        is_free_tier = False

-        # 获取费率倍数
+        # 获取费率倍数（支持按 API 格式查询）
        if provider_api_key_id:
            rate_multiplier = await ProviderCacheService.get_provider_api_key_rate_multiplier(
-                db, provider_api_key_id
+                db, provider_api_key_id, api_format
            )
            if rate_multiplier is not None:
                actual_rate_multiplier = rate_multiplier
@@ -160,8 +173,9 @@ class ProviderCacheService:

    @staticmethod
    async def invalidate_provider_api_key_cache(provider_api_key_id: str) -> None:
-        """清除 ProviderAPIKey 缓存"""
-        await CacheService.delete(f"provider_api_key:rate_multiplier:{provider_api_key_id}")
+        """清除 ProviderAPIKey 缓存（包括所有 API 格式的缓存）"""
+        # 使用模式匹配删除所有格式的缓存
+        await CacheService.delete_pattern(f"provider_api_key:rate_multiplier:{provider_api_key_id}:*")
        logger.debug(f"ProviderAPIKey 缓存已清除: {provider_api_key_id[:8]}...")

    @staticmethod
--- a/src/services/health/endpoint.py
+++ b/src/services/health/endpoint.py
@@ -70,20 +70,21 @@ class EndpointHealthService:
            db.query(ProviderEndpoint).join(Provider).filter(Provider.is_active.is_(True)).all()
        )

-        # 收集所有 endpoint_ids
-        all_endpoint_ids = [ep.id for ep in endpoints]
+        # 收集所有 provider_ids
+        all_provider_ids = list(set(ep.provider_id for ep in endpoints))

-        # 批量查询所有密钥
+        # 批量查询所有密钥（通过 provider_id 关联）
        all_keys = (
            db.query(ProviderAPIKey)
-            .filter(ProviderAPIKey.endpoint_id.in_(all_endpoint_ids))
+            .filter(ProviderAPIKey.provider_id.in_(all_provider_ids))
            .all()
-        ) if all_endpoint_ids else []
+        ) if all_provider_ids else []

-        # 按 endpoint_id 分组密钥
-        keys_by_endpoint: Dict[str, List[ProviderAPIKey]] = defaultdict(list)
+        # 按 api_format 分组密钥（通过 api_formats 字段）
+        keys_by_format: Dict[str, List[ProviderAPIKey]] = defaultdict(list)
        for key in all_keys:
-            keys_by_endpoint[key.endpoint_id].append(key)
+            for fmt in (key.api_formats or []):
+                keys_by_format[fmt].append(key)

        # 按 API 格式聚合
        format_stats = defaultdict(
@@ -106,18 +107,36 @@ class EndpointHealthService:
            format_stats[api_format]["endpoint_ids"].append(ep.id)
            format_stats[api_format]["provider_ids"].add(ep.provider_id)

-            # 从预加载的密钥中获取
-            keys = keys_by_endpoint.get(ep.id, [])
-            format_stats[api_format]["total_keys"] += len(keys)
+        # 统计每个格式的密钥（直接从 keys_by_format 获取）
+        for api_format, keys in keys_by_format.items():
+            if api_format not in format_stats:
+                # 如果有 Key 但没有对应的 Endpoint，跳过
+                continue

-            # 统计活跃密钥和健康度
-            if ep.is_active:
-                for key in keys:
-                    format_stats[api_format]["key_ids"].append(key.id)
-                    if key.is_active and not key.circuit_breaker_open:
-                        format_stats[api_format]["active_keys"] += 1
-                        health_score = key.health_score if key.health_score is not None else 1.0
-                        format_stats[api_format]["health_scores"].append(health_score)
+            # 去重（同一个 Key 可能支持多个格式）
+            seen_key_ids = set()
+            unique_keys = []
+            for key in keys:
+                if key.id not in seen_key_ids:
+                    seen_key_ids.add(key.id)
+                    unique_keys.append(key)
+
+            format_stats[api_format]["total_keys"] = len(unique_keys)
+
+            for key in unique_keys:
+                format_stats[api_format]["key_ids"].append(key.id)
+                # 检查该格式的熔断器状态
+                circuit_by_format = key.circuit_breaker_by_format or {}
+                format_circuit = circuit_by_format.get(api_format, {})
+                is_circuit_open = format_circuit.get("open", False)
+
+                if key.is_active and not is_circuit_open:
+                    format_stats[api_format]["active_keys"] += 1
+                    # 获取该格式的健康度
+                    health_by_format = key.health_by_format or {}
+                    format_health = health_by_format.get(api_format, {})
+                    health_score = float(format_health.get("health_score") or 1.0)
+                    format_stats[api_format]["health_scores"].append(health_score)

        # 批量生成所有格式的时间线数据
        all_key_ids = []
@@ -372,7 +391,7 @@ class EndpointHealthService:
        segments: int = 100,
    ) -> Dict[str, Any]:
        """
-        从真实使用记录生成时间线数据（兼容旧接口，使用批量查询优化）
+        从真实使用记录生成时间线数据（使用批量查询优化）

        Args:
            db: 数据库会话
@@ -391,13 +410,34 @@ class EndpointHealthService:
                "time_range_end": None,
            }

-        # 先查询该 API 格式下的所有密钥
-        key_ids = [
-            k.id
-            for k in db.query(ProviderAPIKey.id)
-            .filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids))
+        # 基于 endpoint_ids 反推 provider_ids 与 api_format，再选出支持该格式的 keys
+        endpoint_rows = (
+            db.query(ProviderEndpoint.provider_id, ProviderEndpoint.api_format)
+            .filter(ProviderEndpoint.id.in_(endpoint_ids))
            .all()
-        ]
+        )
+
+        if not endpoint_rows:
+            return {
+                "timeline": ["unknown"] * 100,
+                "time_range_start": None,
+                "time_range_end": None,
+            }
+
+        provider_ids = {str(pid) for pid, _fmt in endpoint_rows}
+        # 同一调用中 endpoint_ids 来自同一 api_format（上层已按格式分组）
+        api_format = (
+            endpoint_rows[0][1].value
+            if hasattr(endpoint_rows[0][1], "value")
+            else str(endpoint_rows[0][1])
+        )
+
+        keys = (
+            db.query(ProviderAPIKey.id, ProviderAPIKey.api_formats)
+            .filter(ProviderAPIKey.provider_id.in_(provider_ids))
+            .all()
+        )
+        key_ids = [str(key_id) for key_id, formats in keys if api_format in (formats or [])]

        if not key_ids:
            return {
--- a/src/services/health/monitor.py
+++ b/src/services/health/monitor.py
@@ -1,11 +1,15 @@
 """
-健康监控器 - Endpoint 和 Key 的健康度追踪
+健康监控器 - Endpoint 和 Key 的健康度追踪（按 API 格式区分）

 功能：
-1. 基于滑动窗口的错误率计算
-2. 三态熔断器：关闭 -> 打开 -> 半开 -> 关闭
+1. 基于滑动窗口的错误率计算（按 API 格式独立）
+2. 三态熔断器：关闭 -> 打开 -> 半开 -> 关闭（按 API 格式独立）
 3. 半开状态允许少量请求验证服务恢复
 4. 提供健康度查询和管理 API
+
+数据结构：
+- health_by_format: {"CLAUDE": {"health_score": 1.0, "consecutive_failures": 0, ...}, ...}
+- circuit_breaker_by_format: {"CLAUDE": {"open": false, "open_at": null, ...}, ...}
 """

 import os
@@ -30,8 +34,30 @@ class CircuitState:
    HALF_OPEN = "half_open"  # 半开（验证恢复）


+# 默认健康度数据结构
+def _default_health_data() -> Dict[str, Any]:
+    return {
+        "health_score": 1.0,
+        "consecutive_failures": 0,
+        "last_failure_at": None,
+        "request_results_window": [],
+    }
+
+
+# 默认熔断器数据结构
+def _default_circuit_data() -> Dict[str, Any]:
+    return {
+        "open": False,
+        "open_at": None,
+        "next_probe_at": None,
+        "half_open_until": None,
+        "half_open_successes": 0,
+        "half_open_failures": 0,
+    }
+
+
 class HealthMonitor:
-    """健康监控器（滑动窗口 + 半开状态模式）"""
+    """健康监控器（滑动窗口 + 半开状态模式，按 API 格式区分）"""

    # === 滑动窗口配置 ===
    WINDOW_SIZE = int(os.getenv("HEALTH_WINDOW_SIZE", str(CircuitBreakerDefaults.WINDOW_SIZE)))
@@ -96,6 +122,38 @@ class HealthMonitor:
    _circuit_history: List[Dict[str, Any]] = []
    _open_circuit_keys: int = 0

+    # ==================== 数据访问辅助方法 ====================
+
+    @classmethod
+    def _get_health_data(cls, key: ProviderAPIKey, api_format: str) -> Dict[str, Any]:
+        """获取指定格式的健康度数据，不存在则返回默认值"""
+        health_by_format = key.health_by_format or {}
+        if api_format not in health_by_format:
+            return _default_health_data()
+        return health_by_format[api_format]
+
+    @classmethod
+    def _set_health_data(cls, key: ProviderAPIKey, api_format: str, data: Dict[str, Any]) -> None:
+        """设置指定格式的健康度数据"""
+        health_by_format = dict(key.health_by_format or {})
+        health_by_format[api_format] = data
+        key.health_by_format = health_by_format  # type: ignore[assignment]
+
+    @classmethod
+    def _get_circuit_data(cls, key: ProviderAPIKey, api_format: str) -> Dict[str, Any]:
+        """获取指定格式的熔断器数据，不存在则返回默认值"""
+        circuit_by_format = key.circuit_breaker_by_format or {}
+        if api_format not in circuit_by_format:
+            return _default_circuit_data()
+        return circuit_by_format[api_format]
+
+    @classmethod
+    def _set_circuit_data(cls, key: ProviderAPIKey, api_format: str, data: Dict[str, Any]) -> None:
+        """设置指定格式的熔断器数据"""
+        circuit_by_format = dict(key.circuit_breaker_by_format or {})
+        circuit_by_format[api_format] = data
+        key.circuit_breaker_by_format = circuit_by_format  # type: ignore[assignment]
+
    # ==================== 核心方法 ====================

    @classmethod
@@ -103,9 +161,21 @@ class HealthMonitor:
        cls,
        db: Session,
        key_id: Optional[str] = None,
+        api_format: Optional[str] = None,
        response_time_ms: Optional[int] = None,
    ) -> None:
-        """记录成功请求"""
+        """记录成功请求（按 API 格式）
+
+        Args:
+            db: 数据库会话
+            key_id: Key ID（必需）
+            api_format: API 格式（必需，用于区分不同格式的健康度）
+            response_time_ms: 响应时间（可选）
+
+        Note:
+            api_format 在逻辑上是必需的，但为了向后兼容保持 Optional 签名。
+            如果未提供，会尝试从 Key 的 api_formats 中获取第一个格式作为 fallback。
+        """
        try:
            if not key_id:
                return
@@ -114,39 +184,96 @@ class HealthMonitor:
            if not key:
                return

+            # api_format 兼容处理：如果未提供，尝试使用 Key 的第一个格式
+            effective_api_format = api_format
+            if not effective_api_format:
+                if key.api_formats and len(key.api_formats) > 0:
+                    effective_api_format = key.api_formats[0]
+                    logger.debug(
+                        f"record_success: api_format 未提供，使用默认格式 {effective_api_format}"
+                    )
+                else:
+                    logger.warning(
+                        f"record_success: api_format 未提供且 Key 无可用格式: key_id={key_id[:8]}..."
+                    )
+                    return
+
            now = datetime.now(timezone.utc)
            now_ts = now.timestamp()

+            # 获取当前格式的健康度数据
+            health_data = cls._get_health_data(key, effective_api_format)
+            circuit_data = cls._get_circuit_data(key, effective_api_format)
+
            # 1. 更新滑动窗口
-            cls._add_to_window(key, now_ts, success=True)
+            window = health_data.get("request_results_window") or []
+            window.append({"ts": now_ts, "ok": True})
+            cutoff_ts = now_ts - cls.WINDOW_SECONDS
+            window = [r for r in window if r["ts"] > cutoff_ts]
+            if len(window) > cls.WINDOW_SIZE:
+                window = window[-cls.WINDOW_SIZE :]
+            health_data["request_results_window"] = window

            # 2. 更新健康度（用于展示）
-            new_score = min(float(key.health_score or 0) + cls.SUCCESS_INCREMENT, 1.0)
-            key.health_score = new_score  # type: ignore[assignment]
+            current_score = float(health_data.get("health_score") or 0)
+            new_score = min(current_score + cls.SUCCESS_INCREMENT, 1.0)
+            health_data["health_score"] = new_score

            # 3. 更新统计
-            key.consecutive_failures = 0  # type: ignore[assignment]
-            key.last_failure_at = None  # type: ignore[assignment]
+            health_data["consecutive_failures"] = 0
+            health_data["last_failure_at"] = None
+
+            # 4. 处理熔断器状态
+            state = cls._get_circuit_state_from_data(circuit_data, now)
+
+            if state == CircuitState.HALF_OPEN:
+                # 半开状态：记录成功
+                circuit_data["half_open_successes"] = int(
+                    circuit_data.get("half_open_successes") or 0
+                ) + 1
+
+                if circuit_data["half_open_successes"] >= cls.HALF_OPEN_SUCCESS_THRESHOLD:
+                    # 达到成功阈值，关闭熔断器
+                    cls._close_circuit_data(circuit_data, health_data, reason="半开状态验证成功")
+                    cls._push_circuit_event(
+                        {
+                            "event": "closed",
+                            "key_id": key.id,
+                            "api_format": effective_api_format,
+                            "reason": "半开状态验证成功",
+                            "timestamp": now.isoformat(),
+                        }
+                    )
+                    logger.info(
+                        f"[CLOSED] Key 熔断器关闭: {key.id[:8]}.../{effective_api_format} | 原因: 半开状态验证成功"
+                    )
+
+            elif state == CircuitState.OPEN:
+                # 打开状态下的成功（探测成功），进入半开状态
+                cls._enter_half_open_data(circuit_data, now)
+                cls._push_circuit_event(
+                    {
+                        "event": "half_open",
+                        "key_id": key.id,
+                        "api_format": effective_api_format,
+                        "timestamp": now.isoformat(),
+                    }
+                )
+                logger.info(
+                    f"[HALF-OPEN] Key 进入半开状态: {key.id[:8]}.../{effective_api_format} | "
+                    f"需要 {cls.HALF_OPEN_SUCCESS_THRESHOLD} 次成功关闭熔断器"
+                )
+
+            # 保存数据
+            cls._set_health_data(key, effective_api_format, health_data)
+            cls._set_circuit_data(key, effective_api_format, circuit_data)
+
+            # 更新全局统计
            key.success_count = int(key.success_count or 0) + 1  # type: ignore[assignment]
            key.request_count = int(key.request_count or 0) + 1  # type: ignore[assignment]
            if response_time_ms:
                key.total_response_time_ms = int(key.total_response_time_ms or 0) + response_time_ms  # type: ignore[assignment]

-            # 4. 处理熔断器状态
-            state = cls._get_circuit_state(key, now)
-
-            if state == CircuitState.HALF_OPEN:
-                # 半开状态：记录成功
-                key.half_open_successes = int(key.half_open_successes or 0) + 1  # type: ignore[assignment]
-
-                if int(key.half_open_successes or 0) >= cls.HALF_OPEN_SUCCESS_THRESHOLD:
-                    # 达到成功阈值，关闭熔断器
-                    cls._close_circuit(key, now, reason="半开状态验证成功")
-
-            elif state == CircuitState.OPEN:
-                # 打开状态下的成功（探测成功），进入半开状态
-                cls._enter_half_open(key, now)
-
            db.flush()
            get_batch_committer().mark_dirty(db)

@@ -159,9 +286,21 @@ class HealthMonitor:
        cls,
        db: Session,
        key_id: Optional[str] = None,
+        api_format: Optional[str] = None,
        error_type: Optional[str] = None,
    ) -> None:
-        """记录失败请求"""
+        """记录失败请求（按 API 格式）
+
+        Args:
+            db: 数据库会话
+            key_id: Key ID（必需）
+            api_format: API 格式（必需，用于区分不同格式的健康度）
+            error_type: 错误类型（可选）
+
+        Note:
+            api_format 在逻辑上是必需的，但为了向后兼容保持 Optional 签名。
+            如果未提供，会尝试从 Key 的 api_formats 中获取第一个格式作为 fallback。
+        """
        try:
            if not key_id:
                return
@@ -170,46 +309,117 @@ class HealthMonitor:
            if not key:
                return

+            # api_format 兼容处理：如果未提供，尝试使用 Key 的第一个格式
+            effective_api_format = api_format
+            if not effective_api_format:
+                if key.api_formats and len(key.api_formats) > 0:
+                    effective_api_format = key.api_formats[0]
+                    logger.debug(
+                        f"record_failure: api_format 未提供，使用默认格式 {effective_api_format}"
+                    )
+                else:
+                    logger.warning(
+                        f"record_failure: api_format 未提供且 Key 无可用格式: key_id={key_id[:8]}..."
+                    )
+                    return
+
            now = datetime.now(timezone.utc)
            now_ts = now.timestamp()

+            # 获取当前格式的健康度数据
+            health_data = cls._get_health_data(key, effective_api_format)
+            circuit_data = cls._get_circuit_data(key, effective_api_format)
+
            # 1. 更新滑动窗口
-            cls._add_to_window(key, now_ts, success=False)
+            window = health_data.get("request_results_window") or []
+            window.append({"ts": now_ts, "ok": False})
+            cutoff_ts = now_ts - cls.WINDOW_SECONDS
+            window = [r for r in window if r["ts"] > cutoff_ts]
+            if len(window) > cls.WINDOW_SIZE:
+                window = window[-cls.WINDOW_SIZE :]
+            health_data["request_results_window"] = window

            # 2. 更新健康度（用于展示）
-            new_score = max(float(key.health_score or 1) - cls.FAILURE_DECREMENT, 0.0)
-            key.health_score = new_score  # type: ignore[assignment]
+            current_score = float(health_data.get("health_score") or 1)
+            new_score = max(current_score - cls.FAILURE_DECREMENT, 0.0)
+            health_data["health_score"] = new_score

            # 3. 更新统计
-            key.consecutive_failures = int(key.consecutive_failures or 0) + 1  # type: ignore[assignment]
-            key.last_failure_at = now  # type: ignore[assignment]
-            key.error_count = int(key.error_count or 0) + 1  # type: ignore[assignment]
-            key.request_count = int(key.request_count or 0) + 1  # type: ignore[assignment]
+            health_data["consecutive_failures"] = (
+                int(health_data.get("consecutive_failures") or 0) + 1
+            )
+            health_data["last_failure_at"] = now.isoformat()

            # 4. 处理熔断器状态
-            state = cls._get_circuit_state(key, now)
+            state = cls._get_circuit_state_from_data(circuit_data, now)

            if state == CircuitState.HALF_OPEN:
                # 半开状态：记录失败
-                key.half_open_failures = int(key.half_open_failures or 0) + 1  # type: ignore[assignment]
+                circuit_data["half_open_failures"] = int(
+                    circuit_data.get("half_open_failures") or 0
+                ) + 1

-                if int(key.half_open_failures or 0) >= cls.HALF_OPEN_FAILURE_THRESHOLD:
+                if circuit_data["half_open_failures"] >= cls.HALF_OPEN_FAILURE_THRESHOLD:
                    # 达到失败阈值，重新打开熔断器
-                    cls._open_circuit(key, now, reason="半开状态验证失败")
+                    # 注意：半开状态本身就是打开状态的子状态，不需要增加计数
+                    consecutive = int(health_data.get("consecutive_failures") or 0)
+                    recovery_seconds = cls._calculate_recovery_seconds(consecutive)
+                    cls._open_circuit_data(
+                        circuit_data, now, recovery_seconds, reason="半开状态验证失败"
+                    )
+                    cls._push_circuit_event(
+                        {
+                            "event": "opened",
+                            "key_id": key.id,
+                            "api_format": effective_api_format,
+                            "reason": "半开状态验证失败",
+                            "recovery_seconds": recovery_seconds,
+                            "timestamp": now.isoformat(),
+                        }
+                    )
+                    logger.warning(
+                        f"[OPEN] Key 熔断器打开: {key.id[:8]}.../{effective_api_format} | 原因: 半开状态验证失败 | "
+                        f"{recovery_seconds}秒后进入半开状态"
+                    )

            elif state == CircuitState.CLOSED:
                # 关闭状态：检查是否需要打开熔断器
-                error_rate = cls._calculate_error_rate(key, now_ts)
-                window = key.request_results_window or []
+                error_rate = cls._calculate_error_rate_from_window(window, now_ts)

                if len(window) >= cls.MIN_REQUESTS and error_rate >= cls.ERROR_RATE_THRESHOLD:
-                    cls._open_circuit(
-                        key, now, reason=f"错误率 {error_rate:.0%} 超过阈值 {cls.ERROR_RATE_THRESHOLD:.0%}"
+                    consecutive = int(health_data.get("consecutive_failures") or 0)
+                    recovery_seconds = cls._calculate_recovery_seconds(consecutive)
+                    reason = f"错误率 {error_rate:.0%} 超过阈值 {cls.ERROR_RATE_THRESHOLD:.0%}"
+                    cls._open_circuit_data(circuit_data, now, recovery_seconds, reason=reason)
+                    cls._open_circuit_keys += 1
+                    health_open_circuits.set(cls._open_circuit_keys)
+                    cls._push_circuit_event(
+                        {
+                            "event": "opened",
+                            "key_id": key.id,
+                            "api_format": effective_api_format,
+                            "reason": reason,
+                            "recovery_seconds": recovery_seconds,
+                            "timestamp": now.isoformat(),
+                        }
+                    )
+                    logger.warning(
+                        f"[OPEN] Key 熔断器打开: {key.id[:8]}.../{effective_api_format} | 原因: {reason} | "
+                        f"{recovery_seconds}秒后进入半开状态"
                    )

+            # 保存数据
+            cls._set_health_data(key, effective_api_format, health_data)
+            cls._set_circuit_data(key, effective_api_format, circuit_data)
+
+            # 更新全局统计
+            key.error_count = int(key.error_count or 0) + 1  # type: ignore[assignment]
+            key.request_count = int(key.request_count or 0) + 1  # type: ignore[assignment]
+            key.last_error_at = now  # type: ignore[assignment]
+
            logger.debug(
-                f"[WARN] Key 健康度下降: {key_id[:8]}... -> {new_score:.2f} "
-                f"(连续失败 {key.consecutive_failures} 次, error_type={error_type})"
+                f"[WARN] Key 健康度下降: {key_id[:8]}.../{effective_api_format} -> {new_score:.2f} "
+                f"(连续失败 {health_data['consecutive_failures']} 次, error_type={error_type})"
            )

            db.flush()
@@ -222,31 +432,13 @@ class HealthMonitor:
    # ==================== 滑动窗口方法 ====================

    @classmethod
-    def _add_to_window(cls, key: ProviderAPIKey, now_ts: float, success: bool) -> None:
-        """添加请求结果到滑动窗口"""
-        window: List[Dict[str, Any]] = key.request_results_window or []
-
-        # 添加新记录
-        window.append({"ts": now_ts, "ok": success})
-
-        # 清理过期记录
-        cutoff_ts = now_ts - cls.WINDOW_SECONDS
-        window = [r for r in window if r["ts"] > cutoff_ts]
-
-        # 限制窗口大小
-        if len(window) > cls.WINDOW_SIZE:
-            window = window[-cls.WINDOW_SIZE :]
-
-        key.request_results_window = window  # type: ignore[assignment]
-
-    @classmethod
-    def _calculate_error_rate(cls, key: ProviderAPIKey, now_ts: float) -> float:
-        """计算滑动窗口内的错误率"""
-        window: List[Dict[str, Any]] = key.request_results_window or []
+    def _calculate_error_rate_from_window(
+        cls, window: List[Dict[str, Any]], now_ts: float
+    ) -> float:
+        """从窗口数据计算错误率"""
        if not window:
            return 0.0

-        # 过滤过期记录
        cutoff_ts = now_ts - cls.WINDOW_SECONDS
        valid_records = [r for r in window if r["ts"] > cutoff_ts]

@@ -256,157 +448,158 @@ class HealthMonitor:
        failures = sum(1 for r in valid_records if not r["ok"])
        return failures / len(valid_records)

-    # ==================== 熔断器状态方法 ====================
+    # ==================== 熔断器状态方法（操作数据字典）====================

    @classmethod
-    def _get_circuit_state(cls, key: ProviderAPIKey, now: datetime) -> str:
-        """获取当前熔断器状态"""
-        if not key.circuit_breaker_open:
+    def _get_circuit_state_from_data(cls, circuit_data: Dict[str, Any], now: datetime) -> str:
+        """从数据字典获取当前熔断器状态"""
+        if not circuit_data.get("open"):
            return CircuitState.CLOSED

        # 检查是否在半开状态
-        if key.half_open_until and now < key.half_open_until:
-            return CircuitState.HALF_OPEN
+        half_open_until_str = circuit_data.get("half_open_until")
+        if half_open_until_str:
+            half_open_until = datetime.fromisoformat(half_open_until_str)
+            if now < half_open_until:
+                return CircuitState.HALF_OPEN

        # 检查是否到了探测时间（进入半开）
-        if key.next_probe_at and now >= key.next_probe_at:
-            return CircuitState.HALF_OPEN
+        next_probe_str = circuit_data.get("next_probe_at")
+        if next_probe_str:
+            next_probe_at = datetime.fromisoformat(next_probe_str)
+            if now >= next_probe_at:
+                return CircuitState.HALF_OPEN

        return CircuitState.OPEN

    @classmethod
-    def _open_circuit(cls, key: ProviderAPIKey, now: datetime, reason: str) -> None:
-        """打开熔断器"""
-        was_open = key.circuit_breaker_open
-
-        key.circuit_breaker_open = True  # type: ignore[assignment]
-        key.circuit_breaker_open_at = now  # type: ignore[assignment]
-        key.half_open_until = None  # type: ignore[assignment]
-        key.half_open_successes = 0  # type: ignore[assignment]
-        key.half_open_failures = 0  # type: ignore[assignment]
-
-        # 计算下次探测时间（进入半开状态的时间）
-        consecutive = int(key.consecutive_failures or 0)
-        recovery_seconds = cls._calculate_recovery_seconds(consecutive)
-        key.next_probe_at = now + timedelta(seconds=recovery_seconds)  # type: ignore[assignment]
-
-        if not was_open:
-            cls._open_circuit_keys += 1
-            health_open_circuits.set(cls._open_circuit_keys)
-
-        logger.warning(
-            f"[OPEN] Key 熔断器打开: {key.id[:8]}... | 原因: {reason} | "
-            f"{recovery_seconds}秒后进入半开状态"
-        )
-
-        cls._push_circuit_event(
-            {
-                "event": "opened",
-                "key_id": key.id,
-                "reason": reason,
-                "recovery_seconds": recovery_seconds,
-                "timestamp": now.isoformat(),
-            }
-        )
+    def _open_circuit_data(
+        cls,
+        circuit_data: Dict[str, Any],
+        now: datetime,
+        recovery_seconds: int,
+        reason: str,
+    ) -> None:
+        """打开熔断器（操作数据字典）"""
+        circuit_data["open"] = True
+        circuit_data["open_at"] = now.isoformat()
+        circuit_data["half_open_until"] = None
+        circuit_data["half_open_successes"] = 0
+        circuit_data["half_open_failures"] = 0
+        circuit_data["next_probe_at"] = (now + timedelta(seconds=recovery_seconds)).isoformat()

    @classmethod
-    def _enter_half_open(cls, key: ProviderAPIKey, now: datetime) -> None:
-        """进入半开状态"""
-        key.half_open_until = now + timedelta(seconds=cls.HALF_OPEN_DURATION)  # type: ignore[assignment]
-        key.half_open_successes = 0  # type: ignore[assignment]
-        key.half_open_failures = 0  # type: ignore[assignment]
-
-        logger.info(
-            f"[HALF-OPEN] Key 进入半开状态: {key.id[:8]}... | "
-            f"需要 {cls.HALF_OPEN_SUCCESS_THRESHOLD} 次成功关闭熔断器"
-        )
-
-        cls._push_circuit_event(
-            {
-                "event": "half_open",
-                "key_id": key.id,
-                "timestamp": now.isoformat(),
-            }
-        )
+    def _enter_half_open_data(cls, circuit_data: Dict[str, Any], now: datetime) -> None:
+        """进入半开状态（操作数据字典）"""
+        circuit_data["half_open_until"] = (
+            now + timedelta(seconds=cls.HALF_OPEN_DURATION)
+        ).isoformat()
+        circuit_data["half_open_successes"] = 0
+        circuit_data["half_open_failures"] = 0

    @classmethod
-    def _close_circuit(cls, key: ProviderAPIKey, now: datetime, reason: str) -> None:
-        """关闭熔断器"""
-        key.circuit_breaker_open = False  # type: ignore[assignment]
-        key.circuit_breaker_open_at = None  # type: ignore[assignment]
-        key.next_probe_at = None  # type: ignore[assignment]
-        key.half_open_until = None  # type: ignore[assignment]
-        key.half_open_successes = 0  # type: ignore[assignment]
-        key.half_open_failures = 0  # type: ignore[assignment]
+    def _close_circuit_data(
+        cls, circuit_data: Dict[str, Any], health_data: Dict[str, Any], reason: str
+    ) -> None:
+        """关闭熔断器（操作数据字典）"""
+        circuit_data["open"] = False
+        circuit_data["open_at"] = None
+        circuit_data["next_probe_at"] = None
+        circuit_data["half_open_until"] = None
+        circuit_data["half_open_successes"] = 0
+        circuit_data["half_open_failures"] = 0

        # 快速恢复健康度
-        key.health_score = max(float(key.health_score or 0), cls.PROBE_RECOVERY_SCORE)  # type: ignore[assignment]
+        current_score = float(health_data.get("health_score") or 0)
+        health_data["health_score"] = max(current_score, cls.PROBE_RECOVERY_SCORE)

        cls._open_circuit_keys = max(0, cls._open_circuit_keys - 1)
        health_open_circuits.set(cls._open_circuit_keys)

-        logger.info(f"[CLOSED] Key 熔断器关闭: {key.id[:8]}... | 原因: {reason}")
-
-        cls._push_circuit_event(
-            {
-                "event": "closed",
-                "key_id": key.id,
-                "reason": reason,
-                "timestamp": now.isoformat(),
-            }
-        )
-
    @classmethod
    def _calculate_recovery_seconds(cls, consecutive_failures: int) -> int:
        """计算恢复等待时间（指数退避）"""
-        # 指数退避：30s -> 60s -> 120s -> 240s -> 300s（上限）
-        exponent = min(consecutive_failures // 5, 4)  # 每5次失败增加一级
+        exponent = min(consecutive_failures // 5, 4)
        seconds = cls.INITIAL_RECOVERY_SECONDS * (cls.RECOVERY_BACKOFF**exponent)
        return min(int(seconds), cls.MAX_RECOVERY_SECONDS)

    # ==================== 状态查询方法 ====================

    @classmethod
-    def is_circuit_breaker_closed(cls, resource: ProviderAPIKey) -> bool:
-        """检查熔断器是否允许请求通过"""
-        if not resource.circuit_breaker_open:
+    def is_circuit_breaker_closed(
+        cls, resource: ProviderAPIKey, api_format: Optional[str] = None
+    ) -> bool:
+        """检查熔断器是否允许请求通过（按 API 格式）"""
+        if not api_format:
+            # 兼容旧调用：检查是否有任何格式的熔断器开启
+            circuit_by_format = resource.circuit_breaker_by_format or {}
+            for fmt, circuit_data in circuit_by_format.items():
+                if circuit_data.get("open"):
+                    return False
+            return True
+
+        circuit_data = cls._get_circuit_data(resource, api_format)
+
+        if not circuit_data.get("open"):
            return True

        now = datetime.now(timezone.utc)
-        state = cls._get_circuit_state(resource, now)
+        state = cls._get_circuit_state_from_data(circuit_data, now)

        # 半开状态允许请求通过
        if state == CircuitState.HALF_OPEN:
            return True

        # 检查是否到了探测时间
-        if resource.next_probe_at and now >= resource.next_probe_at:
-            # 自动进入半开状态
-            cls._enter_half_open(resource, now)
-            return True
+        next_probe_str = circuit_data.get("next_probe_at")
+        if next_probe_str:
+            next_probe_at = datetime.fromisoformat(next_probe_str)
+            if now >= next_probe_at:
+                # 自动进入半开状态
+                cls._enter_half_open_data(circuit_data, now)
+                cls._set_circuit_data(resource, api_format, circuit_data)
+                return True

        return False

    @classmethod
    def get_circuit_breaker_status(
-        cls, resource: ProviderAPIKey
+        cls, resource: ProviderAPIKey, api_format: Optional[str] = None
    ) -> Tuple[bool, Optional[str]]:
-        """获取熔断器详细状态"""
-        if not resource.circuit_breaker_open:
+        """获取熔断器详细状态（按 API 格式）"""
+        if not api_format:
+            # 兼容旧调用：返回第一个开启的熔断器状态
+            circuit_by_format = resource.circuit_breaker_by_format or {}
+            for fmt, circuit_data in circuit_by_format.items():
+                if circuit_data.get("open"):
+                    return cls._get_status_from_circuit_data(circuit_data)
+            return True, None
+
+        circuit_data = cls._get_circuit_data(resource, api_format)
+        return cls._get_status_from_circuit_data(circuit_data)
+
+    @classmethod
+    def _get_status_from_circuit_data(
+        cls, circuit_data: Dict[str, Any]
+    ) -> Tuple[bool, Optional[str]]:
+        """从熔断器数据获取状态描述"""
+        if not circuit_data.get("open"):
            return True, None

        now = datetime.now(timezone.utc)
-        state = cls._get_circuit_state(resource, now)
+        state = cls._get_circuit_state_from_data(circuit_data, now)

        if state == CircuitState.HALF_OPEN:
-            successes = int(resource.half_open_successes or 0)
+            successes = int(circuit_data.get("half_open_successes") or 0)
            return True, f"半开状态({successes}/{cls.HALF_OPEN_SUCCESS_THRESHOLD}成功)"

-        if resource.next_probe_at:
-            if now >= resource.next_probe_at:
+        next_probe_str = circuit_data.get("next_probe_at")
+        if next_probe_str:
+            next_probe_at = datetime.fromisoformat(next_probe_str)
+            if now >= next_probe_at:
                return True, None

-            remaining = resource.next_probe_at - now
+            remaining = next_probe_at - now
            remaining_seconds = int(remaining.total_seconds())
            if remaining_seconds >= 60:
                time_str = f"{remaining_seconds // 60}min{remaining_seconds % 60}s"
@@ -417,8 +610,10 @@ class HealthMonitor:
        return False, "熔断中"

    @classmethod
-    def get_key_health(cls, db: Session, key_id: str) -> Optional[Dict[str, Any]]:
-        """获取 Key 健康状态"""
+    def get_key_health(
+        cls, db: Session, key_id: str, api_format: Optional[str] = None
+    ) -> Optional[Dict[str, Any]]:
+        """获取 Key 健康状态（支持按格式查询）"""
        try:
            key = db.query(ProviderAPIKey).filter(ProviderAPIKey.id == key_id).first()
            if not key:
@@ -427,24 +622,15 @@ class HealthMonitor:
            now = datetime.now(timezone.utc)
            now_ts = now.timestamp()

-            # 计算当前错误率
-            error_rate = cls._calculate_error_rate(key, now_ts)
-            window = key.request_results_window or []
-            valid_window = [r for r in window if r["ts"] > now_ts - cls.WINDOW_SECONDS]
-
            avg_response_time_ms = (
                int(key.total_response_time_ms or 0) / int(key.success_count or 1)
                if key.success_count
                else 0
            )

-            return {
+            # 全局统计
+            result = {
                "key_id": key.id,
-                "health_score": float(key.health_score or 1.0),
-                "error_rate": error_rate,
-                "window_size": len(valid_window),
-                "consecutive_failures": int(key.consecutive_failures or 0),
-                "last_failure_at": key.last_failure_at.isoformat() if key.last_failure_at else None,
                "is_active": key.is_active,
                "statistics": {
                    "request_count": int(key.request_count or 0),
@@ -457,25 +643,84 @@ class HealthMonitor:
                    ),
                    "avg_response_time_ms": round(avg_response_time_ms, 2),
                },
-                "circuit_breaker": {
-                    "state": cls._get_circuit_state(key, now),
-                    "open": key.circuit_breaker_open,
-                    "open_at": (
-                        key.circuit_breaker_open_at.isoformat()
-                        if key.circuit_breaker_open_at
-                        else None
-                    ),
-                    "next_probe_at": (
-                        key.next_probe_at.isoformat() if key.next_probe_at else None
-                    ),
-                    "half_open_until": (
-                        key.half_open_until.isoformat() if key.half_open_until else None
-                    ),
-                    "half_open_successes": int(key.half_open_successes or 0),
-                    "half_open_failures": int(key.half_open_failures or 0),
-                },
            }

+            # 按格式的健康度数据
+            health_by_format = key.health_by_format or {}
+            circuit_by_format = key.circuit_breaker_by_format or {}
+
+            if api_format:
+                # 查询单个格式
+                health_data = cls._get_health_data(key, api_format)
+                circuit_data = cls._get_circuit_data(key, api_format)
+                window = health_data.get("request_results_window") or []
+                valid_window = [r for r in window if r["ts"] > now_ts - cls.WINDOW_SECONDS]
+
+                result["api_format"] = api_format
+                result["health_score"] = float(health_data.get("health_score") or 1.0)
+                result["error_rate"] = cls._calculate_error_rate_from_window(window, now_ts)
+                result["window_size"] = len(valid_window)
+                result["consecutive_failures"] = int(
+                    health_data.get("consecutive_failures") or 0
+                )
+                result["last_failure_at"] = health_data.get("last_failure_at")
+                result["circuit_breaker"] = {
+                    "state": cls._get_circuit_state_from_data(circuit_data, now),
+                    "open": circuit_data.get("open", False),
+                    "open_at": circuit_data.get("open_at"),
+                    "next_probe_at": circuit_data.get("next_probe_at"),
+                    "half_open_until": circuit_data.get("half_open_until"),
+                    "half_open_successes": int(circuit_data.get("half_open_successes") or 0),
+                    "half_open_failures": int(circuit_data.get("half_open_failures") or 0),
+                }
+            else:
+                # 返回所有格式的健康度数据
+                formats_health = {}
+                for fmt in (key.api_formats or []):
+                    health_data = health_by_format.get(fmt, _default_health_data())
+                    circuit_data = circuit_by_format.get(fmt, _default_circuit_data())
+                    window = health_data.get("request_results_window") or []
+                    valid_window = [r for r in window if r["ts"] > now_ts - cls.WINDOW_SECONDS]
+
+                    formats_health[fmt] = {
+                        "health_score": float(health_data.get("health_score") or 1.0),
+                        "error_rate": cls._calculate_error_rate_from_window(window, now_ts),
+                        "window_size": len(valid_window),
+                        "consecutive_failures": int(
+                            health_data.get("consecutive_failures") or 0
+                        ),
+                        "last_failure_at": health_data.get("last_failure_at"),
+                        "circuit_breaker": {
+                            "state": cls._get_circuit_state_from_data(circuit_data, now),
+                            "open": circuit_data.get("open", False),
+                            "open_at": circuit_data.get("open_at"),
+                            "next_probe_at": circuit_data.get("next_probe_at"),
+                            "half_open_until": circuit_data.get("half_open_until"),
+                            "half_open_successes": int(
+                                circuit_data.get("half_open_successes") or 0
+                            ),
+                            "half_open_failures": int(
+                                circuit_data.get("half_open_failures") or 0
+                            ),
+                        },
+                    }
+
+                result["health_by_format"] = formats_health
+
+                # 计算整体健康度（取最低值）
+                if formats_health:
+                    result["health_score"] = min(
+                        h["health_score"] for h in formats_health.values()
+                    )
+                    result["any_circuit_open"] = any(
+                        h["circuit_breaker"]["open"] for h in formats_health.values()
+                    )
+                else:
+                    result["health_score"] = 1.0
+                    result["any_circuit_open"] = False
+
+            return result
+
        except Exception as e:
            logger.error(f"获取 Key 健康状态失败: {e}")
            return None
@@ -507,23 +752,24 @@ class HealthMonitor:
    # ==================== 管理方法 ====================

    @classmethod
-    def reset_health(cls, db: Session, key_id: Optional[str] = None) -> bool:
-        """重置健康度"""
+    def reset_health(
+        cls, db: Session, key_id: Optional[str] = None, api_format: Optional[str] = None
+    ) -> bool:
+        """重置健康度（支持按格式重置）"""
        try:
            if key_id:
                key = db.query(ProviderAPIKey).filter(ProviderAPIKey.id == key_id).first()
                if key:
-                    key.health_score = 1.0  # type: ignore[assignment]
-                    key.consecutive_failures = 0  # type: ignore[assignment]
-                    key.last_failure_at = None  # type: ignore[assignment]
-                    key.request_results_window = []  # type: ignore[assignment]
-                    key.circuit_breaker_open = False  # type: ignore[assignment]
-                    key.circuit_breaker_open_at = None  # type: ignore[assignment]
-                    key.next_probe_at = None  # type: ignore[assignment]
-                    key.half_open_until = None  # type: ignore[assignment]
-                    key.half_open_successes = 0  # type: ignore[assignment]
-                    key.half_open_failures = 0  # type: ignore[assignment]
-                    logger.info(f"[RESET] 重置 Key 健康度: {key_id}")
+                    if api_format:
+                        # 重置单个格式
+                        cls._set_health_data(key, api_format, _default_health_data())
+                        cls._set_circuit_data(key, api_format, _default_circuit_data())
+                        logger.info(f"[RESET] 重置 Key 健康度: {key_id}/{api_format}")
+                    else:
+                        # 重置所有格式
+                        key.health_by_format = {}  # type: ignore[assignment]
+                        key.circuit_breaker_by_format = {}  # type: ignore[assignment]
+                        logger.info(f"[RESET] 重置 Key 所有格式健康度: {key_id}")

            db.flush()
            get_batch_committer().mark_dirty(db)
@@ -542,7 +788,9 @@ class HealthMonitor:
                key = db.query(ProviderAPIKey).filter(ProviderAPIKey.id == key_id).first()
                if key and not key.is_active:
                    key.is_active = True  # type: ignore[assignment]
-                    key.consecutive_failures = 0  # type: ignore[assignment]
+                    # 重置所有格式的健康度
+                    key.health_by_format = {}  # type: ignore[assignment]
+                    key.circuit_breaker_by_format = {}  # type: ignore[assignment]
                    logger.info(f"[OK] 手动启用 Key: {key_id}")

            db.flush()
@@ -566,14 +814,28 @@ class HealthMonitor:
                ),
            ).first()

-            key_stats = db.query(
-                func.count(ProviderAPIKey.id).label("total"),
-                func.sum(case((ProviderAPIKey.is_active == True, 1), else_=0)).label("active"),
-                func.sum(case((ProviderAPIKey.health_score < 0.5, 1), else_=0)).label("unhealthy"),
-                func.sum(case((ProviderAPIKey.circuit_breaker_open == True, 1), else_=0)).label(
-                    "circuit_open"
-                ),
-            ).first()
+            # 统计 Key（需要遍历 JSON 字段计算熔断状态）
+            keys = db.query(ProviderAPIKey).all()
+            total_keys = len(keys)
+            active_keys = sum(1 for k in keys if k.is_active)
+            unhealthy_keys = 0
+            circuit_open_keys = 0
+
+            for key in keys:
+                health_by_format = key.health_by_format or {}
+                circuit_by_format = key.circuit_breaker_by_format or {}
+
+                # 检查是否有任何格式健康度低于 0.5
+                for fmt, health_data in health_by_format.items():
+                    if float(health_data.get("health_score") or 1.0) < 0.5:
+                        unhealthy_keys += 1
+                        break
+
+                # 检查是否有任何格式熔断器开启
+                for fmt, circuit_data in circuit_by_format.items():
+                    if circuit_data.get("open"):
+                        circuit_open_keys += 1
+                        break

            return {
                "endpoints": {
@@ -582,10 +844,10 @@ class HealthMonitor:
                    "unhealthy": int(endpoint_stats.unhealthy or 0) if endpoint_stats else 0,
                },
                "keys": {
-                    "total": key_stats.total or 0 if key_stats else 0,
-                    "active": int(key_stats.active or 0) if key_stats else 0,
-                    "unhealthy": int(key_stats.unhealthy or 0) if key_stats else 0,
-                    "circuit_open": int(key_stats.circuit_open or 0) if key_stats else 0,
+                    "total": total_keys,
+                    "active": active_keys,
+                    "unhealthy": unhealthy_keys,
+                    "circuit_open": circuit_open_keys,
                },
            }

@@ -618,8 +880,9 @@ class HealthMonitor:
        db: Session,
        endpoint_id: Optional[str] = None,
        key_id: Optional[str] = None,
+        api_format: Optional[str] = None,
    ) -> bool:
-        """检查是否有资格进行探测（兼容旧接口）"""
+        """检查是否有资格进行探测（按 API 格式）"""
        if not cls.ALLOW_AUTO_RECOVER:
            return False

@@ -628,13 +891,53 @@ class HealthMonitor:

        if key_id:
            key = db.query(ProviderAPIKey).filter(ProviderAPIKey.id == key_id).first()
-            if key and key.circuit_breaker_open:
-                now = datetime.now(timezone.utc)
-                state = cls._get_circuit_state(key, now)
-                return state == CircuitState.HALF_OPEN
+            if key:
+                if api_format:
+                    circuit_data = cls._get_circuit_data(key, api_format)
+                    if circuit_data.get("open"):
+                        now = datetime.now(timezone.utc)
+                        state = cls._get_circuit_state_from_data(circuit_data, now)
+                        return state == CircuitState.HALF_OPEN
+                else:
+                    # 兼容旧调用：检查是否有任何格式处于半开状态
+                    circuit_by_format = key.circuit_breaker_by_format or {}
+                    now = datetime.now(timezone.utc)
+                    for fmt, circuit_data in circuit_by_format.items():
+                        if circuit_data.get("open"):
+                            state = cls._get_circuit_state_from_data(circuit_data, now)
+                            if state == CircuitState.HALF_OPEN:
+                                return True

        return False

+    # ==================== 便捷方法 ====================
+
+    @classmethod
+    def get_health_score(
+        cls, key: ProviderAPIKey, api_format: Optional[str] = None
+    ) -> float:
+        """获取指定格式的健康度分数"""
+        if not api_format:
+            # 返回所有格式中的最低健康度
+            health_by_format = key.health_by_format or {}
+            if not health_by_format:
+                return 1.0
+            return min(
+                float(h.get("health_score") or 1.0) for h in health_by_format.values()
+            )
+
+        health_data = cls._get_health_data(key, api_format)
+        return float(health_data.get("health_score") or 1.0)
+
+    @classmethod
+    def is_any_circuit_open(cls, key: ProviderAPIKey) -> bool:
+        """检查是否有任何格式的熔断器开启"""
+        circuit_by_format = key.circuit_breaker_by_format or {}
+        for circuit_data in circuit_by_format.values():
+            if circuit_data.get("open"):
+                return True
+        return False
+

 # 全局健康监控器实例
 health_monitor = HealthMonitor()
--- a/src/services/model/service.py
+++ b/src/services/model/service.py
@@ -216,7 +216,7 @@ class ModelService:
    def delete_model(db: Session, model_id: str):  # UUID
        """删除模型

-        新架构删除逻辑：
+        删除逻辑：
        - Model 只是 Provider 对 GlobalModel 的实现，删除不影响 GlobalModel
        - 检查是否是该 GlobalModel 的最后一个实现（如果是，警告但允许删除）
        """
@@ -384,7 +384,7 @@ class ModelService:

    @staticmethod
    def convert_to_response(model: Model) -> ModelResponse:
-        """转换为响应模型（新架构：从 GlobalModel 获取显示信息和默认值）"""
+        """转换为响应模型（从 GlobalModel 获取显示信息和默认值）"""
        return ModelResponse(
            id=model.id,
            provider_id=model.provider_id,
--- a/src/services/orchestration/candidate_resolver.py
+++ b/src/services/orchestration/candidate_resolver.py
@@ -171,7 +171,8 @@ class CandidateResolver:
                )
                candidate_record_map[(candidate_index, 0)] = record_id
            else:
-                max_retries_for_candidate = endpoint.max_retries if candidate.is_cached else 1
+                # max_retries 已从 Endpoint 迁移到 Provider（Endpoint 仍可能保留旧字段用于兼容）
+                max_retries_for_candidate = int(provider.max_retries or 2) if candidate.is_cached else 1

                for retry_index in range(max_retries_for_candidate):
                    record_id = str(uuid.uuid4())
@@ -236,7 +237,7 @@ class CandidateResolver:
        total = 0
        for candidate in all_candidates:
            if not candidate.is_skipped:
-                endpoint = candidate.endpoint
-                max_retries = int(endpoint.max_retries) if candidate.is_cached else 1
+                provider = candidate.provider
+                max_retries = int(provider.max_retries or 2) if candidate.is_cached else 1
                total += max_retries
        return total
--- a/src/services/orchestration/error_classifier.py
+++ b/src/services/orchestration/error_classifier.py
@@ -26,7 +26,7 @@ from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint
 from src.services.cache.aware_scheduler import CacheAwareScheduler
 from src.services.health.monitor import health_monitor
 from src.services.provider.format import normalize_api_format
-from src.services.rate_limit.adaptive_concurrency import get_adaptive_manager
+from src.services.rate_limit.adaptive_rpm import get_adaptive_rpm_manager
 from src.services.rate_limit.detector import RateLimitType, detect_rate_limit_type


@@ -112,7 +112,7 @@ class ErrorClassifier:
            cache_scheduler: 缓存调度器（可选）
        """
        self.db = db
-        self.adaptive_manager = adaptive_manager or get_adaptive_manager()
+        self.adaptive_manager = adaptive_manager or get_adaptive_rpm_manager()
        self.cache_scheduler = cache_scheduler

    # 表示客户端错误的 error type（不区分大小写）
@@ -361,7 +361,7 @@ class ErrorClassifier:
        self,
        key: ProviderAPIKey,
        provider_name: str,
-        current_concurrent: Optional[int],
+        current_rpm: Optional[int],
        exception: ProviderRateLimitException,
        request_id: Optional[str] = None,
    ) -> str:
@@ -371,7 +371,7 @@ class ErrorClassifier:
        Args:
            key: API Key 对象
            provider_name: 提供商名称
-            current_concurrent: 当前并发数
+            current_rpm: 当前分钟内的请求数
            exception: 速率限制异常
            request_id: 请求 ID（用于日志）

@@ -388,27 +388,27 @@ class ErrorClassifier:
            rate_limit_info = detect_rate_limit_type(
                headers=response_headers,
                provider_name=provider_name,
-                current_concurrent=current_concurrent,
+                current_usage=current_rpm,
            )

            logger.info(f"  [{request_id}] 429错误分析: "
                f"类型={rate_limit_info.limit_type}, "
                f"retry_after={rate_limit_info.retry_after}s, "
-                f"当前并发={current_concurrent}")
+                f"当前RPM={current_rpm}")

            # 调用自适应管理器处理
            new_limit = self.adaptive_manager.handle_429_error(
                db=self.db,
                key=key,
                rate_limit_info=rate_limit_info,
-                current_concurrent=current_concurrent,
+                current_rpm=current_rpm,
            )

            if rate_limit_info.limit_type == RateLimitType.CONCURRENT:
-                logger.warning(f"  [{request_id}] 自适应调整: " f"Key {key.id[:8]}... 并发限制 -> {new_limit}")
+                logger.warning(f"  [{request_id}] 并发限制触发（不调整RPM）")
                return "concurrent"
            elif rate_limit_info.limit_type == RateLimitType.RPM:
-                logger.info(f"  [{request_id}] [RPM] RPM限制，需要切换Provider")
+                logger.warning(f"  [{request_id}] 自适应调整: Key {key.id[:8]}... RPM限制 -> {new_limit}")
                return "rpm"
            else:
                return "unknown"
@@ -439,18 +439,18 @@ class ErrorClassifier:
        # 提取可读的错误消息
        extracted_message = self._extract_error_message(error_response_text)

-        # 构建详细错误信息
+        # 构建详细错误信息（仅用于日志，不暴露给客户端）
        if extracted_message:
-            detailed_message = f"提供商 '{provider_name}' 返回错误 {status}: {extracted_message}"
+            detailed_message = f"上游服务返回错误 {status}: {extracted_message}"
        else:
-            detailed_message = f"提供商 '{provider_name}' 返回错误: {status}"
+            detailed_message = f"上游服务返回错误: {status}"

        if status == 401:
            return ProviderAuthException(provider_name=provider_name)

        if status == 429:
            return ProviderRateLimitException(
-                message=error_response_text or f"提供商 '{provider_name}' 速率限制",
+                message="请求过于频繁，请稍后重试",
                provider_name=provider_name,
                response_headers=dict(error.response.headers) if error.response else None,
                retry_after=(
@@ -583,6 +583,7 @@ class ErrorClassifier:
                health_monitor.record_failure(
                    db=self.db,
                    key_id=str(key.id),
+                    api_format=api_format_str,
                    error_type="ProviderAuthException",
                )
            return extra_data
@@ -592,7 +593,7 @@ class ErrorClassifier:
            await self.handle_rate_limit(
                key=key,
                provider_name=provider_name,
-                current_concurrent=captured_key_concurrent,
+                current_rpm=captured_key_concurrent,
                exception=converted_error,
                request_id=request_id,
            )
@@ -620,6 +621,7 @@ class ErrorClassifier:
            health_monitor.record_failure(
                db=self.db,
                key_id=str(key.id),
+                api_format=api_format_str,
                error_type=type(converted_error).__name__,
            )

@@ -675,7 +677,7 @@ class ErrorClassifier:
            await self.handle_rate_limit(
                key=key,
                provider_name=provider_name,
-                current_concurrent=captured_key_concurrent,
+                current_rpm=captured_key_concurrent,
                exception=error,
                request_id=request_id,
            )
@@ -702,5 +704,6 @@ class ErrorClassifier:
            health_monitor.record_failure(
                db=self.db,
                key_id=str(key.id),
+                api_format=api_format_str,
                error_type=type(error).__name__,
            )
--- a/src/services/orchestration/fallback_orchestrator.py
+++ b/src/services/orchestration/fallback_orchestrator.py
@@ -44,7 +44,7 @@ from src.services.cache.aware_scheduler import (
    get_cache_aware_scheduler,
 )
 from src.services.provider.format import normalize_api_format
-from src.services.rate_limit.adaptive_concurrency import get_adaptive_manager
+from src.services.rate_limit.adaptive_rpm import get_adaptive_rpm_manager
 from src.services.rate_limit.concurrency_manager import get_concurrency_manager
 from src.services.request.candidate import RequestCandidateService
 from src.services.request.executor import ExecutionError, RequestExecutor
@@ -87,7 +87,7 @@ class FallbackOrchestrator:
        self.redis = redis_client
        self.cache_scheduler: Optional[CacheAwareScheduler] = None
        self.concurrency_manager: Any = None
-        self.adaptive_manager = get_adaptive_manager()  # 自适应并发管理器
+        self.adaptive_manager = get_adaptive_rpm_manager()  # 自适应 RPM 管理器
        self.request_executor: Optional[RequestExecutor] = None

        # 拆分后的组件（延迟初始化）
@@ -558,7 +558,8 @@ class FallbackOrchestrator:
        """尝试单个候选（含重试逻辑），返回执行结果"""
        provider = candidate.provider
        endpoint = candidate.endpoint
-        max_retries_for_candidate = int(endpoint.max_retries) if candidate.is_cached else 1
+        # 从 Provider 读取 max_retries（已从 Endpoint 迁移）
+        max_retries_for_candidate = int(provider.max_retries or 2) if candidate.is_cached else 1
        last_error: Optional[Exception] = None

        for retry_index in range(max_retries_for_candidate):
@@ -710,7 +711,7 @@ class FallbackOrchestrator:
                upstream_status = getattr(last_error, "upstream_status", None)
                upstream_response = getattr(last_error, "upstream_response", None)

-            # 如果响应为空或无效，使用异常的字符串表示
+            # 如果响应为空或无效，使用异常的字符串表示作为 upstream_response
            if (
                not upstream_response
                or not upstream_response.strip()
@@ -718,8 +719,17 @@ class FallbackOrchestrator:
            ):
                upstream_response = str(last_error)

+        # 构建友好的错误消息（用于返回给客户端，不暴露内部信息）
+        # 如果 last_error 有 message 属性，优先使用（已经是友好提示）
+        # 否则使用通用提示
+        friendly_message = "服务暂时不可用，请稍后重试"
+        if last_error:
+            last_error_message = getattr(last_error, "message", None)
+            if last_error_message and isinstance(last_error_message, str):
+                friendly_message = last_error_message
+
        raise ProviderNotAvailableException(
-            f"所有Provider均不可用，已尝试{max_attempts}个组合",
+            friendly_message,
            request_metadata=request_metadata,
            upstream_status=upstream_status,
            upstream_response=upstream_response,
--- a/src/services/rate_limit/init.py
+++ b/src/services/rate_limit/init.py
@@ -1,19 +1,23 @@
 """
 限流服务模块

-包含自适应并发控制、RPM限流、IP限流等功能。
+包含自适应 RPM 控制、并发管理、IP限流等功能。
 """

-from src.services.rate_limit.adaptive_concurrency import AdaptiveConcurrencyManager
+from src.services.rate_limit.adaptive_rpm import (
+    AdaptiveConcurrencyManager,  # 向后兼容别名
+    AdaptiveRPMManager,
+    get_adaptive_rpm_manager,
+)
 from src.services.rate_limit.concurrency_manager import ConcurrencyManager
 from src.services.rate_limit.detector import RateLimitDetector
 from src.services.rate_limit.ip_limiter import IPRateLimiter
-from src.services.rate_limit.rpm_limiter import RPMLimiter

 __all__ = [
-    "AdaptiveConcurrencyManager",
+    "AdaptiveConcurrencyManager",  # 向后兼容
+    "AdaptiveRPMManager",
    "ConcurrencyManager",
    "IPRateLimiter",
-    "RPMLimiter",
    "RateLimitDetector",
+    "get_adaptive_rpm_manager",
 ]
--- a/src/services/rate_limit/adaptive_reservation.py
+++ b/src/services/rate_limit/adaptive_reservation.py
@@ -98,7 +98,7 @@ class AdaptiveReservationManager:
    def calculate_reservation(
        self,
        key: "ProviderAPIKey",
-        current_concurrent: int = 0,
+        current_usage: int = 0,
        effective_limit: Optional[int] = None,
    ) -> ReservationResult:
        """
@@ -106,8 +106,8 @@ class AdaptiveReservationManager:

        Args:
            key: ProviderAPIKey 对象
-            current_concurrent: 当前并发数
-            effective_limit: 有效并发限制（学习值或配置值）
+            current_usage: 当前使用量（RPM 计数）
+            effective_limit: 有效限制（学习值或配置值）

        Returns:
            ReservationResult 包含预留比例和详细信息
@@ -116,7 +116,7 @@ class AdaptiveReservationManager:
        total_requests = self._get_total_requests(key)

        # 计算负载率
-        load_ratio = self._calculate_load_ratio(current_concurrent, effective_limit)
+        load_ratio = self._calculate_load_ratio(current_usage, effective_limit)

        # 阶段1: 探测阶段
        if total_requests < self.config.probe_phase_requests:
@@ -165,12 +165,12 @@ class AdaptiveReservationManager:
        return request_count

    def _calculate_load_ratio(
-        self, current_concurrent: int, effective_limit: Optional[int]
+        self, current_usage: int, effective_limit: Optional[int]
    ) -> float:
        """计算当前负载率"""
        if not effective_limit or effective_limit <= 0:
            return 0.0
-        return min(current_concurrent / effective_limit, 1.0)
+        return min(current_usage / effective_limit, 1.0)

    def _calculate_confidence(self, key: "ProviderAPIKey") -> float:
        """
--- a/src/services/rate_limit/adaptive_concurrency.py
+++ b/src/services/rate_limit/adaptive_concurrency.py
@@ -1,16 +1,16 @@
 """
-自适应并发调整器 - 基于边界记忆的并发限制调整
+自适应 RPM 调整器 - 基于边界记忆的 RPM 限制调整

 核心算法：边界记忆 + 渐进探测
- 触发 429 时记录边界（last_concurrent_peak），这就是真实上限
- 缩容策略：新限制 = 边界 - 1，而非乘性减少
+- 触发 429 时记录边界（last_rpm_peak），这就是真实上限
+- 缩容策略：新限制 = 边界 - 步长，而非乘性减少
 - 扩容策略：不超过已知边界，除非是探测性扩容
 - 探测性扩容：长时间无 429 时尝试突破边界

 设计原则：
 1. 快速收敛：一次 429 就能找到接近真实的限制
 2. 避免过度保守：不会因为多次 429 而无限下降
-3. 安全探测：允许在稳定后尝试更高并发
+3. 安全探测：允许在稳定后尝试更高 RPM
 """

 from datetime import datetime, timezone
@@ -18,7 +18,7 @@ from typing import Any, Dict, List, Optional, cast

 from sqlalchemy.orm import Session

-from src.config.constants import ConcurrencyDefaults
+from src.config.constants import RPMDefaults
 from src.core.batch_committer import get_batch_committer
 from src.core.logger import logger
 from src.models.database import ProviderAPIKey
@@ -33,14 +33,14 @@ class AdaptiveStrategy:
    AGGRESSIVE = "aggressive"  # 激进策略（快速探测）


-class AdaptiveConcurrencyManager:
+class AdaptiveRPMManager:
    """
-    自适应并发管理器
+    自适应 RPM 管理器

    核心算法：边界记忆 + 渐进探测
-    - 触发 429 时记录边界（last_concurrent_peak = 触发时的并发数）
-    - 缩容：新限制 = 边界 - 1（快速收敛到真实限制附近）
-    - 扩容：不超过边界（即 last_concurrent_peak），允许回到边界值尝试
+    - 触发 429 时记录边界（last_rpm_peak = 触发时的 RPM）
+    - 缩容：新限制 = 边界 - 步长（快速收敛到真实限制附近）
+    - 扩容：不超过边界（即 last_rpm_peak），允许回到边界值尝试
    - 探测性扩容：长时间（30分钟）无 429 时，可以尝试 +1 突破边界

    扩容条件（满足任一即可）：
@@ -50,35 +50,35 @@ class AdaptiveConcurrencyManager:
    关键特性：
    1. 快速收敛：一次 429 就能学到接近真实的限制值
    2. 边界保护：普通扩容不会超过已知边界
-    3. 安全探测：长时间稳定后允许尝试更高并发
+    3. 安全探测：长时间稳定后允许尝试更高 RPM
    4. 区分并发限制和 RPM 限制
    """

    # 默认配置 - 使用统一常量
-    DEFAULT_INITIAL_LIMIT = ConcurrencyDefaults.INITIAL_LIMIT
-    MIN_CONCURRENT_LIMIT = ConcurrencyDefaults.MIN_CONCURRENT_LIMIT
-    MAX_CONCURRENT_LIMIT = ConcurrencyDefaults.MAX_CONCURRENT_LIMIT
+    DEFAULT_INITIAL_LIMIT = RPMDefaults.INITIAL_LIMIT
+    MIN_RPM_LIMIT = RPMDefaults.MIN_RPM_LIMIT
+    MAX_RPM_LIMIT = RPMDefaults.MAX_RPM_LIMIT

    # AIMD 参数
-    INCREASE_STEP = ConcurrencyDefaults.INCREASE_STEP
+    INCREASE_STEP = RPMDefaults.INCREASE_STEP

    # 滑动窗口参数
-    UTILIZATION_WINDOW_SIZE = ConcurrencyDefaults.UTILIZATION_WINDOW_SIZE
-    UTILIZATION_WINDOW_SECONDS = ConcurrencyDefaults.UTILIZATION_WINDOW_SECONDS
-    UTILIZATION_THRESHOLD = ConcurrencyDefaults.UTILIZATION_THRESHOLD
-    HIGH_UTILIZATION_RATIO = ConcurrencyDefaults.HIGH_UTILIZATION_RATIO
-    MIN_SAMPLES_FOR_DECISION = ConcurrencyDefaults.MIN_SAMPLES_FOR_DECISION
+    UTILIZATION_WINDOW_SIZE = RPMDefaults.UTILIZATION_WINDOW_SIZE
+    UTILIZATION_WINDOW_SECONDS = RPMDefaults.UTILIZATION_WINDOW_SECONDS
+    UTILIZATION_THRESHOLD = RPMDefaults.UTILIZATION_THRESHOLD
+    HIGH_UTILIZATION_RATIO = RPMDefaults.HIGH_UTILIZATION_RATIO
+    MIN_SAMPLES_FOR_DECISION = RPMDefaults.MIN_SAMPLES_FOR_DECISION

    # 探测性扩容参数
-    PROBE_INCREASE_INTERVAL_MINUTES = ConcurrencyDefaults.PROBE_INCREASE_INTERVAL_MINUTES
-    PROBE_INCREASE_MIN_REQUESTS = ConcurrencyDefaults.PROBE_INCREASE_MIN_REQUESTS
+    PROBE_INCREASE_INTERVAL_MINUTES = RPMDefaults.PROBE_INCREASE_INTERVAL_MINUTES
+    PROBE_INCREASE_MIN_REQUESTS = RPMDefaults.PROBE_INCREASE_MIN_REQUESTS

    # 记录历史数量
    MAX_HISTORY_RECORDS = 20

    def __init__(self, strategy: str = AdaptiveStrategy.AIMD):
        """
-        初始化自适应并发管理器
+        初始化自适应 RPM 管理器

        Args:
            strategy: 调整策略
@@ -90,54 +90,54 @@ class AdaptiveConcurrencyManager:
        db: Session,
        key: ProviderAPIKey,
        rate_limit_info: RateLimitInfo,
-        current_concurrent: Optional[int] = None,
+        current_rpm: Optional[int] = None,
    ) -> int:
        """
-        处理429错误，调整并发限制
+        处理429错误，调整 RPM 限制

        Args:
            db: 数据库会话
            key: API Key对象
            rate_limit_info: 速率限制信息
-            current_concurrent: 当前并发数
+            current_rpm: 当前分钟内的请求数

        Returns:
-            调整后的并发限制
+            调整后的 RPM 限制
        """
-        # max_concurrent=NULL 表示启用自适应，max_concurrent=数字 表示固定限制
-        is_adaptive = key.max_concurrent is None
+        # rpm_limit=NULL 表示启用自适应，rpm_limit=数字 表示固定限制
+        is_adaptive = key.rpm_limit is None

        if not is_adaptive:
            logger.debug(
-                f"Key {key.id} 设置了固定并发限制 ({key.max_concurrent})，跳过自适应调整"
+                f"Key {key.id} 设置了固定 RPM 限制 ({key.rpm_limit})，跳过自适应调整"
            )
-            return int(key.max_concurrent)  # type: ignore[arg-type]
+            return int(key.rpm_limit)  # type: ignore[arg-type]

        # 更新429统计
        key.last_429_at = datetime.now(timezone.utc)  # type: ignore[assignment]
        key.last_429_type = rate_limit_info.limit_type  # type: ignore[assignment]
-        # 仅在并发限制且拿到并发数时记录边界（RPM/UNKNOWN 不应覆盖并发边界记忆）
+        # 仅在 RPM 限制且拿到 RPM 数时记录边界
        if (
-            rate_limit_info.limit_type == RateLimitType.CONCURRENT
-            and current_concurrent is not None
-            and current_concurrent > 0
+            rate_limit_info.limit_type == RateLimitType.RPM
+            and current_rpm is not None
+            and current_rpm > 0
        ):
-            key.last_concurrent_peak = current_concurrent  # type: ignore[assignment]
+            key.last_rpm_peak = current_rpm  # type: ignore[assignment]

        # 遇到 429 错误，清空利用率采样窗口（重新开始收集）
        key.utilization_samples = []  # type: ignore[assignment]

-        if rate_limit_info.limit_type == RateLimitType.CONCURRENT:
-            # 并发限制：减少并发数
-            key.concurrent_429_count = int(key.concurrent_429_count or 0) + 1  # type: ignore[assignment]
+        if rate_limit_info.limit_type == RateLimitType.RPM:
+            # RPM 限制：减少 RPM 限制
+            key.rpm_429_count = int(key.rpm_429_count or 0) + 1  # type: ignore[assignment]

-            # 获取当前有效限制（自适应模式使用 learned_max_concurrent）
-            old_limit = int(key.learned_max_concurrent or self.DEFAULT_INITIAL_LIMIT)
-            new_limit = self._decrease_limit(old_limit, current_concurrent)
+            # 获取当前有效限制（自适应模式使用 learned_rpm_limit）
+            old_limit = int(key.learned_rpm_limit or self.DEFAULT_INITIAL_LIMIT)
+            new_limit = self._decrease_limit(old_limit, current_rpm)

            logger.warning(
-                f"[CONCURRENT] 并发限制触发: Key {key.id[:8]}... | "
-                f"当前并发: {current_concurrent} | "
+                f"[RPM] RPM 限制触发: Key {key.id[:8]}... | "
+                f"当前 RPM: {current_rpm} | "
                f"调整: {old_limit} -> {new_limit}"
            )

@@ -146,79 +146,78 @@ class AdaptiveConcurrencyManager:
                key,
                old_limit=old_limit,
                new_limit=new_limit,
-                reason="concurrent_429",
-                current_concurrent=current_concurrent,
+                reason="rpm_429",
+                current_rpm=current_rpm,
            )

-            # 更新学习到的并发限制
-            key.learned_max_concurrent = new_limit  # type: ignore[assignment]
+            # 更新学习到的 RPM 限制
+            key.learned_rpm_limit = new_limit  # type: ignore[assignment]

-        elif rate_limit_info.limit_type == RateLimitType.RPM:
-            # RPM限制：不调整并发，只记录
-            key.rpm_429_count = int(key.rpm_429_count or 0) + 1  # type: ignore[assignment]
+        elif rate_limit_info.limit_type == RateLimitType.CONCURRENT:
+            # 并发限制：不调整 RPM，只记录
+            key.concurrent_429_count = int(key.concurrent_429_count or 0) + 1  # type: ignore[assignment]

            logger.info(
-                f"[RPM] RPM限制触发: Key {key.id[:8]}... | "
-                f"retry_after: {rate_limit_info.retry_after}s | "
-                f"不调整并发限制"
+                f"[CONCURRENT] 并发限制触发: Key {key.id[:8]}... | "
+                f"不调整 RPM 限制（这是并发问题，非 RPM 问题）"
            )

        else:
            # 未知类型：保守处理，轻微减少
            logger.warning(
                f"[UNKNOWN] 未知429类型: Key {key.id[:8]}... | "
-                f"当前并发: {current_concurrent} | "
-                f"保守减少并发"
+                f"当前 RPM: {current_rpm} | "
+                f"保守减少 RPM"
            )

-            old_limit = int(key.learned_max_concurrent or self.DEFAULT_INITIAL_LIMIT)
-            new_limit = max(int(old_limit * 0.9), self.MIN_CONCURRENT_LIMIT)  # 减少10%
+            old_limit = int(key.learned_rpm_limit or self.DEFAULT_INITIAL_LIMIT)
+            new_limit = max(int(old_limit * 0.9), self.MIN_RPM_LIMIT)  # 减少10%

            self._record_adjustment(
                key,
                old_limit=old_limit,
                new_limit=new_limit,
                reason="unknown_429",
-                current_concurrent=current_concurrent,
+                current_rpm=current_rpm,
            )

-            key.learned_max_concurrent = new_limit  # type: ignore[assignment]
+            key.learned_rpm_limit = new_limit  # type: ignore[assignment]

        db.flush()
        get_batch_committer().mark_dirty(db)

-        return int(key.learned_max_concurrent or self.DEFAULT_INITIAL_LIMIT)
+        return int(key.learned_rpm_limit or self.DEFAULT_INITIAL_LIMIT)

    def handle_success(
        self,
        db: Session,
        key: ProviderAPIKey,
-        current_concurrent: int,
+        current_rpm: int,
    ) -> Optional[int]:
        """
-        处理成功请求，基于滑动窗口利用率考虑增加并发限制
+        处理成功请求，基于滑动窗口利用率考虑增加 RPM 限制

        Args:
            db: 数据库会话
            key: API Key对象
-            current_concurrent: 当前并发数（必需，用于计算利用率）
+            current_rpm: 当前分钟内的请求数（必需，用于计算利用率）

        Returns:
-            调整后的并发限制（如果有调整），否则返回 None
+            调整后的 RPM 限制（如果有调整），否则返回 None
        """
-        # max_concurrent=NULL 表示启用自适应
-        is_adaptive = key.max_concurrent is None
+        # rpm_limit=NULL 表示启用自适应
+        is_adaptive = key.rpm_limit is None

        if not is_adaptive:
            return None

-        current_limit = int(key.learned_max_concurrent or self.DEFAULT_INITIAL_LIMIT)
+        current_limit = int(key.learned_rpm_limit or self.DEFAULT_INITIAL_LIMIT)

-        # 获取已知边界（上次触发 429 时的并发数）
-        known_boundary = key.last_concurrent_peak
+        # 获取已知边界（上次触发 429 时的 RPM）
+        known_boundary = key.last_rpm_peak

        # 计算当前利用率
-        utilization = float(current_concurrent / current_limit) if current_limit > 0 else 0.0
+        utilization = float(current_rpm / current_limit) if current_limit > 0 else 0.0

        now = datetime.now(timezone.utc)
        now_ts = now.timestamp()
@@ -229,7 +228,7 @@ class AdaptiveConcurrencyManager:
        # 检查是否满足扩容条件
        increase_reason = self._check_increase_conditions(key, samples, now, known_boundary)

-        if increase_reason and current_limit < self.MAX_CONCURRENT_LIMIT:
+        if increase_reason and current_limit < self.MAX_RPM_LIMIT:
            old_limit = current_limit
            is_probe = increase_reason == "probe_increase"
            new_limit = self._increase_limit(current_limit, known_boundary, is_probe)
@@ -262,12 +261,12 @@ class AdaptiveConcurrencyManager:
                avg_utilization=round(avg_util, 2),
                high_util_ratio=round(high_util_ratio, 2),
                sample_count=len(samples),
-                current_concurrent=current_concurrent,
+                current_rpm=current_rpm,
                known_boundary=known_boundary,
            )

            # 更新限制
-            key.learned_max_concurrent = new_limit  # type: ignore[assignment]
+            key.learned_rpm_limit = new_limit  # type: ignore[assignment]

            # 如果是探测性扩容，更新探测时间
            if is_probe:
@@ -334,7 +333,7 @@ class AdaptiveConcurrencyManager:
            key: API Key对象
            samples: 利用率采样列表
            now: 当前时间
-            known_boundary: 已知边界（触发 429 时的并发数）
+            known_boundary: 已知边界（触发 429 时的 RPM）

        Returns:
            扩容原因（如果满足条件），否则返回 None
@@ -343,7 +342,7 @@ class AdaptiveConcurrencyManager:
        if self._is_in_cooldown(key):
            return None

-        current_limit = int(key.learned_max_concurrent or self.DEFAULT_INITIAL_LIMIT)
+        current_limit = int(key.learned_rpm_limit or self.DEFAULT_INITIAL_LIMIT)

        # 条件1：滑动窗口扩容（不超过边界）
        if len(samples) >= self.MIN_SAMPLES_FOR_DECISION:
@@ -353,7 +352,7 @@ class AdaptiveConcurrencyManager:
            if high_util_ratio >= self.HIGH_UTILIZATION_RATIO:
                # 检查是否还有扩容空间（边界保护）
                if known_boundary:
-                    # 允许扩容到边界值（而非 boundary - 1），因为缩容时已经 -1 了
+                    # 允许扩容到边界值（而非 boundary - 1），因为缩容时已经 -步长 了
                    if current_limit < known_boundary:
                        return "high_utilization"
                    # 已达边界，不触发普通扩容
@@ -429,34 +428,37 @@ class AdaptiveConcurrencyManager:

        last_429_at = cast(datetime, key.last_429_at)
        time_since_429 = (datetime.now(timezone.utc) - last_429_at).total_seconds()
-        cooldown_seconds = ConcurrencyDefaults.COOLDOWN_AFTER_429_MINUTES * 60
+        cooldown_seconds = RPMDefaults.COOLDOWN_AFTER_429_MINUTES * 60

        return bool(time_since_429 < cooldown_seconds)

    def _decrease_limit(
        self,
        current_limit: int,
-        current_concurrent: Optional[int] = None,
+        current_rpm: Optional[int] = None,
    ) -> int:
        """
-        减少并发限制（基于边界记忆策略）
+        减少 RPM 限制（基于边界记忆策略）

        策略：
-        - 如果知道触发 429 时的并发数，新限制 = 并发数 - 1
-        - 这样可以快速收敛到真实限制附近，而不会过度保守
-        - 例如：真实限制 8，触发时并发 8 -> 新限制 7（而非 8*0.85=6）
+        - 如果知道触发 429 时的 RPM，新限制 = RPM * 0.90（保留 10% 安全边际）
+        - 10% 的安全边际更保守，考虑到：
+          1. RPM 报告可能存在延迟，实际触发时的 RPM 可能略高于报告值
+          2. 上游 API 的限制可能有波动
+          3. 避免频繁在边界附近触发 429
+        - 相比固定步长，百分比方式更适应不同量级的限制值
        """
-        if current_concurrent is not None and current_concurrent > 0:
-            # 边界记忆策略：新限制 = 触发边界 - 1
-            candidate = current_concurrent - 1
+        if current_rpm is not None and current_rpm > 0:
+            # 边界记忆策略：新限制 = 触发边界 * 0.90（10% 安全边际）
+            candidate = int(current_rpm * 0.90)
        else:
-            # 没有并发信息时，保守减少 1
-            candidate = current_limit - 1
+            # 没有 RPM 信息时，减少 10%
+            candidate = int(current_limit * 0.9)

-        # 保证不会“缩容变扩容”（例如 current_concurrent > current_limit 的异常场景）
+        # 保证不会"缩容变扩容"
        candidate = min(candidate, current_limit - 1)

-        new_limit = max(candidate, self.MIN_CONCURRENT_LIMIT)
+        new_limit = max(candidate, self.MIN_RPM_LIMIT)

        return new_limit

@@ -467,16 +469,15 @@ class AdaptiveConcurrencyManager:
        is_probe: bool = False,
    ) -> int:
        """
-        增加并发限制（考虑边界保护）
+        增加 RPM 限制（考虑边界保护）

        策略：
        - 普通扩容：每次 +INCREASE_STEP，但不超过 known_boundary
-          （因为缩容时已经 -1 了，这里允许回到边界值尝试）
        - 探测性扩容：每次只 +1，可以突破边界，但要谨慎

        Args:
            current_limit: 当前限制
-            known_boundary: 已知边界（last_concurrent_peak），即触发 429 时的并发数
+            known_boundary: 已知边界（last_rpm_peak），即触发 429 时的 RPM
            is_probe: 是否是探测性扩容（可以突破边界）
        """
        if is_probe:
@@ -486,13 +487,13 @@ class AdaptiveConcurrencyManager:
            # 普通模式：每次 +INCREASE_STEP
            new_limit = current_limit + self.INCREASE_STEP

-            # 边界保护：普通扩容不超过 known_boundary（允许回到边界值尝试）
+            # 边界保护：普通扩容不超过 known_boundary
            if known_boundary:
                if new_limit > known_boundary:
                    new_limit = known_boundary

        # 全局上限保护
-        new_limit = min(new_limit, self.MAX_CONCURRENT_LIMIT)
+        new_limit = min(new_limit, self.MAX_RPM_LIMIT)

        # 确保有增长（否则返回原值表示不扩容）
        if new_limit <= current_limit:
@@ -509,7 +510,7 @@ class AdaptiveConcurrencyManager:
        **extra_data: Any,
    ) -> None:
        """
-        记录并发调整历史
+        记录 RPM 调整历史

        Args:
            key: API Key对象
@@ -548,10 +549,10 @@ class AdaptiveConcurrencyManager:
        history: List[Dict[str, Any]] = list(key.adjustment_history or [])
        samples: List[Dict[str, Any]] = list(key.utilization_samples or [])

-        # max_concurrent=NULL 表示自适应，否则为固定限制
-        is_adaptive = key.max_concurrent is None
-        current_limit = int(key.learned_max_concurrent or self.DEFAULT_INITIAL_LIMIT)
-        effective_limit = current_limit if is_adaptive else int(key.max_concurrent)  # type: ignore
+        # rpm_limit=NULL 表示自适应，否则为固定限制
+        is_adaptive = key.rpm_limit is None
+        current_limit = int(key.learned_rpm_limit or self.DEFAULT_INITIAL_LIMIT)
+        effective_limit = current_limit if is_adaptive else int(key.rpm_limit)  # type: ignore

        # 计算窗口统计
        avg_utilization: Optional[float] = None
@@ -570,15 +571,15 @@ class AdaptiveConcurrencyManager:
            last_probe_at_str = cast(datetime, key.last_probe_increase_at).isoformat()

        # 边界信息
-        known_boundary = key.last_concurrent_peak
+        known_boundary = key.last_rpm_peak

        return {
            "adaptive_mode": is_adaptive,
-            "max_concurrent": key.max_concurrent,  # NULL=自适应，数字=固定限制
+            "rpm_limit": key.rpm_limit,  # NULL=自适应，数字=固定限制
            "effective_limit": effective_limit,  # 当前有效限制
-            "learned_limit": key.learned_max_concurrent,  # 学习到的限制
+            "learned_limit": key.learned_rpm_limit,  # 学习到的限制
            # 边界记忆相关
-            "known_boundary": known_boundary,  # 触发 429 时的并发数（已知上限）
+            "known_boundary": known_boundary,  # 触发 429 时的 RPM（已知上限）
            "concurrent_429_count": int(key.concurrent_429_count or 0),
            "rpm_429_count": int(key.rpm_429_count or 0),
            "last_429_at": last_429_at_str,
@@ -607,12 +608,12 @@ class AdaptiveConcurrencyManager:
        """
        logger.info(f"[RESET] 重置学习状态: Key {key.id[:8]}...")

-        key.learned_max_concurrent = None  # type: ignore[assignment]
+        key.learned_rpm_limit = None  # type: ignore[assignment]
        key.concurrent_429_count = 0  # type: ignore[assignment]
        key.rpm_429_count = 0  # type: ignore[assignment]
        key.last_429_at = None  # type: ignore[assignment]
        key.last_429_type = None  # type: ignore[assignment]
-        key.last_concurrent_peak = None  # type: ignore[assignment]
+        key.last_rpm_peak = None  # type: ignore[assignment]
        key.adjustment_history = []  # type: ignore[assignment]
        key.utilization_samples = []  # type: ignore[assignment]
        key.last_probe_increase_at = None  # type: ignore[assignment]
@@ -622,12 +623,17 @@ class AdaptiveConcurrencyManager:


 # 全局单例
-_adaptive_manager: Optional[AdaptiveConcurrencyManager] = None
+_adaptive_rpm_manager: Optional[AdaptiveRPMManager] = None


-def get_adaptive_manager() -> AdaptiveConcurrencyManager:
-    """获取全局自适应管理器单例"""
-    global _adaptive_manager
-    if _adaptive_manager is None:
-        _adaptive_manager = AdaptiveConcurrencyManager()
-    return _adaptive_manager
+def get_adaptive_rpm_manager() -> AdaptiveRPMManager:
+    """获取全局自适应 RPM 管理器单例"""
+    global _adaptive_rpm_manager
+    if _adaptive_rpm_manager is None:
+        _adaptive_rpm_manager = AdaptiveRPMManager()
+    return _adaptive_rpm_manager
+
+
+# 向后兼容别名
+AdaptiveConcurrencyManager = AdaptiveRPMManager
+get_adaptive_manager = get_adaptive_rpm_manager
--- a/src/services/rate_limit/concurrency_manager.py
+++ b/src/services/rate_limit/concurrency_manager.py
@@ -1,29 +1,33 @@
 """
-并发管理器 - 支持 Redis 或内存的并发控制
+RPM 限制管理器 - 支持 Redis 或内存的 Key 级别 RPM 限制

 功能：
-1. Endpoint 级别的并发限制
-2. ProviderAPIKey 级别的并发限制
-3. 分布式环境下优先使用 Redis，多实例共享
-4. 在开发/单实例场景下自动降级为内存计数
-5. 自动释放和异常处理（Redis 提供 TTL，内存模式请确保手动释放）
+1. ProviderAPIKey 级别的 RPM 限制（按分钟窗口计数）
+2. 分布式环境下优先使用 Redis，多实例共享
+3. 在开发/单实例场景下自动降级为内存计数
+4. 支持缓存用户优先级（预留槽位机制）
 """

 import asyncio
 import math
+import os
+import time
 from contextlib import asynccontextmanager
-from datetime import timedelta  # noqa: F401 - kept for potential future use
-from typing import Optional, Tuple
+from typing import Optional

 import redis.asyncio as aioredis
+
+from src.config.constants import RPMDefaults
 from src.core.logger import logger


 class ConcurrencyManager:
-    """分布式并发管理器"""
+    """Key RPM 限制管理器"""

    _instance: Optional["ConcurrencyManager"] = None
    _redis: Optional[aioredis.Redis] = None
+    _key_rpm_bucket_seconds: int = 60
+    _key_rpm_key_ttl_seconds: int = 120  # 2 分钟，足够覆盖当前分钟与边界

    def __new__(cls):
        """单例模式"""
@@ -37,9 +41,22 @@ class ConcurrencyManager:
            return

        self._memory_lock: asyncio.Lock = asyncio.Lock()
-        self._memory_endpoint_counts: dict[str, int] = {}
-        self._memory_key_counts: dict[str, int] = {}
+        # Key RPM 计数器：{key_id: (bucket, count)}，bucket = floor(now / 60)
+        self._memory_key_rpm_counts: dict[str, tuple[int, int]] = {}
        self._owns_redis: bool = False
+        self._last_cleanup_bucket: int = 0  # 上次清理时的 bucket，用于定期清理过期数据
+        self._last_cleanup_time: float = 0  # 上次清理的时间戳，用于强制定期清理
+        self._cleanup_interval_seconds: int = 300  # 强制清理间隔（5 分钟）
+        self._cleanup_task: Optional[asyncio.Task] = None  # 后台清理任务
+
+        # 内存模式下的最大条目限制，防止内存泄漏（支持环境变量覆盖）
+        self._max_memory_rpm_entries: int = int(
+            os.getenv("RPM_MAX_MEMORY_ENTRIES", str(RPMDefaults.MAX_MEMORY_RPM_ENTRIES))
+        )
+        # 早期告警阈值（达到此比例时记录警告）
+        self._memory_warning_threshold: float = float(
+            os.getenv("RPM_MEMORY_WARNING_THRESHOLD", str(RPMDefaults.MEMORY_WARNING_THRESHOLD))
+        )
        self._memory_initialized = True

    async def initialize(self) -> None:
@@ -56,212 +73,304 @@ class ConcurrencyManager:
            if self._redis:
                logger.info("[OK] ConcurrencyManager 已复用全局 Redis 客户端")
            else:
-                logger.warning("[WARN] Redis 不可用，并发控制降级为内存模式（仅在单实例环境下安全）")
+                logger.warning("[WARN] Redis 不可用，RPM 限制降级为内存模式（仅在单实例环境下安全）")
+                # 内存模式下启动后台清理任务
+                self._start_background_cleanup()
        except Exception as e:
            logger.error(f"[ERROR] 获取全局 Redis 客户端失败: {e}")
-            logger.warning("[WARN] 并发控制将降级为内存模式（仅在单实例环境下安全）")
+            logger.warning("[WARN] RPM 限制将降级为内存模式（仅在单实例环境下安全）")
            self._redis = None
            self._owns_redis = False
+            # 内存模式下启动后台清理任务
+            self._start_background_cleanup()
+
+    def _start_background_cleanup(self) -> None:
+        """启动后台定期清理任务（仅内存模式需要）"""
+        if self._cleanup_task is not None:
+            return  # 已经启动
+
+        async def cleanup_loop():
+            """后台清理循环"""
+            while True:
+                try:
+                    await asyncio.sleep(60)  # 每分钟检查一次
+                    async with self._memory_lock:
+                        current_bucket = self._get_rpm_bucket()
+                        self._cleanup_expired_memory_rpm_counts(current_bucket, force=False)
+                except asyncio.CancelledError:
+                    break
+                except Exception as e:
+                    logger.debug(f"后台清理任务异常: {e}")
+
+        try:
+            self._cleanup_task = asyncio.create_task(cleanup_loop())
+            logger.debug("[OK] 内存模式后台清理任务已启动")
+        except RuntimeError:
+            # 没有事件循环时忽略
+            pass

    async def close(self) -> None:
        """关闭 Redis 连接"""
+        # 停止后台清理任务
+        if self._cleanup_task is not None:
+            self._cleanup_task.cancel()
+            try:
+                await self._cleanup_task
+            except asyncio.CancelledError:
+                pass
+            self._cleanup_task = None
+
        if self._redis and self._owns_redis:
            await self._redis.close()
            logger.info("ConcurrencyManager Redis 连接已关闭")
        self._redis = None
        self._owns_redis = False

-    def _get_endpoint_key(self, endpoint_id: str) -> str:
-        """获取 Endpoint 并发计数的 Redis Key"""
-        return f"concurrency:endpoint:{endpoint_id}"
+    @classmethod
+    def _get_rpm_bucket(cls, now_ts: Optional[float] = None) -> int:
+        """获取当前 RPM 计数桶（按分钟）"""
+        ts = now_ts if now_ts is not None else time.time()
+        return int(ts // cls._key_rpm_bucket_seconds)

-    def _get_key_key(self, key_id: str) -> str:
-        """获取 ProviderAPIKey 并发计数的 Redis Key"""
-        return f"concurrency:key:{key_id}"
+    @classmethod
+    def _get_key_key(cls, key_id: str, bucket: Optional[int] = None) -> str:
+        """获取 ProviderAPIKey RPM 计数的 Redis Key（按分钟桶）"""
+        b = bucket if bucket is not None else cls._get_rpm_bucket()
+        return f"rpm:key:{key_id}:{b}"

-    async def get_current_concurrency(
-        self, endpoint_id: Optional[str] = None, key_id: Optional[str] = None
-    ) -> Tuple[int, int]:
+    def _get_memory_key_rpm_count(self, key_id: str, bucket: int) -> int:
+        """获取内存模式下 Key 在指定 bucket 的 RPM 计数"""
+        stored = self._memory_key_rpm_counts.get(key_id)
+        if not stored:
+            return 0
+        stored_bucket, count = stored
+        if stored_bucket != bucket:
+            # 旧桶数据已过期，删除以防止内存泄漏
+            del self._memory_key_rpm_counts[key_id]
+            return 0
+        return count
+
+    def _set_memory_key_rpm_count(self, key_id: str, bucket: int, count: int) -> None:
+        """设置内存模式下 Key 在指定 bucket 的 RPM 计数"""
+        current_size = len(self._memory_key_rpm_counts)
+        warning_threshold = int(self._max_memory_rpm_entries * self._memory_warning_threshold)
+        high_threshold = int(self._max_memory_rpm_entries * 0.8)
+        critical_threshold = int(self._max_memory_rpm_entries * 0.95)
+
+        # 分级告警：根据使用率记录不同级别的日志
+        if current_size >= critical_threshold and key_id not in self._memory_key_rpm_counts:
+            logger.critical(
+                f"[CRITICAL] 内存 RPM 计数器接近上限 ({current_size}/{self._max_memory_rpm_entries})，"
+                f"强烈建议启用 Redis！继续增长可能导致 RPM 限制失效"
+            )
+        elif current_size >= high_threshold and key_id not in self._memory_key_rpm_counts:
+            # 每 100 个条目告警一次，避免日志过多
+            if current_size % 100 == 0:
+                logger.error(
+                    f"[ERROR] 内存 RPM 计数器使用率过高 ({current_size}/{self._max_memory_rpm_entries})，"
+                    f"建议启用 Redis"
+                )
+        elif current_size >= warning_threshold and key_id not in self._memory_key_rpm_counts:
+            if current_size == warning_threshold:
+                logger.warning(
+                    f"[WARN] 内存 RPM 计数器达到 {self._memory_warning_threshold:.0%} 阈值 "
+                    f"({current_size}/{self._max_memory_rpm_entries})，建议启用 Redis"
+                )
+
+        # 检查是否超过最大条目限制
+        if (
+            key_id not in self._memory_key_rpm_counts
+            and current_size >= self._max_memory_rpm_entries
+        ):
+            # 触发强制清理
+            self._cleanup_expired_memory_rpm_counts(bucket, force=True)
+            # 如果清理后仍然超过限制，执行 LRU 淘汰（删除最旧的 20%）
+            if len(self._memory_key_rpm_counts) >= self._max_memory_rpm_entries:
+                evict_count = max(1, self._max_memory_rpm_entries // 5)
+                # 按 bucket（时间）排序，删除最旧的
+                sorted_keys = sorted(
+                    self._memory_key_rpm_counts.items(),
+                    key=lambda x: x[1][0]  # 按 bucket 排序
+                )
+                for k, _ in sorted_keys[:evict_count]:
+                    del self._memory_key_rpm_counts[k]
+                logger.warning(
+                    f"[WARN] 内存 RPM 计数器达到上限，已淘汰 {evict_count} 个最旧条目"
+                )
+        self._memory_key_rpm_counts[key_id] = (bucket, count)
+
+    def _cleanup_expired_memory_rpm_counts(self, current_bucket: int, force: bool = False) -> None:
        """
-        获取当前并发数
+        清理内存中过期的 RPM 计数（必须在持有 _memory_lock 时调用）
+
+        清理策略：
+        - 常规清理：每分钟最多执行一次（当 bucket 变化时）
+        - 强制清理：每 5 分钟执行一次（防止长时间无请求导致内存泄漏）
+        """
+        now = time.time()
+
+        # 检查是否需要清理
+        should_cleanup = (
+            current_bucket != self._last_cleanup_bucket  # 分钟切换
+            or force  # 强制清理
+            or (now - self._last_cleanup_time > self._cleanup_interval_seconds)  # 超时清理
+        )
+
+        if not should_cleanup:
+            return
+
+        self._last_cleanup_bucket = current_bucket
+        self._last_cleanup_time = now
+        expired_keys = []
+
+        for key_id, (stored_bucket, _count) in self._memory_key_rpm_counts.items():
+            if stored_bucket < current_bucket:
+                expired_keys.append(key_id)
+
+        for key_id in expired_keys:
+            del self._memory_key_rpm_counts[key_id]
+
+        if expired_keys:
+            logger.debug(f"[CLEANUP] 清理了 {len(expired_keys)} 个过期的内存 RPM 计数")
+
+    async def get_key_rpm_count(self, key_id: str) -> int:
+        """
+        获取 Key 当前 RPM 计数

        Args:
-            endpoint_id: Endpoint ID（可选）
-            key_id: ProviderAPIKey ID（可选）
+            key_id: ProviderAPIKey ID

        Returns:
-            (endpoint_concurrency, key_concurrency)
+            当前分钟窗口内的请求数
        """
        if self._redis is None:
            async with self._memory_lock:
-                endpoint_count = (
-                    self._memory_endpoint_counts.get(endpoint_id, 0) if endpoint_id else 0
-                )
-                key_count = self._memory_key_counts.get(key_id, 0) if key_id else 0
-                return endpoint_count, key_count
-
-        endpoint_count = 0
-        key_count = 0
+                bucket = self._get_rpm_bucket()
+                # 定期清理过期数据，避免内存泄漏
+                self._cleanup_expired_memory_rpm_counts(bucket)
+                return self._get_memory_key_rpm_count(key_id, bucket)

        try:
-            if endpoint_id:
-                endpoint_key = self._get_endpoint_key(endpoint_id)
-                result = await self._redis.get(endpoint_key)
-                endpoint_count = int(result) if result else 0
-
-            if key_id:
-                key_key = self._get_key_key(key_id)
-                result = await self._redis.get(key_key)
-                key_count = int(result) if result else 0
-
+            key_key = self._get_key_key(key_id)
+            result = await self._redis.get(key_key)
+            return int(result) if result else 0
        except Exception as e:
-            logger.error(f"获取并发数失败: {e}")
+            logger.error(f"获取 RPM 计数失败: {e}")
+            return 0

-        return endpoint_count, key_count
-
-    async def check_available(
+    async def check_rpm_available(
        self,
-        endpoint_id: str,
-        endpoint_max_concurrent: Optional[int],
        key_id: str,
-        key_max_concurrent: Optional[int],
+        key_rpm_limit: Optional[int],
+        is_cached_user: bool = False,
+        cache_reservation_ratio: Optional[float] = None,
    ) -> bool:
        """
-        检查是否可以获取并发槽位（不实际获取）
+        检查是否可以通过 RPM 限制（不实际增加计数）

        Args:
-            endpoint_id: Endpoint ID
-            endpoint_max_concurrent: Endpoint 最大并发数（None 表示不限制）
            key_id: ProviderAPIKey ID
-            key_max_concurrent: Key 最大并发数（None 表示不限制）
+            key_rpm_limit: Key RPM 限制（每分钟最大请求数，None 表示不限制）
+            is_cached_user: 是否是缓存用户
+            cache_reservation_ratio: 缓存预留比例

        Returns:
            是否可用（True/False）
        """
-        if self._redis is None:
-            async with self._memory_lock:
-                endpoint_count = self._memory_endpoint_counts.get(endpoint_id, 0)
-                key_count = self._memory_key_counts.get(key_id, 0)
+        if key_rpm_limit is None:
+            return True

-                if (
-                    endpoint_max_concurrent is not None
-                    and endpoint_count >= endpoint_max_concurrent
-                ):
-                    return False
+        # 从配置读取默认值
+        from src.config.settings import config

-                if key_max_concurrent is not None and key_count >= key_max_concurrent:
-                    return False
+        if cache_reservation_ratio is None:
+            cache_reservation_ratio = config.cache_reservation_ratio

-                return True
+        key_count = await self.get_key_rpm_count(key_id)

-        endpoint_count, key_count = await self.get_current_concurrency(endpoint_id, key_id)
+        if is_cached_user:
+            return key_count < key_rpm_limit
+        else:
+            # 新用户只能使用 (1 - cache_reservation_ratio) 的槽位
+            available_for_new = max(1, math.floor(key_rpm_limit * (1 - cache_reservation_ratio)))
+            return key_count < available_for_new

-        # 检查 Endpoint 级别限制
-        if endpoint_max_concurrent is not None and endpoint_count >= endpoint_max_concurrent:
-            return False
-
-        # 检查 Key 级别限制
-        if key_max_concurrent is not None and key_count >= key_max_concurrent:
-            return False
-
-        return True
-
-    async def acquire_slot(
+    async def acquire_rpm_slot(
        self,
-        endpoint_id: str,
-        endpoint_max_concurrent: Optional[int],
        key_id: str,
-        key_max_concurrent: Optional[int],
-        is_cached_user: bool = False,  # 新增：是否是缓存用户
-        cache_reservation_ratio: Optional[float] = None,  # 缓存预留比例，None 时从配置读取
-        ttl_seconds: Optional[int] = None,  # TTL 秒数，None 时从配置读取
+        key_rpm_limit: Optional[int],
+        is_cached_user: bool = False,
+        cache_reservation_ratio: Optional[float] = None,
    ) -> bool:
        """
-        尝试获取并发槽位（支持缓存用户优先级）
+        尝试获取 RPM 槽位（支持缓存用户优先级）

        Args:
-            endpoint_id: Endpoint ID
-            endpoint_max_concurrent: Endpoint 最大并发数（None 表示不限制）
            key_id: ProviderAPIKey ID
-            key_max_concurrent: Key 最大并发数（None 表示不限制）
+            key_rpm_limit: Key RPM 限制（每分钟最大请求数，None 表示不限制）
            is_cached_user: 是否是缓存用户（缓存用户可使用全部槽位）
            cache_reservation_ratio: 缓存预留比例，None 时从配置读取
-            ttl_seconds: TTL 秒数，None 时从配置读取

        Returns:
            是否成功获取（True/False）

        缓存预留机制说明:
-        - 假设 key_max_concurrent = 10, cache_reservation_ratio = 0.3
-        - 新用户最多使用: 7个槽位 (10 * (1 - 0.3))
-        - 缓存用户最多使用: 10个槽位（全部）
-        - 预留的3个槽位专门给缓存用户，保证他们的请求优先
+        - 假设 key_rpm_limit = 100, cache_reservation_ratio = 0.3
+        - 新用户最多使用: 70 RPM (100 * (1 - 0.3))
+        - 缓存用户最多使用: 100 RPM（全部）
+        - 预留的 30 RPM 专门给缓存用户，保证他们的请求优先
        """
        # 从配置读取默认值
        from src.config.settings import config

        if cache_reservation_ratio is None:
            cache_reservation_ratio = config.cache_reservation_ratio
-        if ttl_seconds is None:
-            ttl_seconds = config.concurrency_slot_ttl

        if self._redis is None:
            async with self._memory_lock:
-                endpoint_count = self._memory_endpoint_counts.get(endpoint_id, 0)
-                key_count = self._memory_key_counts.get(key_id, 0)
+                bucket = self._get_rpm_bucket()
+                # 定期清理过期数据，避免内存泄漏
+                self._cleanup_expired_memory_rpm_counts(bucket)

-                # Endpoint 限制
-                if (
-                    endpoint_max_concurrent is not None
-                    and endpoint_count >= endpoint_max_concurrent
-                ):
-                    return False
+                key_count = self._get_memory_key_rpm_count(key_id, bucket)

-                # Key 限制，包含缓存预留
-                if key_max_concurrent is not None:
+                # Key RPM 限制，包含缓存预留
+                if key_rpm_limit is not None:
                    if is_cached_user:
-                        if key_count >= key_max_concurrent:
+                        if key_count >= key_rpm_limit:
                            return False
                    else:
+                        # 新用户只能使用 (1 - cache_reservation_ratio) 的槽位
                        available_for_new = max(
-                            1, math.ceil(key_max_concurrent * (1 - cache_reservation_ratio))
+                            1, math.floor(key_rpm_limit * (1 - cache_reservation_ratio))
                        )
                        if key_count >= available_for_new:
                            return False

                # 通过限制，更新计数
-                self._memory_endpoint_counts[endpoint_id] = endpoint_count + 1
-                self._memory_key_counts[key_id] = key_count + 1
+                self._set_memory_key_rpm_count(key_id, bucket, key_count + 1)
                return True

-        endpoint_key = self._get_endpoint_key(endpoint_id)
-        key_key = self._get_key_key(key_id)
+        bucket = self._get_rpm_bucket()
+        key_key = self._get_key_key(key_id, bucket=bucket)

        try:
-            # 使用 Lua 脚本保证原子性（新增缓存预留逻辑）
+            # 使用 Lua 脚本保证原子性（支持缓存预留逻辑）
            lua_script = """
-            local endpoint_key = KEYS[1]
-            local key_key = KEYS[2]
-            local endpoint_max = tonumber(ARGV[1])
-            local key_max = tonumber(ARGV[2])
-            local ttl = tonumber(ARGV[3])
-            local is_cached = tonumber(ARGV[4])  -- 0=新用户, 1=缓存用户
-            local cache_ratio = tonumber(ARGV[5])  -- 缓存预留比例
+            local key_key = KEYS[1]
+            local key_max = tonumber(ARGV[1])
+            local key_ttl = tonumber(ARGV[2])
+            local is_cached = tonumber(ARGV[3])  -- 0=新用户, 1=缓存用户
+            local cache_ratio = tonumber(ARGV[4])  -- 缓存预留比例

            -- 获取当前值
-            local endpoint_count = tonumber(redis.call('GET', endpoint_key) or '0')
            local key_count = tonumber(redis.call('GET', key_key) or '0')

-            -- 检查 endpoint 限制（-1 表示不限制）
-            if endpoint_max >= 0 and endpoint_count >= endpoint_max then
-                return 0  -- 失败：endpoint 已满
-            end
-
            -- 检查 key 限制（支持缓存预留）
            if key_max >= 0 then
                if is_cached == 0 then
                    -- 新用户：只能使用 (1 - cache_ratio) 的槽位
-                    local available_for_new = math.floor(key_max * (1 - cache_ratio))
+                    local available_for_new = math.max(1, math.floor(key_max * (1 - cache_ratio)))
                    if key_count >= available_for_new then
                        return 0  -- 失败：新用户配额已满
                    end
@@ -274,10 +383,8 @@ class ConcurrencyManager:
            end

            -- 增加计数
-            redis.call('INCR', endpoint_key)
-            redis.call('EXPIRE', endpoint_key, ttl)
            redis.call('INCR', key_key)
-            redis.call('EXPIRE', key_key, ttl)
+            redis.call('EXPIRE', key_key, key_ttl)

            return 1  -- 成功
            """
@@ -285,12 +392,10 @@ class ConcurrencyManager:
            # 执行脚本
            result = await self._redis.eval(
                lua_script,
-                2,  # 2 个 KEYS
-                endpoint_key,
+                1,  # 1 个 KEY
                key_key,
-                endpoint_max_concurrent if endpoint_max_concurrent is not None else -1,
-                key_max_concurrent if key_max_concurrent is not None else -1,
-                ttl_seconds,
+                key_rpm_limit if key_rpm_limit is not None else -1,
+                self._key_rpm_key_ttl_seconds,
                1 if is_cached_user else 0,  # 缓存用户标志
                cache_reservation_ratio,  # 预留比例
            )
@@ -299,143 +404,68 @@ class ConcurrencyManager:

            if success:
                user_type = "缓存用户" if is_cached_user else "新用户"
-                logger.debug(
-                    f"[OK] 获取并发槽位成功: endpoint={endpoint_id}, key={key_id}, "
-                    f"类型={user_type}"
-                )
+                logger.debug(f"[OK] 获取 RPM 槽位成功: key={key_id}, 类型={user_type}")
            else:
-                endpoint_count, key_count = await self.get_current_concurrency(endpoint_id, key_id)
+                key_count = await self.get_key_rpm_count(key_id)

-                # 计算新用户可用槽位
-                if key_max_concurrent and not is_cached_user:
-                    available_for_new = int(key_max_concurrent * (1 - cache_reservation_ratio))
+                # 计算新用户可用 RPM
+                if key_rpm_limit and not is_cached_user:
+                    available_for_new = int(key_rpm_limit * (1 - cache_reservation_ratio))
                    user_info = f"新用户配额={available_for_new}, 当前={key_count}"
                else:
-                    user_info = f"缓存用户, 当前={key_count}/{key_max_concurrent}"
+                    user_info = f"缓存用户, 当前={key_count}/{key_rpm_limit}"

-                logger.warning(
-                    f"[WARN] 并发槽位已满: endpoint={endpoint_id}({endpoint_count}/{endpoint_max_concurrent}), "
-                    f"key={key_id}({user_info})"
-                )
+                logger.warning(f"[WARN] RPM 限制已达上限: key={key_id}({user_info})")

            return success

        except Exception as e:
-            logger.error(f"获取并发槽位失败，降级到内存模式: {e}")
+            logger.error(f"获取 RPM 槽位失败，降级到内存模式: {e}")
            # Redis 异常时降级到内存模式进行保守限流
-            # 使用较低的限制值（原限制的 50%）避免上游 API 被打爆
            async with self._memory_lock:
-                endpoint_count = self._memory_endpoint_counts.get(endpoint_id, 0)
-                key_count = self._memory_key_counts.get(key_id, 0)
+                bucket = self._get_rpm_bucket()
+                self._cleanup_expired_memory_rpm_counts(bucket)
+
+                key_count = self._get_memory_key_rpm_count(key_id, bucket)

                # 降级模式下使用更保守的限制（50%）
-                fallback_endpoint_limit = (
-                    max(1, endpoint_max_concurrent // 2)
-                    if endpoint_max_concurrent is not None
-                    else None
-                )
-                fallback_key_limit = (
-                    max(1, key_max_concurrent // 2) if key_max_concurrent is not None else None
+                fallback_rpm_limit = (
+                    max(1, key_rpm_limit // 2) if key_rpm_limit is not None else None
                )

-                if (
-                    fallback_endpoint_limit is not None
-                    and endpoint_count >= fallback_endpoint_limit
-                ):
+                if fallback_rpm_limit is not None and key_count >= fallback_rpm_limit:
                    logger.warning(
-                        f"[FALLBACK] Endpoint 并发达到降级限制: {endpoint_count}/{fallback_endpoint_limit}"
-                    )
-                    return False
-
-                if fallback_key_limit is not None and key_count >= fallback_key_limit:
-                    logger.warning(
-                        f"[FALLBACK] Key 并发达到降级限制: {key_count}/{fallback_key_limit}"
+                        f"[FALLBACK] Key RPM 达到降级限制: {key_count}/{fallback_rpm_limit}"
                    )
                    return False

                # 更新内存计数
-                self._memory_endpoint_counts[endpoint_id] = endpoint_count + 1
-                self._memory_key_counts[key_id] = key_count + 1
-                logger.debug(
-                    f"[FALLBACK] 使用内存模式获取槽位: endpoint={endpoint_id}, key={key_id}"
-                )
+                self._set_memory_key_rpm_count(key_id, bucket, key_count + 1)
+                logger.debug(f"[FALLBACK] 使用内存模式获取 RPM 槽位: key={key_id}")
                return True

-    async def release_slot(self, endpoint_id: str, key_id: str) -> None:
-        """
-        释放并发槽位
-
-        Args:
-            endpoint_id: Endpoint ID
-            key_id: ProviderAPIKey ID
-        """
-        if self._redis is None:
-            async with self._memory_lock:
-                if endpoint_id in self._memory_endpoint_counts:
-                    self._memory_endpoint_counts[endpoint_id] = max(
-                        0, self._memory_endpoint_counts[endpoint_id] - 1
-                    )
-                    if self._memory_endpoint_counts[endpoint_id] == 0:
-                        self._memory_endpoint_counts.pop(endpoint_id, None)
-
-                if key_id in self._memory_key_counts:
-                    self._memory_key_counts[key_id] = max(0, self._memory_key_counts[key_id] - 1)
-                    if self._memory_key_counts[key_id] == 0:
-                        self._memory_key_counts.pop(key_id, None)
-            return
-
-        endpoint_key = self._get_endpoint_key(endpoint_id)
-        key_key = self._get_key_key(key_id)
-
-        try:
-            # 使用 Lua 脚本保证原子性（不会减到负数）
-            lua_script = """
-            local endpoint_key = KEYS[1]
-            local key_key = KEYS[2]
-
-            local endpoint_count = tonumber(redis.call('GET', endpoint_key) or '0')
-            local key_count = tonumber(redis.call('GET', key_key) or '0')
-
-            if endpoint_count > 0 then
-                redis.call('DECR', endpoint_key)
-            end
-
-            if key_count > 0 then
-                redis.call('DECR', key_key)
-            end
-
-            return 1
-            """
-
-            await self._redis.eval(lua_script, 2, endpoint_key, key_key)
-
-            logger.debug(f"[OK] 释放并发槽位: endpoint={endpoint_id}, key={key_id}")
-
-        except Exception as e:
-            logger.error(f"释放并发槽位失败: {e}")
-
    @asynccontextmanager
-    async def concurrency_guard(
+    async def rpm_guard(
        self,
-        endpoint_id: str,
-        endpoint_max_concurrent: Optional[int],
        key_id: str,
-        key_max_concurrent: Optional[int],
-        is_cached_user: bool = False,  # 新增：是否是缓存用户
-        cache_reservation_ratio: Optional[float] = None,  # 缓存预留比例，None 时从配置读取
+        key_rpm_limit: Optional[int],
+        is_cached_user: bool = False,
+        cache_reservation_ratio: Optional[float] = None,
    ):
        """
-        并发控制上下文管理器（支持缓存用户优先级）
+        RPM 限制上下文管理器（支持缓存用户优先级）

        用法：
-            async with manager.concurrency_guard(
-                endpoint_id, endpoint_max, key_id, key_max,
+            async with manager.rpm_guard(
+                key_id, key_rpm_limit,
                is_cached_user=True  # 缓存用户
            ):
                # 执行请求
                response = await send_request(...)

        如果获取失败，会抛出 ConcurrencyLimitError 异常
+
+        注意：RPM 是按分钟窗口计数，不需要在请求结束后释放
        """
        # 从配置读取默认值
        from src.config.settings import config
@@ -444,11 +474,9 @@ class ConcurrencyManager:
            cache_reservation_ratio = config.cache_reservation_ratio

        # 尝试获取槽位（传递缓存用户参数）
-        acquired = await self.acquire_slot(
-            endpoint_id,
-            endpoint_max_concurrent,
+        acquired = await self.acquire_rpm_slot(
            key_id,
-            key_max_concurrent,
+            key_rpm_limit,
            is_cached_user,
            cache_reservation_ratio,
        )
@@ -458,7 +486,7 @@ class ConcurrencyManager:

            user_type = "缓存用户" if is_cached_user else "新用户"
            raise ConcurrencyLimitError(
-                f"并发限制已达上限: endpoint={endpoint_id}, key={key_id}, 类型={user_type}"
+                f"RPM 限制已达上限: key={key_id}, 类型={user_type}"
            )

        # 记录开始时间和状态
@@ -469,7 +497,7 @@ class ConcurrencyManager:

        try:
            yield  # 执行请求
-        except Exception as e:
+        except Exception:
            # 记录异常
            exception_occurred = True
            raise
@@ -498,7 +526,7 @@ class ConcurrencyManager:
                # 告警：槽位占用时间过长（超过 60 秒）
                if slot_duration > 60:
                    logger.warning(
-                        f"[WARN] 并发槽位占用时间过长: "
+                        f"[WARN] 请求耗时过长: "
                        f"key_id={key_id[:8] if key_id else 'unknown'}..., "
                        f"duration={slot_duration:.1f}s, "
                        f"exception={exception_occurred}"
@@ -506,67 +534,64 @@ class ConcurrencyManager:

            except Exception as metric_error:
                # 指标记录失败不应影响业务逻辑
-                logger.debug(f"记录并发指标失败: {metric_error}")
+                logger.debug(f"记录指标失败: {metric_error}")

-            # 自动释放槽位（即使发生异常）
-            await self.release_slot(endpoint_id, key_id)
+            # 注意：RPM 计数不需要在请求结束后释放，它会在分钟窗口过期后自动重置

-    async def reset_concurrency(
-        self, endpoint_id: Optional[str] = None, key_id: Optional[str] = None
-    ) -> None:
+    async def reset_key_rpm(self, key_id: str) -> None:
        """
-        重置并发计数（管理功能，慎用）
+        重置 Key RPM 计数（管理功能，慎用）

        Args:
-            endpoint_id: Endpoint ID（可选，None 表示重置所有 endpoint）
-            key_id: ProviderAPIKey ID（可选，None 表示重置所有 key）
+            key_id: ProviderAPIKey ID
        """
        if self._redis is None:
            async with self._memory_lock:
-                if endpoint_id:
-                    self._memory_endpoint_counts.pop(endpoint_id, None)
-                    logger.info(f"[RESET] 重置 Endpoint 并发计数(内存): {endpoint_id}")
-                else:
-                    count = len(self._memory_endpoint_counts)
-                    self._memory_endpoint_counts.clear()
-                    if count:
-                        logger.info(f"[RESET] 重置所有 Endpoint 并发计数(内存): {count} 个")
-
-                if key_id:
-                    self._memory_key_counts.pop(key_id, None)
-                    logger.info(f"[RESET] 重置 Key 并发计数(内存): {key_id}")
-                else:
-                    count = len(self._memory_key_counts)
-                    self._memory_key_counts.clear()
-                    if count:
-                        logger.info(f"[RESET] 重置所有 Key 并发计数(内存): {count} 个")
+                self._memory_key_rpm_counts.pop(key_id, None)
+                logger.info(f"[RESET] 重置 Key RPM 计数(内存): {key_id}")
            return

        try:
-            if endpoint_id:
-                endpoint_key = self._get_endpoint_key(endpoint_id)
-                await self._redis.delete(endpoint_key)
-                logger.info(f"[RESET] 重置 Endpoint 并发计数: {endpoint_id}")
-            else:
-                # 重置所有 endpoint
-                keys = await self._redis.keys("concurrency:endpoint:*")
-                if keys:
-                    await self._redis.delete(*keys)
-                    logger.info(f"[RESET] 重置所有 Endpoint 并发计数: {len(keys)} 个")
-
-            if key_id:
-                key_key = self._get_key_key(key_id)
-                await self._redis.delete(key_key)
-                logger.info(f"[RESET] 重置 Key 并发计数: {key_id}")
-            else:
-                # 重置所有 key
-                keys = await self._redis.keys("concurrency:key:*")
-                if keys:
-                    await self._redis.delete(*keys)
-                    logger.info(f"[RESET] 重置所有 Key 并发计数: {len(keys)} 个")
-
+            deleted_count = await self._scan_and_delete(f"rpm:key:{key_id}:*")
+            logger.info(f"[RESET] 重置 Key RPM 计数: {key_id}, 删除 {deleted_count} 个键")
        except Exception as e:
-            logger.error(f"重置并发计数失败: {e}")
+            logger.error(f"重置 Key RPM 计数失败: {e}")
+
+    async def reset_all_rpm(self) -> None:
+        """重置所有 Key RPM 计数（管理功能，慎用）"""
+        if self._redis is None:
+            async with self._memory_lock:
+                count = len(self._memory_key_rpm_counts)
+                self._memory_key_rpm_counts.clear()
+                if count:
+                    logger.info(f"[RESET] 重置所有 Key RPM 计数(内存): {count} 个")
+            return
+
+        try:
+            deleted_count = await self._scan_and_delete("rpm:key:*")
+            if deleted_count:
+                logger.info(f"[RESET] 重置所有 Key RPM 计数: {deleted_count} 个")
+        except Exception as e:
+            logger.error(f"重置所有 Key RPM 计数失败: {e}")
+
+    async def _scan_and_delete(self, pattern: str, batch_size: int = 100) -> int:
+        """使用 SCAN 遍历并分批删除匹配的键，避免阻塞 Redis"""
+        if self._redis is None:
+            return 0
+
+        deleted_count = 0
+        cursor = 0
+        while True:
+            cursor, keys = await self._redis.scan(cursor, match=pattern, count=batch_size)
+            if keys:
+                # 分批删除，每批最多 batch_size 个
+                for i in range(0, len(keys), batch_size):
+                    batch = keys[i : i + batch_size]
+                    await self._redis.delete(*batch)
+                    deleted_count += len(batch)
+            if cursor == 0:
+                break
+        return deleted_count


 # 全局单例
--- a/src/services/rate_limit/detector.py
+++ b/src/services/rate_limit/detector.py
@@ -3,7 +3,7 @@
 """

 from datetime import datetime, timezone
-from typing import Any, Dict, Optional, Tuple
+from typing import Dict, Optional

 from src.core.logger import logger

@@ -62,7 +62,7 @@ class RateLimitDetector:
    def detect_from_headers(
        headers: Dict[str, str],
        provider_name: str = "unknown",
-        current_concurrent: Optional[int] = None,
+        current_usage: Optional[int] = None,
    ) -> RateLimitInfo:
        """
        从响应头中检测速率限制类型
@@ -70,7 +70,7 @@ class RateLimitDetector:
        Args:
            headers: 429响应的HTTP头
            provider_name: 提供商名称（用于选择解析策略）
-            current_concurrent: 当前并发数（用于判断是否为并发限制）
+            current_usage: 当前使用量（RPM 计数，用于启发式判断是否为并发限制）

        Returns:
            RateLimitInfo对象
@@ -80,16 +80,16 @@ class RateLimitDetector:

        # 根据提供商选择解析策略
        if "anthropic" in provider_name.lower() or "claude" in provider_name.lower():
-            return RateLimitDetector._parse_anthropic_headers(headers_lower, current_concurrent)
+            return RateLimitDetector._parse_anthropic_headers(headers_lower, current_usage)
        elif "openai" in provider_name.lower():
-            return RateLimitDetector._parse_openai_headers(headers_lower, current_concurrent)
+            return RateLimitDetector._parse_openai_headers(headers_lower, current_usage)
        else:
-            return RateLimitDetector._parse_generic_headers(headers_lower, current_concurrent)
+            return RateLimitDetector._parse_generic_headers(headers_lower, current_usage)

    @staticmethod
    def _parse_anthropic_headers(
        headers: Dict[str, str],
-        current_concurrent: Optional[int] = None,
+        current_usage: Optional[int] = None,
    ) -> RateLimitInfo:
        """
        解析 Anthropic Claude API 的速率限制头
@@ -127,29 +127,66 @@ class RateLimitDetector:
                raw_headers=headers,
            )

-        # 2. 可能的并发限制判断（多条件综合）
-        # 条件：当前并发数存在，且 remaining > 0（说明不是 RPM 耗尽）
-        # 同时 retry_after 较短（并发限制通常 retry_after 较短，如 1-10 秒）
-        is_likely_concurrent = (
-            current_concurrent is not None
-            and current_concurrent >= 2  # 至少有 2 个并发
-            and (requests_remaining is None or requests_remaining > 0)  # RPM 未耗尽
-            and (retry_after is None or retry_after <= 30)  # 短暂等待
-        )
+        # 2. 并发限制判断（多条件策略）
+        # 注意：current_usage 是 RPM 计数（当前分钟请求数），不是真正的并发数
+        #
+        # 判断条件（满足任一即可）：
+        # A. 强判断：remaining > 0 且 retry_after <= 30（Provider 明确告知还有配额但需要等待）
+        # B. 弱判断：只有 retry_after <= 5 且缺少 remaining 头（短等待时间是并发限制的典型特征）
+        #
+        # 选择保守的 retry_after 阈值：
+        # - 强判断用 30 秒（有 remaining 头时）
+        # - 弱判断用 5 秒（无 remaining 头时，更保守）
+        is_likely_concurrent = False
+        concurrent_reason = ""
+
+        # 条件 A：remaining > 0 且 retry_after <= 30
+        if (
+            requests_remaining is not None
+            and requests_remaining > 0
+            and retry_after is not None
+            and retry_after <= 30
+        ):
+            is_likely_concurrent = True
+            concurrent_reason = f"remaining={requests_remaining} > 0, retry_after={retry_after}s <= 30s"
+
+        # 条件 B：无 remaining 头但 retry_after 很短（<= 5 秒）
+        elif (
+            requests_remaining is None
+            and retry_after is not None
+            and retry_after <= 5
+        ):
+            is_likely_concurrent = True
+            concurrent_reason = f"no remaining header, retry_after={retry_after}s <= 5s"

        if is_likely_concurrent:
-            logger.info(
-                f"检测到可能的并发限制: current_concurrent={current_concurrent}, "
-                f"remaining={requests_remaining}, retry_after={retry_after}"
-            )
+            logger.info(f"检测到并发限制: {concurrent_reason}")
            return RateLimitInfo(
                limit_type=RateLimitType.CONCURRENT,
                retry_after=retry_after,
-                current_usage=current_concurrent,
+                current_usage=current_usage,
                raw_headers=headers,
            )

-        # 3. 未知类型
+        # 3. 默认视为 RPM 限制（更保守的处理）
+        # 无法明确区分时，视为 RPM 限制让系统降低 RPM
+        # 这比误判为并发限制（不降 RPM）更安全
+        if retry_after is not None or requests_limit is not None:
+            logger.info(
+                f"无法明确区分限制类型，保守视为 RPM 限制: "
+                f"remaining={requests_remaining}, retry_after={retry_after}"
+            )
+            return RateLimitInfo(
+                limit_type=RateLimitType.RPM,
+                retry_after=retry_after,
+                limit_value=requests_limit,
+                remaining=requests_remaining,
+                reset_at=requests_reset,
+                current_usage=current_usage,
+                raw_headers=headers,
+            )
+
+        # 4. 完全没有信息，标记为未知
        return RateLimitInfo(
            limit_type=RateLimitType.UNKNOWN,
            retry_after=retry_after,
@@ -159,7 +196,7 @@ class RateLimitDetector:
    @staticmethod
    def _parse_openai_headers(
        headers: Dict[str, str],
-        current_concurrent: Optional[int] = None,
+        current_usage: Optional[int] = None,
    ) -> RateLimitInfo:
        """
        解析 OpenAI API 的速率限制头
@@ -195,23 +232,55 @@ class RateLimitDetector:
                raw_headers=headers,
            )

-        # 2. 可能的并发限制（多条件综合判断）
-        is_likely_concurrent = (
-            current_concurrent is not None
-            and current_concurrent >= 2
-            and (requests_remaining is None or requests_remaining > 0)
-            and (retry_after is None or retry_after <= 30)
-        )
+        # 2. 并发限制判断（多条件策略）
+        # 判断条件（满足任一即可）：
+        # A. 强判断：remaining > 0 且 retry_after <= 30
+        # B. 弱判断：只有 retry_after <= 5 且缺少 remaining 头
+        is_likely_concurrent = False
+        concurrent_reason = ""
+
+        if (
+            requests_remaining is not None
+            and requests_remaining > 0
+            and retry_after is not None
+            and retry_after <= 30
+        ):
+            is_likely_concurrent = True
+            concurrent_reason = f"remaining={requests_remaining} > 0, retry_after={retry_after}s <= 30s"
+        elif (
+            requests_remaining is None
+            and retry_after is not None
+            and retry_after <= 5
+        ):
+            is_likely_concurrent = True
+            concurrent_reason = f"no remaining header, retry_after={retry_after}s <= 5s"

        if is_likely_concurrent:
+            logger.info(f"检测到并发限制: {concurrent_reason}")
            return RateLimitInfo(
                limit_type=RateLimitType.CONCURRENT,
                retry_after=retry_after,
-                current_usage=current_concurrent,
+                current_usage=current_usage,
                raw_headers=headers,
            )

-        # 3. 未知类型
+        # 3. 默认视为 RPM 限制（更保守的处理）
+        if retry_after is not None or requests_limit is not None:
+            logger.info(
+                f"无法明确区分限制类型，保守视为 RPM 限制: "
+                f"remaining={requests_remaining}, retry_after={retry_after}"
+            )
+            return RateLimitInfo(
+                limit_type=RateLimitType.RPM,
+                retry_after=retry_after,
+                limit_value=requests_limit,
+                remaining=requests_remaining,
+                reset_at=requests_reset,
+                current_usage=current_usage,
+                raw_headers=headers,
+            )
+
+        # 4. 完全没有信息，标记为未知
        return RateLimitInfo(
            limit_type=RateLimitType.UNKNOWN,
            retry_after=retry_after,
@@ -221,7 +290,7 @@ class RateLimitDetector:
    @staticmethod
    def _parse_generic_headers(
        headers: Dict[str, str],
-        current_concurrent: Optional[int] = None,
+        current_usage: Optional[int] = None,
    ) -> RateLimitInfo:
        """
        解析通用的速率限制头
@@ -247,23 +316,54 @@ class RateLimitDetector:
                raw_headers=headers,
            )

-        # 2. 可能的并发限制
-        is_likely_concurrent = (
-            current_concurrent is not None
-            and current_concurrent >= 2
-            and (remaining is None or remaining > 0)
-            and (retry_after is None or retry_after <= 30)
-        )
+        # 2. 并发限制判断（多条件策略）
+        # 判断条件（满足任一即可）：
+        # A. 强判断：remaining > 0 且 retry_after <= 30
+        # B. 弱判断：只有 retry_after <= 5 且缺少 remaining 头
+        is_likely_concurrent = False
+        concurrent_reason = ""
+
+        if (
+            remaining is not None
+            and remaining > 0
+            and retry_after is not None
+            and retry_after <= 30
+        ):
+            is_likely_concurrent = True
+            concurrent_reason = f"remaining={remaining} > 0, retry_after={retry_after}s <= 30s"
+        elif (
+            remaining is None
+            and retry_after is not None
+            and retry_after <= 5
+        ):
+            is_likely_concurrent = True
+            concurrent_reason = f"no remaining header, retry_after={retry_after}s <= 5s"

        if is_likely_concurrent:
+            logger.info(f"检测到并发限制: {concurrent_reason}")
            return RateLimitInfo(
                limit_type=RateLimitType.CONCURRENT,
                retry_after=retry_after,
-                current_usage=current_concurrent,
+                current_usage=current_usage,
                raw_headers=headers,
            )

-        # 3. 未知类型
+        # 3. 默认视为 RPM 限制（更保守的处理）
+        if retry_after is not None or limit_value is not None:
+            logger.info(
+                f"无法明确区分限制类型，保守视为 RPM 限制: "
+                f"remaining={remaining}, retry_after={retry_after}"
+            )
+            return RateLimitInfo(
+                limit_type=RateLimitType.RPM,
+                retry_after=retry_after,
+                limit_value=limit_value,
+                remaining=remaining,
+                current_usage=current_usage,
+                raw_headers=headers,
+            )
+
+        # 4. 完全没有信息，标记为未知
        return RateLimitInfo(
            limit_type=RateLimitType.UNKNOWN,
            retry_after=retry_after,
@@ -317,7 +417,7 @@ class RateLimitDetector:
 def detect_rate_limit_type(
    headers: Dict[str, str],
    provider_name: str = "unknown",
-    current_concurrent: Optional[int] = None,
+    current_usage: Optional[int] = None,
 ) -> RateLimitInfo:
    """
    检测速率限制类型（便捷函数）
@@ -325,9 +425,9 @@ def detect_rate_limit_type(
    Args:
        headers: 429响应头
        provider_name: 提供商名称
-        current_concurrent: 当前并发数
+        current_usage: 当前使用量（RPM 计数）

    Returns:
        RateLimitInfo对象
    """
-    return RateLimitDetector.detect_from_headers(headers, provider_name, current_concurrent)
+    return RateLimitDetector.detect_from_headers(headers, provider_name, current_usage)
--- a/src/services/rate_limit/rpm_limiter.py
+++ b/src/services/rate_limit/rpm_limiter.py
@@ -1,135 +0,0 @@
-"""
-RPM (Requests Per Minute) 限流服务
-"""
-
-import time
-from datetime import datetime, timedelta, timezone
-from typing import Dict, Tuple
-
-from sqlalchemy.orm import Session
-
-from src.core.batch_committer import get_batch_committer
-from src.core.logger import logger
-from src.models.database import Provider
-from src.models.database_extensions import ProviderUsageTracking
-
-
-
-class RPMLimiter:
-    """RPM限流器"""
-
-    def __init__(self, db: Session):
-        self.db = db
-        # 内存中的RPM计数器 {provider_id: (count, window_start)}
-        self._rpm_counters: Dict[str, Tuple[int, float]] = {}
-
-    def check_and_increment(self, provider_id: str) -> bool:
-        """
-        检查并递增RPM计数
-
-        Returns:
-            True if allowed, False if rate limited
-        """
-        provider = self.db.query(Provider).filter(Provider.id == provider_id).first()
-        if not provider:
-            return True
-
-        rpm_limit = provider.rpm_limit
-        if rpm_limit is None:
-            # 未设置限制
-            return True
-
-        if rpm_limit == 0:
-            logger.warning(f"Provider {provider.name} is fully restricted by RPM limit=0")
-            return False
-
-        current_time = time.time()
-
-        # 检查是否需要重置
-        if provider.rpm_reset_at and provider.rpm_reset_at < datetime.now(timezone.utc):
-            provider.rpm_used = 0
-            provider.rpm_reset_at = datetime.fromtimestamp(current_time + 60, tz=timezone.utc)
-            self.db.commit()  # 立即提交事务,释放数据库锁
-
-        # 检查是否超限
-        if provider.rpm_used >= rpm_limit:
-            logger.warning(f"Provider {provider.name} RPM limit exceeded")
-            return False
-
-        # 递增计数
-        provider.rpm_used += 1
-        if not provider.rpm_reset_at:
-            provider.rpm_reset_at = datetime.fromtimestamp(current_time + 60, tz=timezone.utc)
-
-        self.db.commit()  # 立即提交事务,释放数据库锁
-        return True
-
-    def record_usage(
-        self, provider_id: str, success: bool, response_time_ms: float, cost_usd: float
-    ):
-        """记录使用情况到追踪表"""
-
-        # 获取当前分钟窗口
-        now = datetime.now(timezone.utc)
-        window_start = now.replace(second=0, microsecond=0)
-        window_end = window_start + timedelta(minutes=1)
-
-        # 查找或创建追踪记录
-        tracking = (
-            self.db.query(ProviderUsageTracking)
-            .filter(
-                ProviderUsageTracking.provider_id == provider_id,
-                ProviderUsageTracking.window_start == window_start,
-            )
-            .first()
-        )
-
-        if not tracking:
-            tracking = ProviderUsageTracking(
-                provider_id=provider_id, window_start=window_start, window_end=window_end
-            )
-            self.db.add(tracking)
-
-        # 更新统计
-        tracking.total_requests += 1
-        if success:
-            tracking.successful_requests += 1
-        else:
-            tracking.failed_requests += 1
-
-        tracking.total_response_time_ms += response_time_ms
-        tracking.avg_response_time_ms = tracking.total_response_time_ms / tracking.total_requests
-        tracking.total_cost_usd += cost_usd
-
-        self.db.flush()  # 只 flush，不立即 commit
-        # RPM 使用统计是非关键数据，使用批量提交
-        get_batch_committer().mark_dirty(self.db)
-
-        logger.debug(f"Recorded usage for provider {provider_id}")
-
-    def get_rpm_status(self, provider_id: str) -> Dict:
-        """获取提供商的RPM状态"""
-        provider = self.db.query(Provider).filter(Provider.id == provider_id).first()
-        if not provider:
-            return {"error": "Provider not found"}
-
-        return {
-            "provider_id": provider_id,
-            "provider_name": provider.name,
-            "rpm_limit": provider.rpm_limit,
-            "rpm_used": provider.rpm_used,
-            "rpm_reset_at": provider.rpm_reset_at.isoformat() if provider.rpm_reset_at else None,
-            "available": (
-                provider.rpm_limit - provider.rpm_used if provider.rpm_limit is not None else None
-            ),
-        }
-
-    def reset_rpm_counter(self, provider_id: str):
-        """手动重置RPM计数器"""
-        provider = self.db.query(Provider).filter(Provider.id == provider_id).first()
-        if provider:
-            provider.rpm_used = 0
-            provider.rpm_reset_at = None
-            self.db.commit()  # 立即提交事务,释放数据库锁
-
-            logger.info(f"Reset RPM counter for provider {provider.name}")
--- a/src/services/request/executor.py
+++ b/src/services/request/executor.py
@@ -89,24 +89,29 @@ class RequestExecutor:
        try:
            # 计算动态预留比例
            reservation_manager = get_adaptive_reservation_manager()
-            # 获取当前并发数用于计算负载
+            # 获取当前 RPM 计数用于计算负载
+            # 注意：key 侧返回的是 RPM 计数（不会在请求结束时减少，靠 TTL 过期）
            try:
-                _, current_key_concurrent = await self.concurrency_manager.get_current_concurrency(
+                _, current_key_rpm = await self.concurrency_manager.get_current_concurrency(
                    endpoint_id=endpoint.id,
                    key_id=key.id,
                )
            except Exception as e:
-                logger.debug(f"获取并发数失败（用于预留计算）: {e}")
-                current_key_concurrent = 0
+                logger.debug(f"获取 RPM 计数失败（用于预留计算）: {e}")
+                current_key_rpm = 0

-            # 获取有效的并发限制（自适应或固定）
-            effective_key_limit = (
-                key.learned_max_concurrent if key.max_concurrent is None else key.max_concurrent
-            )
+            # 获取有效的 RPM 限制（自适应或固定）
+            if key.rpm_limit is None:
+                # 自适应模式：优先使用学习值，否则使用默认初始限制，避免无限制打爆上游
+                from src.config.constants import RPMDefaults
+
+                effective_key_limit = int(key.learned_rpm_limit or RPMDefaults.INITIAL_LIMIT)
+            else:
+                effective_key_limit = int(key.rpm_limit)

            reservation_result = reservation_manager.calculate_reservation(
                key=key,
-                current_concurrent=current_key_concurrent,
+                current_usage=current_key_rpm,
                effective_limit=effective_key_limit,
            )
            dynamic_reservation_ratio = reservation_result.ratio
@@ -115,24 +120,22 @@ class RequestExecutor:
                f"ratio={dynamic_reservation_ratio:.0%}, phase={reservation_result.phase}, "
                f"confidence={reservation_result.confidence:.0%}")

-            async with self.concurrency_manager.concurrency_guard(
-                endpoint_id=endpoint.id,
-                endpoint_max_concurrent=endpoint.max_concurrent,
+            async with self.concurrency_manager.rpm_guard(
                key_id=key.id,
-                key_max_concurrent=effective_key_limit,
+                key_rpm_limit=effective_key_limit,
                is_cached_user=is_cached_user,
                cache_reservation_ratio=dynamic_reservation_ratio,
            ):
+                # 获取当前 RPM 计数（guard 内再次获取以获得最新值）
                try:
-                    _, key_concurrent = await self.concurrency_manager.get_current_concurrency(
-                        endpoint_id=endpoint.id,
+                    key_rpm_count = await self.concurrency_manager.get_key_rpm_count(
                        key_id=key.id,
                    )
                except Exception as e:
-                    logger.debug(f"获取并发数失败（guard 内）: {e}")
-                    key_concurrent = None
+                    logger.debug(f"获取 RPM 计数失败（guard 内）: {e}")
+                    key_rpm_count = None

-                context.concurrent_requests = key_concurrent
+                context.concurrent_requests = key_rpm_count  # 用于记录，实际是 RPM 计数
                context.start_time = time.time()

                response = await request_func(provider, endpoint, key)
@@ -142,15 +145,18 @@ class RequestExecutor:
                health_monitor.record_success(
                    db=self.db,
                    key_id=key.id,
+                    api_format=(
+                        api_format.value if isinstance(api_format, APIFormat) else api_format
+                    ),
                    response_time_ms=context.elapsed_ms,
                )

-                # 自适应模式：max_concurrent = NULL
-                if key.max_concurrent is None and key_concurrent is not None:
+                # 自适应模式：rpm_limit = NULL
+                if key.rpm_limit is None and key_rpm_count is not None:
                    self.adaptive_manager.handle_success(
                        db=self.db,
                        key=key,
-                        current_concurrent=key_concurrent,
+                        current_rpm=key_rpm_count,
                    )

                # 根据是否为流式请求，标记不同状态
@@ -162,7 +168,7 @@ class RequestExecutor:
                        db=self.db,
                        candidate_id=candidate_id,
                        status_code=200,
-                        concurrent_requests=key_concurrent,
+                        concurrent_requests=key_rpm_count,
                    )
                else:
                    # 非流式请求：标记为 success 状态
@@ -171,7 +177,7 @@ class RequestExecutor:
                        candidate_id=candidate_id,
                        status_code=200,
                        latency_ms=context.elapsed_ms,
-                        concurrent_requests=key_concurrent,
+                        concurrent_requests=key_rpm_count,
                        extra_data={
                            "is_cached_user": is_cached_user,
                            "model_name": model_name,
--- a/src/services/request/result.py
+++ b/src/services/request/result.py
@@ -289,10 +289,10 @@ class RequestResult:
            status_code = 500
            error_type = "internal_error"

-        # 构建错误消息：优先使用上游响应作为主要错误信息
-        if isinstance(exception, ProviderNotAvailableException) and exception.upstream_response:
-            error_message = exception.upstream_response
-        else:
+        # 构建错误消息：优先使用友好的 message 属性
+        # upstream_response 仅用于调试/链路追踪，不作为客户端错误消息
+        error_message = getattr(exception, "message", None)
+        if not error_message or not isinstance(error_message, str):
            error_message = str(exception)

        return cls(
--- a/src/services/usage/quota_scheduler.py
+++ b/src/services/usage/quota_scheduler.py
@@ -96,8 +96,6 @@ class QuotaScheduler:
                        logger.info(f"Resetting quota for provider {provider.name}")

                        provider.monthly_used_usd = 0.0
-                        provider.rpm_used = 0  # 同时重置RPM计数
-                        provider.rpm_reset_at = None
                        provider.quota_last_reset_at = now
                        reset_count += 1

@@ -126,8 +124,6 @@ class QuotaScheduler:
                provider = db.query(Provider).filter(Provider.id == provider_id).first()
                if provider and provider.billing_type == ProviderBillingType.MONTHLY_QUOTA:
                    provider.monthly_used_usd = 0.0
-                    provider.rpm_used = 0
-                    provider.rpm_reset_at = None
                    provider.quota_last_reset_at = now
                    db.commit()
                    logger.info(f"Force reset quota for provider {provider.name}")
@@ -140,8 +136,6 @@ class QuotaScheduler:
                )
                for provider in providers:
                    provider.monthly_used_usd = 0.0
-                    provider.rpm_used = 0
-                    provider.rpm_reset_at = None
                    provider.quota_last_reset_at = now
                db.commit()
                logger.info(f"Force reset quotas for {len(providers)} providers")
--- a/src/services/usage/recorder.py
+++ b/src/services/usage/recorder.py
@@ -93,6 +93,10 @@ class UsageRecorder:
        if metadata.original_model and metadata.original_model != metadata.model:
            target_model = metadata.model

+        # 非流式成功时，返回给客户端的是提供商响应头（透传）+ content-type
+        client_response_headers = dict(metadata.provider_response_headers) if metadata.provider_response_headers else {}
+        client_response_headers["content-type"] = "application/json"
+
        await UsageService.record_usage(
            db=self.db,
            user=self.user,
@@ -115,6 +119,7 @@ class UsageRecorder:
            request_body=request_body or result.request_body,
            provider_request_headers=metadata.provider_request_headers,
            response_headers=metadata.provider_response_headers,
+            client_response_headers=client_response_headers,
            response_body=result.response_data if isinstance(result.response_data, dict) else {},
            request_id=self.request_id,
            provider_id=metadata.provider_id,
@@ -181,6 +186,8 @@ class UsageRecorder:
            request_body=request_body or result.request_body,
            provider_request_headers=metadata.provider_request_headers,
            response_headers={},
+            # 失败请求返回给客户端的是 JSON 错误响应
+            client_response_headers={"content-type": "application/json"},
            response_body={"error": result.error_message} if result.error_message else {},
            request_id=self.request_id,
            provider_id=metadata.provider_id,
--- a/src/services/usage/service.py
+++ b/src/services/usage/service.py
@@ -40,6 +40,7 @@ class UsageRecordParams:
    request_body: Optional[Any]
    provider_request_headers: Optional[Dict[str, Any]]
    response_headers: Optional[Dict[str, Any]]
+    client_response_headers: Optional[Dict[str, Any]]
    response_body: Optional[Any]
    request_id: str
    provider_id: Optional[str]
@@ -223,6 +224,7 @@ class UsageService:
        request_body: Optional[Any],
        provider_request_headers: Optional[Dict[str, Any]],
        response_headers: Optional[Dict[str, Any]],
+        client_response_headers: Optional[Dict[str, Any]],
        response_body: Optional[Any],
        request_id: str,
        provider_id: Optional[str],
@@ -288,6 +290,13 @@ class UsageService:
                db, response_headers
            )

+        # 处理返回给客户端的响应头
+        processed_client_response_headers = None
+        if should_log_headers and client_response_headers:
+            processed_client_response_headers = SystemConfigService.mask_sensitive_headers(
+                db, client_response_headers
+            )
+
        # 计算真实成本（表面成本 * 倍率），免费套餐实际费用为 0
        if is_free_tier:
            actual_input_cost = 0.0
@@ -351,6 +360,7 @@ class UsageService:
            "request_body": processed_request_body,
            "provider_request_headers": processed_provider_request_headers,
            "response_headers": processed_response_headers,
+            "client_response_headers": processed_client_response_headers,
            "response_body": processed_response_body,
        }

@@ -360,12 +370,13 @@ class UsageService:
        db: Session,
        provider_api_key_id: Optional[str],
        provider_id: Optional[str],
+        api_format: Optional[str] = None,
    ) -> Tuple[float, bool]:
        """获取费率倍数和是否免费套餐（使用缓存）"""
        from src.services.cache.provider_cache import ProviderCacheService

        return await ProviderCacheService.get_rate_multiplier_and_free_tier(
-            db, provider_api_key_id, provider_id
+            db, provider_api_key_id, provider_id, api_format
        )

    @classmethod
@@ -484,6 +495,7 @@ class UsageService:
            existing_usage.provider_request_headers = usage_params["provider_request_headers"]
        existing_usage.response_body = usage_params["response_body"]
        existing_usage.response_headers = usage_params["response_headers"]
+        existing_usage.client_response_headers = usage_params["client_response_headers"]

        # 更新 token 和费用信息
        existing_usage.input_tokens = usage_params["input_tokens"]
@@ -656,9 +668,9 @@ class UsageService:
        Returns:
            (usage_params 字典, total_cost 总成本)
        """
-        # 获取费率倍数和是否免费套餐
+        # 获取费率倍数和是否免费套餐（传递 api_format 支持按格式配置的倍率）
        actual_rate_multiplier, is_free_tier = await cls._get_rate_multiplier_and_free_tier(
-            params.db, params.provider_api_key_id, params.provider_id
+            params.db, params.provider_api_key_id, params.provider_id, params.api_format
        )

        # 计算成本
@@ -704,6 +716,7 @@ class UsageService:
            request_body=params.request_body,
            provider_request_headers=params.provider_request_headers,
            response_headers=params.response_headers,
+            client_response_headers=params.client_response_headers,
            response_body=params.response_body,
            request_id=params.request_id,
            provider_id=params.provider_id,
@@ -753,6 +766,7 @@ class UsageService:
        request_body: Optional[Any] = None,
        provider_request_headers: Optional[Dict[str, Any]] = None,
        response_headers: Optional[Dict[str, Any]] = None,
+        client_response_headers: Optional[Dict[str, Any]] = None,
        response_body: Optional[Any] = None,
        request_id: Optional[str] = None,
        provider_id: Optional[str] = None,
@@ -785,7 +799,8 @@ class UsageService:
            status_code=status_code, error_message=error_message, metadata=metadata,
            request_headers=request_headers, request_body=request_body,
            provider_request_headers=provider_request_headers,
-            response_headers=response_headers, response_body=response_body,
+            response_headers=response_headers, client_response_headers=client_response_headers,
+            response_body=response_body,
            request_id=request_id, provider_id=provider_id,
            provider_endpoint_id=provider_endpoint_id,
            provider_api_key_id=provider_api_key_id, status=status,
@@ -844,6 +859,7 @@ class UsageService:
        request_body: Optional[Any] = None,
        provider_request_headers: Optional[Dict[str, Any]] = None,
        response_headers: Optional[Dict[str, Any]] = None,
+        client_response_headers: Optional[Dict[str, Any]] = None,
        response_body: Optional[Any] = None,
        request_id: Optional[str] = None,
        provider_id: Optional[str] = None,
@@ -878,7 +894,8 @@ class UsageService:
            status_code=status_code, error_message=error_message, metadata=metadata,
            request_headers=request_headers, request_body=request_body,
            provider_request_headers=provider_request_headers,
-            response_headers=response_headers, response_body=response_body,
+            response_headers=response_headers, client_response_headers=client_response_headers,
+            response_body=response_body,
            request_id=request_id, provider_id=provider_id,
            provider_endpoint_id=provider_endpoint_id,
            provider_api_key_id=provider_api_key_id, status=status,
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
fawney19	7faca5512a	feat(ui): 优化密钥添加和仪表盘空状态体验 - KeyFormDialog: 添加模式下保存后不关闭对话框，清除表单以便连续添加 - KeyFormDialog: 按钮文案根据编辑/添加模式动态显示 - Dashboard: 优化统计卡片加载状态和空数据占位显示	2026-01-10 19:32:36 +08:00
fawney19	ad84272084	fix: 修复普通用户无法访问仪表盘接口的权限问题将 DashboardAdapter 的 mode 从 ApiMode.ADMIN 改为 ApiMode.USER，允许普通用户访问 /api/dashboard/stats 和 /api/dashboard/daily-stats 接口。	2026-01-10 19:31:19 +08:00
fawney19	09e0f594ff	refactor: 重构限流系统和健康监控，支持按 API 格式区分 - 将 adaptive_concurrency 重命名为 adaptive_rpm，从并发控制改为 RPM 控制 - 健康监控器支持按 API 格式独立管理健康度和熔断器状态 - 新增 model_permissions 模块，支持按格式配置允许的模型 - 重构前端提供商相关表单组件，新增 Collapsible UI 组件 - 新增数据库迁移脚本支持新的数据结构	2026-01-10 18:48:35 +08:00
fawney19	dd2fbf4424	style(ui): 调整模型详情抽屉关联提供商表格列宽	2026-01-08 13:37:41 +08:00
fawney19	99b12a49c6	Merge pull request #78 from fawney19/perf/optimize perf: 优化 HTTP 客户端连接池复用	2026-01-08 13:37:13 +08:00
fawney19	ea35efe440	perf: 优化 HTTP 客户端连接池复用 - 新增 get_proxy_client() 方法，相同代理配置复用同一客户端 - 添加 LRU 淘汰策略，代理客户端上限 50 个防止内存泄漏 - 新增 get_default_client_async() 异步线程安全版本 - 使用模块级锁避免类属性初始化竞态条件 - 优化 ConcurrencyManager 使用 Redis MGET 批量获取减少往返 - 添加 get_pool_stats() 连接池统计信息接口	2026-01-08 13:34:59 +08:00
fawney19	bf09e740e9	fix(ui): 优化提供商详情页的交互体验 - 模型列表删除按钮仅在 hover 时显示红色 - 批量关联模型对话框：只有全局模型时展开，有多个分组时全部折叠	2026-01-08 11:25:52 +08:00
fawney19	60c77cec56	Merge pull request #77 from AAEE86/ui style(ui): improve text visibility in dark mode for model badges	2026-01-08 10:52:54 +08:00
fawney19	0e4a1dddb5	refactor(ui): 优化批量端点创建的 UI 和性能 - 调整布局: API URL 移至顶部, API 格式选择移至下方 - 优化 checkbox 样式: 使用自定义勾选框替代原生样式 - API 格式按列排序: 基础格式和对应 CLI 格式上下对齐 - 请求配置改为 4 列布局, 更紧凑 - 使用 Promise.allSettled 并发创建端点, 提升性能 - 改进错误提示: 失败时直接展示具体错误信息给用户 - 清理未使用的 Select 组件导入和 selectOpen 变量	2026-01-08 10:50:25 +08:00
AAEE86	1cf18b6e12	feat(ui): support batch endpoint creation with multiple API formats (#76 ) Replace single API format selector with multi-select checkbox interface in endpoint creation dialog. Users can now select multiple API formats to create multiple endpoints simultaneously with shared configuration (URL, path, timeout, etc.). - Change API format selection from dropdown to checkbox grid layout - Add selectedFormats array to track multiple format selections - Implement batch creation logic with individual error handling - Update submit button to show endpoint count being created - Adjust form layout to improve visual hierarchy - Display appropriate success/failure messages for batch operations - Reset selectedFormats on form reset	2026-01-08 10:42:14 +08:00
AAEE86	f9a8be898a	style(ui): improve text visibility in dark mode for model badges	2026-01-08 10:26:58 +08:00