Merge upstream/main into feat/one-click-update

.env.example

@@ -85,12 +85,12 @@ ADMIN_USERNAME=admin123456
 #   MAX = CPU 核数 × 10（AI 网关偏 IO 等待，可激进些；纯 OLTP 用 × 4）
 #   MIN = MAX × 0.2（保留常驻连接应对突发流量，避免冷启动握手开销）
 # 多实例部署时请按 实例数 × MAX 控制总和，PG 端 max_connections 至少为该总和 + 20 余量
-# AETHER_GATEWAY_DATA_POSTGRES_MIN_CONNECTIONS=1
+# AETHER_GATEWAY_DATA_POSTGRES_MIN_CONNECTIONS=4
 # AETHER_GATEWAY_DATA_POSTGRES_MAX_CONNECTIONS=20
 # AETHER_GATEWAY_DATA_POSTGRES_STATEMENT_CACHE_CAPACITY=100
 # AETHER_GATEWAY_DATA_POSTGRES_ACQUIRE_TIMEOUT_MS=3000
 
-# PostgreSQL 性能调优（默认值适合 2核2GB 的最小机器，按实际配置覆盖）
+# PostgreSQL 性能调优（默认值适合 2核4GB 机器，按实际配置覆盖）
 # 参考：shared_buffers ≈ 可用内存 25%，effective_cache_size ≈ 可用内存 50-75%
 # work_mem 是每个连接每个排序操作的内存，不要设太大（并发数 × work_mem 是实际占用）
 # | 系统内存 | shared_buffers | effective_cache_size | work_mem |
@@ -99,7 +99,7 @@ ADMIN_USERNAME=admin123456
 # | 8GB      | 2GB            | 6GB                  | 16MB     |
 # | 16GB     | 4GB            | 12GB                 | 32MB     |
 # | 32GB+    | 8GB            | 24GB                 | 32MB     |
-# POSTGRES_SHARED_BUFFERS=256MB
-# POSTGRES_EFFECTIVE_CACHE_SIZE=768MB
-# POSTGRES_WORK_MEM=4MB
-# POSTGRES_MAINTENANCE_WORK_MEM=64MB
+# POSTGRES_SHARED_BUFFERS=1GB
+# POSTGRES_EFFECTIVE_CACHE_SIZE=3GB
+# POSTGRES_WORK_MEM=16MB
+# POSTGRES_MAINTENANCE_WORK_MEM=256MB

Cargo.lock

@@ -859,6 +859,15 @@ dependencies = [
  "tracing",
 ]
 
+[[package]]
+name = "backon"
+version = "1.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cffb0e931875b666fc4fcb20fee52e9bbd1ef836fd9e9e04ec21555f9f85f7ef"
+dependencies = [
+ "fastrand",
+]
+
 [[package]]
 name = "base64"
 version = "0.21.7"
@@ -1638,6 +1647,12 @@ dependencies = [
  "regex",
 ]
 
+[[package]]
+name = "fastrand"
+version = "2.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9f1f227452a390804cdb637b74a86990f2a7d7ba4b7d5693aac9b4dd6defd8d6"
+
 [[package]]
 name = "filedescriptor"
 version = "0.8.3"
@@ -3390,8 +3405,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e37ec3fd44bea2ec947ba6cc7634d7999a6590aca7c35827c250bc0de502bda6"
 dependencies = [
  "arc-swap",
+ "backon",
  "bytes",
  "combine",
+ "futures-channel",
  "futures-util",
  "itoa",
  "num-bigint",

Cargo.toml

@@ -82,7 +82,7 @@ hmac = "0.12"
 http = "1"
 pbkdf2 = { version = "0.12", default-features = false, features = ["hmac"] }
 reqwest = { version = "0.12", default-features = false, features = ["json", "stream", "rustls-tls", "http2", "socks"] }
-redis = { version = "0.28", default-features = false, features = ["tokio-comp", "script", "streams"] }
+redis = { version = "0.28", default-features = false, features = ["tokio-comp", "script", "streams", "connection-manager"] }
 regex = "1"
 rustls = { version = "0.23", features = ["ring"] }
 semver = "1"

Makefile

@@ -335,6 +335,11 @@ if ! command -v curl >/dev/null 2>&1; then
 	exit 1
 fi
 
+if [ -z "$${RUSTC_WRAPPER:-}" ] && command -v sccache >/dev/null 2>&1; then
+	export RUSTC_WRAPPER="$$(command -v sccache)"
+	echo "=> 启用 Rust 编译缓存: $${RUSTC_WRAPPER}"
+fi
+
 if ! ensure_dev_infra; then
 	exit 1
 fi

apps/aether-gateway/src/admin_api.rs

@@ -7,10 +7,11 @@ pub(crate) use crate::handlers::admin::{
     provider_quota_refresh_endpoint_for_provider, provider_type_supports_quota_refresh,
     reconcile_admin_fixed_provider_template_endpoints,
     refresh_provider_oauth_account_state_after_update, refresh_provider_pool_quota_locally,
-    update_existing_provider_oauth_catalog_key, AdminAppState,
-    AdminGatewayProviderTransportSnapshot, AdminLocalOAuthRefreshError, AdminRequestContext,
-    AdminRouteRequest, AdminRouteResponse, AdminRouteResult, AdminStatsTimeRange,
-    AdminStatsUsageFilter, OAUTH_ACCOUNT_BLOCK_PREFIX, OAUTH_REQUEST_FAILED_PREFIX,
+    store_admin_provider_ops_balance_cache, update_existing_provider_oauth_catalog_key,
+    AdminAppState, AdminGatewayProviderTransportSnapshot, AdminLocalOAuthRefreshError,
+    AdminRequestContext, AdminRouteRequest, AdminRouteResponse, AdminRouteResult,
+    AdminStatsTimeRange, AdminStatsUsageFilter, OAUTH_ACCOUNT_BLOCK_PREFIX,
+    OAUTH_REQUEST_FAILED_PREFIX,
 };
 
 use crate::handlers::admin::{

apps/aether-gateway/src/ai_serving/api.rs

@@ -44,11 +44,11 @@ pub(crate) use aether_ai_formats::api::{
     build_core_error_body_for_client_format, convert_standard_chat_response,
     core_error_background_report_kind, core_error_default_client_api_format,
     core_success_background_report_kind, encode_kiro_sse_events,
-    implicit_sync_finalize_report_kind, is_core_error_finalize_kind,
-    normalize_provider_private_report_context, normalize_provider_private_response_value,
-    provider_private_response_allows_sync_finalize, resolve_claude_stream_spec,
-    resolve_claude_sync_spec, resolve_gemini_stream_spec, resolve_gemini_sync_spec,
-    resolve_local_image_stream_spec, resolve_local_image_sync_spec,
+    extract_provider_private_stream_error_body, implicit_sync_finalize_report_kind,
+    is_core_error_finalize_kind, normalize_provider_private_report_context,
+    normalize_provider_private_response_value, provider_private_response_allows_sync_finalize,
+    resolve_claude_stream_spec, resolve_claude_sync_spec, resolve_gemini_stream_spec,
+    resolve_gemini_sync_spec, resolve_local_image_stream_spec, resolve_local_image_sync_spec,
     resolve_local_same_format_stream_spec, resolve_local_same_format_sync_spec,
     resolve_openai_embedding_sync_spec, sanitize_request_path_and_query, AiControlPlanRequest,
     CanonicalContentPart, CanonicalStreamEvent, CanonicalStreamFrame, ClaudeClientEmitter,

apps/aether-gateway/src/ai_serving/planner/specialized/image/request.rs

@@ -19,9 +19,10 @@ use crate::ai_serving::{
     apply_codex_openai_responses_special_body_edits, apply_codex_openai_responses_special_headers,
     build_chatgpt_web_image_request_body,
     build_gemini_image_request_body_from_openai_image_request,
-    build_openai_image_provider_request_body, default_model_for_openai_image_operation,
-    normalize_openai_image_request, request_conversion_direct_auth, CandidateFailureDiagnostic,
-    GatewayProviderTransportSnapshot, PlannerAppState, RequestConversionKind,
+    build_openai_image_api_provider_request_body, build_openai_image_provider_request_body,
+    default_model_for_openai_image_operation, normalize_openai_image_request,
+    request_conversion_direct_auth, CandidateFailureDiagnostic, GatewayProviderTransportSnapshot,
+    PlannerAppState, RequestConversionKind,
 };
 use crate::image_capabilities::openai_image_normalize_options_for_provider;
 use crate::AppState;
@@ -160,21 +161,31 @@ pub(super) async fn resolve_local_openai_image_candidate_payload_parts(
         .provider_type
         .trim()
         .eq_ignore_ascii_case("grok");
+    let is_codex = transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("codex");
     let transport_profile = crate::ai_serving::transport::resolve_transport_profile(transport);
     let upstream_url = if is_chatgpt_web {
         chatgpt_web_image_internal_url(&transport.endpoint.base_url)
     } else if is_grok {
         build_grok_upstream_url(transport, GROK_CHAT_PATH)
     } else {
-        build_openai_image_upstream_url(transport, parts.uri.query())
+        build_openai_image_upstream_url(transport, Some(parts.uri.path()), parts.uri.query())
     };
     let mut provider_request_body = if is_chatgpt_web {
         match build_chatgpt_web_image_request_body(parts, body_json, body_base64) {
             Ok(body) => body,
             Err(err) => err.to_error_json(),
         }
-    } else {
+    } else if is_codex || is_grok {
         build_openai_image_provider_request_body(&normalized_request)
+    } else {
+        build_openai_image_api_provider_request_body(
+            &normalized_request,
+            Some(prepared_candidate.mapped_model.as_str()),
+        )
     };
     if !is_chatgpt_web {
         apply_codex_openai_responses_special_body_edits(

apps/aether-gateway/src/ai_serving/planner/standard/codex/tests.rs

@@ -85,7 +85,7 @@ fn injects_stable_prompt_cache_key_for_codex_requests() {
 
     assert_eq!(
         body["prompt_cache_key"],
-        "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        "53363264-dbb0-5f9d-b9c7-3e92c45c5bdf"
     );
 }
 

apps/aether-gateway/src/ai_serving/planner/standard/family/request.rs

@@ -24,9 +24,13 @@ use crate::ai_serving::transport::kiro::{
 use crate::ai_serving::transport::{
     build_grok_browser_headers, build_grok_upstream_url, build_kiro_cross_format_upstream_url,
     build_openai_image_headers, build_openai_image_upstream_url,
-    build_standard_provider_request_headers, openai_image_transport_unsupported_reason,
-    resolve_grok_session_auth, resolve_openai_image_auth, GrokHeaderInput,
-    ProviderOpenAiImageHeadersInput, StandardProviderRequestHeadersInput, GROK_CHAT_PATH,
+    build_standard_provider_request_headers, build_windsurf_cascade_headers,
+    build_windsurf_cascade_request_body, build_windsurf_cascade_upstream_url,
+    is_windsurf_provider_transport,
+    local_windsurf_request_transport_unsupported_reason_with_network,
+    openai_image_transport_unsupported_reason, resolve_grok_session_auth,
+    resolve_openai_image_auth, GrokHeaderInput, ProviderOpenAiImageHeadersInput,
+    StandardProviderRequestHeadersInput, GROK_CHAT_PATH, WINDSURF_ENVELOPE_NAME,
 };
 use crate::ai_serving::{
     build_openai_image_request_body_from_gemini_image_request, gemini_request_is_image_generation,
@@ -61,6 +65,134 @@ fn is_grok_text_provider_api_format(provider_api_format: &str) -> bool {
     )
 }
 
+fn provider_preserves_claude_thinking_signatures(provider_type: &str, base_url: &str) -> bool {
+    let provider_type = provider_type.trim().to_ascii_lowercase();
+    let base_url = base_url.trim().to_ascii_lowercase();
+    let is_bedrock_runtime_url = base_url.contains("bedrock-runtime")
+        && (base_url.contains("amazonaws.com")
+            || base_url.contains("amazonaws.com.cn")
+            || base_url.contains("api.aws"));
+    matches!(
+        provider_type.as_str(),
+        "anthropic" | "claude_code" | "bedrock" | "aws_bedrock" | "amazon_bedrock"
+    ) || base_url.contains("api.anthropic.com")
+        || is_bedrock_runtime_url
+}
+
+fn sanitize_claude_thinking_block(block: Value) -> (Option<Value>, bool) {
+    let Some(object) = block.as_object() else {
+        return (Some(block), false);
+    };
+    let block_type = object
+        .get("type")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .unwrap_or_default();
+
+    match block_type {
+        "thinking" => {
+            let thinking_text = object
+                .get("thinking")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .unwrap_or_default();
+            if thinking_text.is_empty() {
+                (None, true)
+            } else {
+                (
+                    Some(serde_json::json!({
+                        "type": "text",
+                        "text": thinking_text,
+                    })),
+                    true,
+                )
+            }
+        }
+        "redacted_thinking" => (None, true),
+        _ => (Some(block), false),
+    }
+}
+
+fn sanitize_claude_message_content_for_non_native_thinking(content: &mut Value) -> bool {
+    const OMITTED_THINKING_TEXT: &str = "Previous thinking omitted.";
+
+    if content.is_object() {
+        let original = std::mem::take(content);
+        let (sanitized, changed) = sanitize_claude_thinking_block(original);
+        if changed {
+            *content = sanitized.unwrap_or_else(|| {
+                serde_json::json!({
+                    "type": "text",
+                    "text": OMITTED_THINKING_TEXT,
+                })
+            });
+        }
+        return changed;
+    }
+
+    let Some(blocks) = content.as_array_mut() else {
+        return false;
+    };
+    let original_blocks = std::mem::take(blocks);
+    let mut changed = false;
+    let mut sanitized_blocks = Vec::with_capacity(original_blocks.len());
+    for block in original_blocks {
+        let (sanitized, block_changed) = sanitize_claude_thinking_block(block);
+        changed |= block_changed;
+        if let Some(sanitized) = sanitized {
+            sanitized_blocks.push(sanitized);
+        }
+    }
+    if changed && sanitized_blocks.is_empty() {
+        sanitized_blocks.push(serde_json::json!({
+            "type": "text",
+            "text": OMITTED_THINKING_TEXT,
+        }));
+    }
+    *blocks = sanitized_blocks;
+    changed
+}
+
+fn sanitize_claude_request_thinking_signatures_for_non_native(body_json: &mut Value) -> bool {
+    body_json
+        .get_mut("messages")
+        .and_then(Value::as_array_mut)
+        .map(|messages| {
+            messages.iter_mut().fold(false, |changed, message| {
+                let is_assistant = message
+                    .get("role")
+                    .and_then(Value::as_str)
+                    .is_some_and(|role| role.trim().eq_ignore_ascii_case("assistant"));
+                if !is_assistant {
+                    return changed;
+                }
+                let content_changed = message
+                    .get_mut("content")
+                    .is_some_and(sanitize_claude_message_content_for_non_native_thinking);
+                changed || content_changed
+            })
+        })
+        .unwrap_or(false)
+}
+
+fn apply_non_native_claude_thinking_signature_compat(
+    provider_request_body: &mut Value,
+    provider_api_format: &str,
+    transport: &GatewayProviderTransportSnapshot,
+) {
+    if crate::ai_serving::normalize_api_format_alias(provider_api_format) != "claude:messages" {
+        return;
+    }
+    if provider_preserves_claude_thinking_signatures(
+        transport.provider.provider_type.as_str(),
+        transport.endpoint.base_url.as_str(),
+    ) {
+        return;
+    }
+
+    let _ = sanitize_claude_request_thinking_signatures_for_non_native(provider_request_body);
+}
+
 pub(crate) async fn resolve_local_standard_candidate_payload_parts(
     state: &AppState,
     parts: &http::request::Parts,
@@ -198,11 +330,18 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
         return None;
     }
 
-    if let Some(skip_reason) = crate::ai_serving::request_pair_transport_unsupported_reason(
-        transport,
-        spec_metadata.api_format,
-        provider_api_format,
-    ) {
+    let is_windsurf_cascade =
+        provider_api_format == "openai:chat" && is_windsurf_provider_transport(transport);
+    let transport_unsupported_reason = if is_windsurf_cascade {
+        local_windsurf_request_transport_unsupported_reason_with_network(transport)
+    } else {
+        crate::ai_serving::request_pair_transport_unsupported_reason(
+            transport,
+            spec_metadata.api_format,
+            provider_api_format,
+        )
+    };
+    if let Some(skip_reason) = transport_unsupported_reason {
         mark_skipped_local_standard_candidate(
             state,
             input,
@@ -321,7 +460,7 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
             provider_api_format,
             parts.uri.path(),
             upstream_is_stream,
-            if is_kiro_claude_cli {
+            if is_kiro_claude_cli || is_windsurf_cascade {
                 None
             } else {
                 transport.endpoint.body_rules.as_ref()
@@ -379,6 +518,11 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
         .await;
         return None;
     }
+    apply_non_native_claude_thinking_signature_compat(
+        &mut provider_request_body,
+        provider_api_format,
+        transport,
+    );
     if let Some(mapping) =
         crate::system_features::reasoning_model_directive_mapping_for_api_format_and_model(
             state,
@@ -422,6 +566,11 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
             .await;
             return None;
         }
+        apply_non_native_claude_thinking_signature_compat(
+            &mut provider_request_body,
+            provider_api_format,
+            transport,
+        );
     }
 
     if let Some(kiro_auth) = kiro_auth.as_ref() {
@@ -443,6 +592,24 @@ pub(crate) async fn resolve_local_standard_candidate_payload_parts(
         )
         .await;
     }
+    if is_windsurf_cascade {
+        return build_windsurf_cross_format_payload_parts(
+            state,
+            parts,
+            trace_id,
+            body_json,
+            input,
+            attempt,
+            transport,
+            provider_api_format,
+            prepared_candidate.mapped_model,
+            prepared_candidate.auth_header,
+            prepared_candidate.auth_value,
+            provider_request_body,
+            upstream_is_stream,
+        )
+        .await;
+    }
 
     let upstream_url = match crate::ai_serving::planner::standard::build_standard_upstream_url(
         parts,
@@ -542,6 +709,121 @@ fn apply_transport_request_body_semantics(
     )
 }
 
+#[allow(clippy::too_many_arguments)]
+async fn build_windsurf_cross_format_payload_parts(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    original_body_json: &serde_json::Value,
+    input: &LocalStandardDecisionInput,
+    attempt: &LocalStandardCandidateAttempt,
+    transport: &Arc<GatewayProviderTransportSnapshot>,
+    provider_api_format: &str,
+    mapped_model: String,
+    auth_header: String,
+    auth_value: String,
+    openai_chat_request_body: Value,
+    upstream_is_stream: bool,
+) -> Option<LocalStandardCandidatePayloadParts> {
+    let candidate = &attempt.eligible.candidate;
+    let effective_headers = input.effective_headers(&parts.headers);
+    let provider_request_body = match build_windsurf_cascade_request_body(
+        &openai_chat_request_body,
+        &mapped_model,
+        &auth_value,
+        transport.endpoint.body_rules.as_ref(),
+        Some(effective_headers),
+        upstream_is_stream,
+    ) {
+        Some(body) => body,
+        None => {
+            mark_skipped_local_standard_candidate_with_extra_data(
+                state,
+                input,
+                trace_id,
+                candidate,
+                attempt.candidate_index,
+                &attempt.candidate_id,
+                "provider_request_body_build_failed",
+                request_body_build_failure_extra_data(
+                    &openai_chat_request_body,
+                    provider_api_format,
+                    provider_api_format,
+                ),
+            )
+            .await;
+            return None;
+        }
+    };
+    let upstream_url = match build_windsurf_cascade_upstream_url(
+        transport.endpoint.base_url.as_str(),
+        parts.uri.query(),
+    ) {
+        Some(url) => url,
+        None => {
+            mark_skipped_local_standard_candidate_with_failure_diagnostic(
+                state,
+                input,
+                trace_id,
+                candidate,
+                attempt.candidate_index,
+                &attempt.candidate_id,
+                "upstream_url_missing",
+                CandidateFailureDiagnostic::upstream_url_missing(
+                    provider_api_format,
+                    provider_api_format,
+                    "standard_family_windsurf_url",
+                ),
+            )
+            .await;
+            return None;
+        }
+    };
+    let provider_request_headers = match build_windsurf_cascade_headers(
+        effective_headers,
+        &provider_request_body,
+        original_body_json,
+        transport.endpoint.header_rules.as_ref(),
+        &auth_header,
+        &auth_value,
+        upstream_is_stream,
+    ) {
+        Some(headers) => headers,
+        None => {
+            mark_skipped_local_standard_candidate_with_failure_diagnostic(
+                state,
+                input,
+                trace_id,
+                candidate,
+                attempt.candidate_index,
+                &attempt.candidate_id,
+                "transport_header_rules_apply_failed",
+                CandidateFailureDiagnostic::header_rules_apply_failed(
+                    provider_api_format,
+                    provider_api_format,
+                    "standard_family_windsurf_headers",
+                ),
+            )
+            .await;
+            return None;
+        }
+    };
+
+    Some(LocalStandardCandidatePayloadParts {
+        auth_header,
+        auth_value,
+        mapped_model,
+        provider_api_format: provider_api_format.to_string(),
+        provider_request_body,
+        provider_request_headers,
+        upstream_url,
+        upstream_is_stream,
+        envelope_name: Some(WINDSURF_ENVELOPE_NAME),
+        transport: Arc::clone(transport),
+        transport_profile: None,
+    })
+}
+
 async fn resolve_local_gemini_image_to_openai_image_candidate_payload_parts(
     state: &AppState,
     parts: &http::request::Parts,
@@ -625,7 +907,8 @@ async fn resolve_local_gemini_image_to_openai_image_candidate_payload_parts(
     };
 
     let upstream_is_stream = true;
-    let upstream_url = build_openai_image_upstream_url(transport, None);
+    let upstream_url =
+        build_openai_image_upstream_url(transport, Some("/v1/images/generations"), None);
     let effective_headers = input.effective_headers(&parts.headers);
     let Some(mut provider_request_headers) =
         build_openai_image_headers(ProviderOpenAiImageHeadersInput {
@@ -798,3 +1081,95 @@ async fn build_kiro_cross_format_payload_parts(
         transport_profile: None,
     })
 }
+
+#[cfg(test)]
+mod tests {
+    use super::{
+        provider_preserves_claude_thinking_signatures,
+        sanitize_claude_request_thinking_signatures_for_non_native,
+    };
+    use serde_json::json;
+
+    #[test]
+    fn sanitizes_historical_claude_thinking_for_non_native_relays() {
+        let mut body = json!({
+            "model": "claude-opus-4-1",
+            "messages": [{
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "thinking",
+                        "thinking": "I should keep this short.",
+                        "signature": "sig_123"
+                    },
+                    {
+                        "type": "redacted_thinking",
+                        "data": "opaque"
+                    },
+                    {
+                        "type": "text",
+                        "text": "Done."
+                    }
+                ]
+            }]
+        });
+
+        assert!(sanitize_claude_request_thinking_signatures_for_non_native(
+            &mut body
+        ));
+        assert_eq!(body["messages"][0]["content"][0]["type"], json!("text"));
+        assert_eq!(
+            body["messages"][0]["content"][0]["text"],
+            json!("I should keep this short.")
+        );
+        assert_eq!(body["messages"][0]["content"].as_array().unwrap().len(), 2);
+        assert_eq!(body["messages"][0]["content"][1]["text"], json!("Done."));
+    }
+
+    #[test]
+    fn inserts_placeholder_when_only_redacted_thinking_would_remain() {
+        let mut body = json!({
+            "model": "claude-opus-4-1",
+            "messages": [{
+                "role": "assistant",
+                "content": [{
+                    "type": "redacted_thinking",
+                    "data": "opaque"
+                }]
+            }]
+        });
+
+        assert!(sanitize_claude_request_thinking_signatures_for_non_native(
+            &mut body
+        ));
+        assert_eq!(body["messages"][0]["content"][0]["type"], json!("text"));
+        assert_eq!(
+            body["messages"][0]["content"][0]["text"],
+            json!("Previous thinking omitted.")
+        );
+    }
+
+    #[test]
+    fn official_claude_providers_preserve_thinking_signatures() {
+        assert!(provider_preserves_claude_thinking_signatures(
+            "anthropic",
+            "https://relay.example.com"
+        ));
+        assert!(provider_preserves_claude_thinking_signatures(
+            "custom",
+            "https://api.anthropic.com"
+        ));
+        assert!(provider_preserves_claude_thinking_signatures(
+            "aws",
+            "https://bedrock-runtime.us-east-1.amazonaws.com"
+        ));
+        assert!(provider_preserves_claude_thinking_signatures(
+            "amazon_bedrock",
+            "https://relay.example.com"
+        ));
+        assert!(!provider_preserves_claude_thinking_signatures(
+            "openai",
+            "https://relay.example.com"
+        ));
+    }
+}

apps/aether-gateway/src/ai_serving/planner/standard/normalize/tests.rs

@@ -209,7 +209,7 @@ fn local_openai_responses_compact_wrapper_strips_include_for_codex_requests() {
     assert_eq!(provider_request_body["instructions"], "");
     assert_eq!(
         provider_request_body["prompt_cache_key"],
-        "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        "3d2e2842-74cb-55dd-803a-b8940b3500c2"
     );
 }
 
@@ -355,7 +355,7 @@ fn injects_codex_prompt_cache_key_for_openai_responses_cross_format_requests() {
 
     assert_eq!(
         provider_request_body["prompt_cache_key"],
-        "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        "b4dfeb75-b105-544c-a706-39b92f0bddb0"
     );
 }
 
@@ -385,6 +385,6 @@ fn injects_codex_prompt_cache_key_for_openai_chat_cross_format_requests() {
 
     assert_eq!(
         provider_request_body["prompt_cache_key"],
-        "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        "4ee6ea6e-3ac6-5a18-8cb8-1f8b956419e5"
     );
 }

apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision.rs

@@ -9,7 +9,6 @@ pub(super) use self::payload::maybe_build_local_openai_chat_decision_payload_for
 pub(super) use self::support::{
     build_lazy_local_openai_chat_candidate_attempt_source,
     build_local_openai_chat_candidate_attempt_source,
-    build_local_openai_chat_image_candidate_attempt_source,
     materialize_local_openai_chat_candidate_attempts, LocalOpenAiChatCandidateAttempt,
     LocalOpenAiChatCandidateAttemptSource, LocalOpenAiChatDecisionInput,
 };

apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision/request.rs

@@ -28,6 +28,12 @@ use crate::ai_serving::transport::kiro::{
     KIRO_ENVELOPE_NAME,
 };
 use crate::ai_serving::transport::local_openai_chat_transport_unsupported_reason;
+use crate::ai_serving::transport::windsurf::{
+    build_windsurf_cascade_headers, build_windsurf_cascade_request_body,
+    build_windsurf_cascade_upstream_url, is_windsurf_provider_transport,
+    local_windsurf_request_transport_unsupported_reason_with_network,
+    resolve_windsurf_cascade_auth, WINDSURF_ENVELOPE_NAME,
+};
 use crate::ai_serving::transport::{
     build_grok_browser_headers, build_grok_upstream_url, build_kiro_cross_format_upstream_url,
     build_openai_image_headers, build_openai_image_upstream_url,
@@ -334,6 +340,25 @@ pub(crate) async fn resolve_local_openai_chat_candidate_payload_parts(
         }));
     }
 
+    if provider_api_format == "openai:chat" && is_windsurf_provider_transport(transport) {
+        return build_windsurf_openai_chat_payload_parts(
+            state,
+            parts,
+            trace_id,
+            body_json,
+            input,
+            eligible,
+            candidate_index,
+            candidate_id,
+            decision_kind,
+            report_kind,
+            transport,
+            upstream_is_stream,
+            redaction.redacted,
+        )
+        .await;
+    }
+
     if provider_api_format == "openai:chat" {
         if let Some(skip_reason) = local_openai_chat_transport_unsupported_reason(transport) {
             mark_skipped_local_openai_chat_candidate(
@@ -909,7 +934,11 @@ async fn resolve_openai_chat_to_openai_image_payload_parts(
     let upstream_url = if is_chatgpt_web {
         chatgpt_web_image_internal_url(&transport.endpoint.base_url)
     } else {
-        build_openai_image_upstream_url(transport, parts.uri.query())
+        build_openai_image_upstream_url(
+            transport,
+            Some("/v1/images/generations"),
+            parts.uri.query(),
+        )
     };
     let Some(mut provider_request_headers) =
         build_openai_image_headers(ProviderOpenAiImageHeadersInput {
@@ -1249,6 +1278,181 @@ fn chatgpt_web_image_internal_url(base_url: &str) -> String {
     format!("{base_url}/__aether/chatgpt-web-image")
 }
 
+#[allow(clippy::too_many_arguments)]
+async fn build_windsurf_openai_chat_payload_parts(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    body_json: &serde_json::Value,
+    input: &LocalOpenAiChatDecisionInput,
+    eligible: &EligibleLocalExecutionCandidate,
+    candidate_index: u32,
+    candidate_id: &str,
+    decision_kind: &str,
+    report_kind: &str,
+    transport: &Arc<GatewayProviderTransportSnapshot>,
+    upstream_is_stream: bool,
+    request_redacted: bool,
+) -> Result<Option<LocalOpenAiChatCandidatePayloadParts>, GatewayError> {
+    let planner_state = crate::ai_serving::PlannerAppState::new(state);
+    let candidate = &eligible.candidate;
+    if let Some(skip_reason) =
+        local_windsurf_request_transport_unsupported_reason_with_network(transport)
+    {
+        mark_skipped_local_openai_chat_candidate(
+            state,
+            input,
+            trace_id,
+            candidate,
+            candidate_index,
+            candidate_id,
+            skip_reason,
+        )
+        .await;
+        return Ok(None);
+    }
+
+    let prepared_candidate = match prepare_header_authenticated_candidate(
+        planner_state,
+        transport,
+        candidate,
+        resolve_windsurf_cascade_auth(transport)
+            .or_else(|| resolve_local_openai_bearer_auth(transport)),
+        OauthPreparationContext {
+            trace_id,
+            api_format: "openai:chat",
+            operation: "openai_chat_windsurf_cascade",
+        },
+    )
+    .await
+    {
+        Ok(prepared) => prepared,
+        Err(skip_reason) => {
+            mark_skipped_local_openai_chat_candidate(
+                state,
+                input,
+                trace_id,
+                candidate,
+                candidate_index,
+                candidate_id,
+                skip_reason,
+            )
+            .await;
+            return Ok(None);
+        }
+    };
+
+    let Some(provider_request_body) = build_windsurf_cascade_request_body(
+        body_json,
+        &prepared_candidate.mapped_model,
+        &prepared_candidate.auth_value,
+        transport.endpoint.body_rules.as_ref(),
+        Some(&parts.headers),
+        upstream_is_stream,
+    ) else {
+        mark_skipped_local_openai_chat_candidate_with_failure_diagnostic(
+            state,
+            input,
+            trace_id,
+            candidate,
+            candidate_index,
+            candidate_id,
+            "provider_request_body_build_failed",
+            CandidateFailureDiagnostic::envelope_build_failed(
+                "openai:chat",
+                "openai:chat",
+                "openai_chat_windsurf_cascade",
+            ),
+        )
+        .await;
+        return Ok(None);
+    };
+
+    let Some(upstream_url) = build_windsurf_cascade_upstream_url(
+        transport.endpoint.base_url.as_str(),
+        parts.uri.query(),
+    ) else {
+        mark_skipped_local_openai_chat_candidate_with_failure_diagnostic(
+            state,
+            input,
+            trace_id,
+            candidate,
+            candidate_index,
+            candidate_id,
+            "upstream_url_missing",
+            CandidateFailureDiagnostic::upstream_url_missing(
+                "openai:chat",
+                "openai:chat",
+                "openai_chat_windsurf_url",
+            ),
+        )
+        .await;
+        return Ok(None);
+    };
+
+    let mut provider_request_headers = match build_windsurf_cascade_headers(
+        &parts.headers,
+        &provider_request_body,
+        body_json,
+        transport.endpoint.header_rules.as_ref(),
+        &prepared_candidate.auth_header,
+        &prepared_candidate.auth_value,
+        upstream_is_stream,
+    ) {
+        Some(headers) => headers,
+        None => {
+            mark_skipped_local_openai_chat_candidate_with_failure_diagnostic(
+                state,
+                input,
+                trace_id,
+                candidate,
+                candidate_index,
+                candidate_id,
+                "transport_header_rules_apply_failed",
+                CandidateFailureDiagnostic::header_rules_apply_failed(
+                    "openai:chat",
+                    "openai:chat",
+                    "openai_chat_windsurf_headers",
+                ),
+            )
+            .await;
+            return Ok(None);
+        }
+    };
+    request_identity_response_encoding_when_redacted(
+        &mut provider_request_headers,
+        request_redacted,
+    );
+
+    let (execution_strategy, conversion_mode) =
+        ai_local_execution_contract_for_formats("openai:chat", "openai:chat");
+    let resolved_report_kind =
+        if decision_kind == OPENAI_CHAT_STREAM_PLAN_KIND || !upstream_is_stream {
+            report_kind.to_string()
+        } else {
+            "openai_chat_sync_finalize".to_string()
+        };
+
+    Ok(Some(LocalOpenAiChatCandidatePayloadParts {
+        client_api_format: "openai:chat".to_string(),
+        auth_header: prepared_candidate.auth_header,
+        auth_value: prepared_candidate.auth_value,
+        mapped_model: prepared_candidate.mapped_model,
+        provider_api_format: "openai:chat".to_string(),
+        provider_request_body,
+        provider_request_headers,
+        upstream_url,
+        execution_strategy,
+        conversion_mode,
+        report_kind: resolved_report_kind,
+        envelope_name: Some(WINDSURF_ENVELOPE_NAME),
+        transport: Arc::clone(transport),
+        request_redacted,
+        transport_profile: None,
+        image_request_summary: None,
+    }))
+}
+
 #[allow(clippy::too_many_arguments)]
 async fn build_kiro_openai_chat_cross_format_payload_parts(
     state: &AppState,

apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision/support.rs

@@ -14,10 +14,7 @@ use crate::ai_serving::planner::candidate_metadata::{
     LocalExecutionCandidateMetadataParts,
 };
 use crate::ai_serving::planner::candidate_resolution::SkippedLocalExecutionCandidate;
-use crate::ai_serving::planner::candidate_source::{
-    preselect_local_execution_candidates_for_api_formats_with_serving,
-    LocalCandidatePreselectionKeyMode,
-};
+use crate::ai_serving::planner::candidate_source::LocalCandidatePreselectionKeyMode;
 use crate::ai_serving::planner::materialization_policy::{
     build_local_candidate_persistence_policy, LocalCandidatePersistencePolicyKind,
 };
@@ -26,7 +23,7 @@ use crate::ai_serving::{
     ai_local_execution_contract_for_formats, extract_pool_sticky_session_token,
     ExecutionRuntimeAuthContext, PlannerAppState,
 };
-use crate::{AppState, GatewayError};
+use crate::AppState;
 
 pub(crate) use crate::ai_serving::planner::candidate_materialization::LocalExecutionCandidateAttempt as LocalOpenAiChatCandidateAttempt;
 pub(crate) use crate::ai_serving::planner::candidate_materialization::LocalExecutionCandidateAttemptSource as LocalOpenAiChatCandidateAttemptSource;
@@ -357,95 +354,3 @@ pub(crate) async fn build_lazy_local_openai_chat_candidate_attempt_source<'a>(
     )
     .await
 }
-
-pub(crate) async fn build_local_openai_chat_image_candidate_attempt_source<'a>(
-    state: &'a AppState,
-    trace_id: &str,
-    input: &LocalOpenAiChatDecisionInput,
-    body_json: &serde_json::Value,
-) -> Result<(LocalOpenAiChatCandidateAttemptSource<'a>, usize), GatewayError> {
-    let planner_state = PlannerAppState::new(state);
-    let sticky_session_token = extract_pool_sticky_session_token(body_json);
-    let auth_context: &ExecutionRuntimeAuthContext = &input.auth_context;
-    let persistence_policy = build_local_candidate_persistence_policy(
-        auth_context,
-        input.required_capabilities.as_ref(),
-        LocalCandidatePersistencePolicyKind::OpenAiChatDecision,
-    );
-    let preselection = preselect_local_execution_candidates_for_api_formats_with_serving(
-        planner_state,
-        "openai:chat",
-        &input.requested_model,
-        false,
-        input.required_capabilities.as_ref(),
-        &input.auth_snapshot,
-        input.routing_policy.as_ref(),
-        input.client_session_affinity.as_ref(),
-        false,
-        LocalCandidatePreselectionKeyMode::ProviderEndpointKeyModelAndApiFormat,
-        vec!["openai:image".to_string()],
-    )
-    .await?;
-
-    Ok(build_local_execution_candidate_attempt_source_with_serving(
-        planner_state,
-        trace_id,
-        "openai:chat",
-        Some(&input.requested_model),
-        Some(&input.auth_snapshot),
-        input.client_session_affinity.as_ref(),
-        input.required_capabilities.as_ref(),
-        input.routing_policy.as_ref(),
-        sticky_session_token.as_deref(),
-        input.request_auth_channel.as_deref(),
-        persistence_policy,
-        preselection.candidates,
-        preselection.skipped_candidates,
-        LocalCandidateResolutionMode::WithoutTransportPairGate,
-        |eligible| {
-            let provider_api_format = eligible.provider_api_format.clone();
-            let (execution_strategy, conversion_mode) =
-                ai_local_execution_contract_for_formats("openai:chat", &provider_api_format);
-            Some(build_local_execution_candidate_contract_metadata(
-                LocalExecutionCandidateMetadataParts {
-                    eligible,
-                    provider_api_format: provider_api_format.as_str(),
-                    client_api_format: "openai:chat",
-                    extra_fields: serde_json::Map::new(),
-                },
-                execution_strategy,
-                conversion_mode,
-                eligible.candidate.endpoint_api_format.trim(),
-            ))
-        },
-        |mut skipped_candidate| {
-            let provider_api_format = skipped_candidate
-                .transport
-                .as_ref()
-                .map(|transport| transport.endpoint.api_format.trim().to_ascii_lowercase())
-                .unwrap_or_else(|| {
-                    skipped_candidate
-                        .candidate
-                        .endpoint_api_format
-                        .trim()
-                        .to_ascii_lowercase()
-                });
-            let (execution_strategy, conversion_mode) =
-                ai_local_execution_contract_for_formats("openai:chat", &provider_api_format);
-            skipped_candidate.extra_data = Some(
-                build_local_execution_candidate_contract_metadata_for_candidate(
-                    &skipped_candidate.candidate,
-                    skipped_candidate.transport_ref(),
-                    provider_api_format.as_str(),
-                    "openai:chat",
-                    serde_json::Map::new(),
-                    execution_strategy,
-                    conversion_mode,
-                    provider_api_format.as_str(),
-                ),
-            );
-            skipped_candidate
-        },
-    )
-    .await)
-}

apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/mod.rs

@@ -11,7 +11,6 @@ mod plans;
 
 use self::decision::{
     build_lazy_local_openai_chat_candidate_attempt_source,
-    build_local_openai_chat_image_candidate_attempt_source,
     maybe_build_local_openai_chat_decision_payload_for_candidate, LocalOpenAiChatCandidateAttempt,
     LocalOpenAiChatCandidateAttemptSource, LocalOpenAiChatDecisionInput,
 };

apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/plans/stream.rs

@@ -1,10 +1,8 @@
 use async_trait::async_trait;
 use tracing::warn;
 
-use super::super::super::openai_request_is_image_generation_intent;
 use super::super::{
     build_lazy_local_openai_chat_candidate_attempt_source,
-    build_local_openai_chat_image_candidate_attempt_source,
     maybe_build_local_openai_chat_decision_payload_for_candidate, AppState, GatewayControlDecision,
     GatewayError, LocalOpenAiChatCandidateAttempt, LocalOpenAiChatCandidateAttemptSource,
     LocalOpenAiChatDecisionInput,
@@ -51,53 +49,14 @@ pub(crate) async fn build_local_openai_chat_stream_attempt_source<'a>(
     };
     let effective_body_json = input.effective_body_json(body_json).clone();
 
-    let image_generation_intent =
-        openai_request_is_image_generation_intent(&input.requested_model, body_json);
-    let (mut candidates, mut candidate_count) = if image_generation_intent {
-        let (image_candidates, image_candidate_count) =
-            build_local_openai_chat_image_candidate_attempt_source(
-                state,
-                trace_id,
-                &input,
-                &effective_body_json,
-            )
-            .await?;
-        if image_candidate_count > 0 {
-            (image_candidates, image_candidate_count)
-        } else {
-            build_lazy_local_openai_chat_candidate_attempt_source(
-                state,
-                trace_id,
-                &input,
-                &effective_body_json,
-                true,
-            )
-            .await
-        }
-    } else {
-        build_lazy_local_openai_chat_candidate_attempt_source(
-            state,
-            trace_id,
-            &input,
-            &effective_body_json,
-            true,
-        )
-        .await
-    };
-    if !image_generation_intent && candidate_count == 0 {
-        let (image_candidates, image_candidate_count) =
-            build_local_openai_chat_image_candidate_attempt_source(
-                state,
-                trace_id,
-                &input,
-                &effective_body_json,
-            )
-            .await?;
-        if image_candidate_count > 0 {
-            candidates = image_candidates;
-            candidate_count = image_candidate_count;
-        }
-    }
+    let (candidates, candidate_count) = build_lazy_local_openai_chat_candidate_attempt_source(
+        state,
+        trace_id,
+        &input,
+        &effective_body_json,
+        true,
+    )
+    .await;
     if candidate_count == 0 {
         set_local_openai_chat_candidate_evaluation_diagnostic(
             state,

apps/aether-gateway/src/ai_serving/planner/standard/openai/image_intent.rs

@@ -1,75 +1,55 @@
-pub(crate) fn openai_request_is_image_generation_intent(
-    requested_model: &str,
-    body_json: &serde_json::Value,
+use serde_json::Value;
+
+pub(super) fn openai_request_is_image_generation_intent(
+    _requested_model: &str,
+    body_json: &Value,
 ) -> bool {
-    openai_model_is_image_generation(requested_model)
-        || body_json
-            .get("model")
-            .and_then(serde_json::Value::as_str)
-            .is_some_and(openai_model_is_image_generation)
-        || openai_tool_choice_selects_image_generation(body_json.get("tool_choice"))
+    request_forces_image_generation_tool(body_json)
 }
 
-fn openai_model_is_image_generation(model: &str) -> bool {
-    model.trim().to_ascii_lowercase().starts_with("gpt-image-")
+fn request_forces_image_generation_tool(body_json: &Value) -> bool {
+    body_json
+        .get("tool_choice")
+        .is_some_and(value_is_image_generation_tool)
 }
 
-fn openai_tool_choice_selects_image_generation(choice: Option<&serde_json::Value>) -> bool {
-    let Some(choice) = choice else {
-        return false;
-    };
-    if let Some(value) = choice.as_str() {
-        return value.trim().eq_ignore_ascii_case("image_generation");
-    }
-    let Some(object) = choice.as_object() else {
-        return false;
-    };
-    object
+fn value_is_image_generation_tool(value: &Value) -> bool {
+    value
         .get("type")
-        .and_then(serde_json::Value::as_str)
-        .is_some_and(|value| value.trim().eq_ignore_ascii_case("image_generation"))
-        || object
-            .get("tool")
-            .and_then(|value| value.get("type"))
-            .and_then(serde_json::Value::as_str)
-            .is_some_and(|value| value.trim().eq_ignore_ascii_case("image_generation"))
-        || object
-            .get("function")
-            .and_then(|value| value.get("name"))
-            .and_then(serde_json::Value::as_str)
-            .is_some_and(|value| value.trim().eq_ignore_ascii_case("image_generation"))
+        .and_then(Value::as_str)
+        .is_some_and(|tool_type| tool_type.trim().eq_ignore_ascii_case("image_generation"))
 }
 
 #[cfg(test)]
 mod tests {
-    use super::openai_request_is_image_generation_intent;
-    use serde_json::json;
+    use super::*;
 
     #[test]
-    fn detects_openai_image_generation_intent_like_compat_proxies() {
-        assert!(openai_request_is_image_generation_intent(
-            "GPT-IMAGE-2",
-            &json!({})
-        ));
-        assert!(openai_request_is_image_generation_intent(
-            "gpt-5",
-            &json!({"model":"gpt-image-2"})
-        ));
-        assert!(openai_request_is_image_generation_intent(
-            "gpt-5",
-            &json!({"tool_choice":{"function":{"name":"image_generation"}}})
-        ));
-        assert!(openai_request_is_image_generation_intent(
-            "gpt-5",
-            &json!({"tool_choice":{"type":"image_generation"}})
-        ));
+    fn tools_declaration_without_tool_choice_does_not_trigger_image_generation() {
+        let body_json = serde_json::json!({
+            "model": "gpt-image-2",
+            "input": "Draw a mountain observatory",
+            "tools": [{"type": "image_generation"}]
+        });
+
         assert!(!openai_request_is_image_generation_intent(
-            "gpt-5",
-            &json!({"tools":[{"type":"image_generation"}]})
+            "gpt-image-2",
+            &body_json
         ));
-        assert!(!openai_request_is_image_generation_intent(
-            "gpt-5",
-            &json!({"messages":[{"role":"user","content":"hello"}]})
+    }
+
+    #[test]
+    fn explicit_image_generation_tool_choice_triggers_image_generation() {
+        let body_json = serde_json::json!({
+            "model": "gpt-image-2",
+            "input": "Draw a mountain observatory",
+            "tools": [{"type": "image_generation"}],
+            "tool_choice": {"type": "image_generation"}
+        });
+
+        assert!(openai_request_is_image_generation_intent(
+            "gpt-image-2",
+            &body_json
         ));
     }
 }

apps/aether-gateway/src/ai_serving/planner/standard/openai/mod.rs

@@ -17,7 +17,7 @@ pub(crate) use chat::{
     set_local_openai_chat_execution_exhausted_diagnostic,
 };
 pub(crate) use embedding::maybe_build_sync_local_openai_embedding_decision_payload;
-pub(super) use image_intent::openai_request_is_image_generation_intent;
+use image_intent::openai_request_is_image_generation_intent;
 pub(crate) use responses::{
     build_local_openai_responses_stream_attempt_source_for_kind,
     build_local_openai_responses_stream_plan_and_reports_for_kind,

apps/aether-gateway/src/ai_serving/planner/standard/openai/responses/decision/request.rs

@@ -39,10 +39,13 @@ use crate::ai_serving::transport::kiro::{
 use crate::ai_serving::transport::{
     build_grok_browser_headers, build_grok_upstream_url, build_kiro_cross_format_upstream_url,
     build_openai_image_headers, build_openai_image_upstream_url,
-    build_standard_provider_request_headers,
-    local_standard_transport_unsupported_reason_with_network,
+    build_standard_provider_request_headers, build_windsurf_cascade_headers,
+    build_windsurf_cascade_request_body, build_windsurf_cascade_upstream_url,
+    is_windsurf_provider_transport, local_standard_transport_unsupported_reason_with_network,
+    local_windsurf_request_transport_unsupported_reason_with_network,
     openai_image_transport_unsupported_reason, resolve_openai_image_auth, GrokHeaderInput,
     ProviderOpenAiImageHeadersInput, StandardProviderRequestHeadersInput, GROK_CHAT_PATH,
+    WINDSURF_ENVELOPE_NAME,
 };
 use crate::ai_serving::{
     ai_local_execution_contract_for_formats, request_conversion_direct_auth,
@@ -114,7 +117,7 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
         .trim()
         .eq_ignore_ascii_case("grok");
 
-    if provider_api_format.eq_ignore_ascii_case("openai:image") {
+    if !is_grok && provider_api_format.eq_ignore_ascii_case("openai:image") {
         return resolve_openai_responses_to_openai_image_payload_parts(
             state,
             parts,
@@ -128,6 +131,8 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
         )
         .await;
     }
+    let is_windsurf_cascade =
+        provider_api_format == "openai:chat" && is_windsurf_provider_transport(transport);
 
     let same_format = api_format_alias_matches(provider_api_format, &client_api_format);
     let conversion_kind = request_conversion_kind(spec_metadata.api_format, provider_api_format);
@@ -139,6 +144,8 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
         local_kiro_request_transport_unsupported_reason_with_network(transport)
     } else if same_format {
         local_standard_transport_unsupported_reason_with_network(transport, provider_api_format)
+    } else if is_windsurf_cascade {
+        local_windsurf_request_transport_unsupported_reason_with_network(transport)
     } else {
         match conversion_kind {
             Some(_) if is_antigravity && provider_api_format == "gemini:generate_content" => None,
@@ -302,7 +309,7 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
                 upstream_is_stream,
                 force_body_stream_field,
                 transport.provider.provider_type.as_str(),
-                if is_kiro_claude_cli {
+                if is_kiro_claude_cli || is_windsurf_cascade {
                     None
                 } else {
                     transport.endpoint.body_rules.as_ref()
@@ -319,7 +326,7 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
                 force_body_stream_field,
                 transport.provider.provider_type.as_str(),
                 provider_api_format,
-                if is_kiro_claude_cli {
+                if is_kiro_claude_cli || is_windsurf_cascade {
                     None
                 } else {
                     transport.endpoint.body_rules.as_ref()
@@ -447,6 +454,27 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
         )
         .await;
     }
+    if is_windsurf_cascade {
+        return build_windsurf_openai_responses_payload_parts(
+            state,
+            parts,
+            trace_id,
+            body_json,
+            input,
+            eligible,
+            candidate_index,
+            candidate_id,
+            spec_metadata.api_format,
+            transport,
+            provider_api_format,
+            mapped_model,
+            auth_header,
+            auth_value,
+            provider_request_body,
+            upstream_is_stream,
+        )
+        .await;
+    }
 
     let Some(upstream_url) = (if is_grok && is_grok_text_provider_api_format(provider_api_format) {
         Some(build_grok_upstream_url(transport, GROK_CHAT_PATH))
@@ -619,6 +647,130 @@ pub(crate) async fn resolve_local_openai_responses_candidate_payload_parts(
     })
 }
 
+#[allow(clippy::too_many_arguments)]
+async fn build_windsurf_openai_responses_payload_parts(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    original_body_json: &serde_json::Value,
+    input: &LocalOpenAiResponsesDecisionInput,
+    eligible: &EligibleLocalExecutionCandidate,
+    candidate_index: u32,
+    candidate_id: &str,
+    client_api_format: &str,
+    transport: &Arc<GatewayProviderTransportSnapshot>,
+    provider_api_format: &str,
+    mapped_model: String,
+    auth_header: String,
+    auth_value: String,
+    openai_chat_request_body: Value,
+    upstream_is_stream: bool,
+) -> Option<LocalOpenAiResponsesCandidatePayloadParts> {
+    let candidate = &eligible.candidate;
+    let effective_headers = input.effective_headers(&parts.headers);
+    let provider_request_body = match build_windsurf_cascade_request_body(
+        &openai_chat_request_body,
+        &mapped_model,
+        &auth_value,
+        transport.endpoint.body_rules.as_ref(),
+        Some(effective_headers),
+        upstream_is_stream,
+    ) {
+        Some(body) => body,
+        None => {
+            mark_skipped_local_openai_responses_candidate_with_failure_diagnostic(
+                state,
+                input,
+                trace_id,
+                candidate,
+                candidate_index,
+                candidate_id,
+                "provider_request_body_build_failed",
+                CandidateFailureDiagnostic::envelope_build_failed(
+                    client_api_format,
+                    provider_api_format,
+                    "openai_responses_windsurf_cascade",
+                ),
+            )
+            .await;
+            return None;
+        }
+    };
+    let upstream_url = match build_windsurf_cascade_upstream_url(
+        transport.endpoint.base_url.as_str(),
+        parts.uri.query(),
+    ) {
+        Some(url) => url,
+        None => {
+            mark_skipped_local_openai_responses_candidate_with_failure_diagnostic(
+                state,
+                input,
+                trace_id,
+                candidate,
+                candidate_index,
+                candidate_id,
+                "upstream_url_missing",
+                CandidateFailureDiagnostic::upstream_url_missing(
+                    client_api_format,
+                    provider_api_format,
+                    "openai_responses_windsurf_url",
+                ),
+            )
+            .await;
+            return None;
+        }
+    };
+    let provider_request_headers = match build_windsurf_cascade_headers(
+        effective_headers,
+        &provider_request_body,
+        original_body_json,
+        transport.endpoint.header_rules.as_ref(),
+        &auth_header,
+        &auth_value,
+        upstream_is_stream,
+    ) {
+        Some(headers) => headers,
+        None => {
+            mark_skipped_local_openai_responses_candidate_with_failure_diagnostic(
+                state,
+                input,
+                trace_id,
+                candidate,
+                candidate_index,
+                candidate_id,
+                "transport_header_rules_apply_failed",
+                CandidateFailureDiagnostic::header_rules_apply_failed(
+                    client_api_format,
+                    provider_api_format,
+                    "openai_responses_windsurf_headers",
+                ),
+            )
+            .await;
+            return None;
+        }
+    };
+    let (execution_strategy, conversion_mode) =
+        ai_local_execution_contract_for_formats(client_api_format, provider_api_format);
+
+    Some(LocalOpenAiResponsesCandidatePayloadParts {
+        auth_header,
+        auth_value,
+        mapped_model,
+        provider_api_format: provider_api_format.to_string(),
+        provider_request_body,
+        provider_request_headers,
+        upstream_url,
+        execution_strategy,
+        conversion_mode,
+        is_antigravity: false,
+        envelope_name: Some(WINDSURF_ENVELOPE_NAME),
+        upstream_is_stream,
+        transport: Arc::clone(transport),
+        transport_profile: None,
+        image_request_summary: None,
+    })
+}
+
 fn api_format_alias_matches(left: &str, right: &str) -> bool {
     crate::ai_serving::api_format_alias_matches(left, right)
 }
@@ -739,7 +891,11 @@ async fn resolve_openai_responses_to_openai_image_payload_parts(
     let upstream_url = if is_chatgpt_web {
         chatgpt_web_image_internal_url(&transport.endpoint.base_url)
     } else {
-        build_openai_image_upstream_url(transport, parts.uri.query())
+        build_openai_image_upstream_url(
+            transport,
+            Some("/v1/images/generations"),
+            parts.uri.query(),
+        )
     };
     let Some(mut provider_request_headers) =
         build_openai_image_headers(ProviderOpenAiImageHeadersInput {

apps/aether-gateway/src/ai_serving/pure/mod.rs

@@ -20,6 +20,7 @@ pub(crate) use aether_ai_formats::api::{
     build_local_openai_responses_request_body,
     build_local_openai_responses_request_body_with_model_directives,
     build_local_success_background_report, build_local_success_conversion_background_report,
+    build_openai_image_api_provider_request_body,
     build_openai_image_provider_body_from_response_stream_sync_body,
     build_openai_image_provider_request_body,
     build_openai_image_request_body_from_gemini_image_request,

apps/aether-gateway/src/ai_serving/transport.rs

@@ -50,6 +50,10 @@ pub(crate) mod vertex {
     pub(crate) use aether_provider_transport::vertex::*;
 }
 
+pub(crate) mod windsurf {
+    pub(crate) use aether_provider_transport::windsurf::*;
+}
+
 pub(crate) use aether_provider_transport::{
     append_transport_diagnostics_to_value, apply_local_body_rules,
     apply_local_body_rules_with_request_headers, apply_local_header_rules,
@@ -69,12 +73,15 @@ pub(crate) use aether_provider_transport::{
     build_standard_plan_fallback_openai_responses_url, build_standard_provider_request_headers,
     build_transport_request_url, build_transport_request_url_for_request_body,
     build_video_create_headers, build_video_create_request_body, build_video_create_upstream_url,
-    candidate_common_transport_skip_reason, candidate_transport_pair_skip_reason,
-    classify_same_format_provider_request_behavior, ensure_upstream_auth_header,
-    gemini_files_transport_unsupported_reason, header_rules_are_locally_supported,
-    header_rules_have_enabled_rules, local_gemini_transport_unsupported_reason_with_network,
+    build_windsurf_cascade_headers, build_windsurf_cascade_request_body,
+    build_windsurf_cascade_upstream_url, candidate_common_transport_skip_reason,
+    candidate_transport_pair_skip_reason, classify_same_format_provider_request_behavior,
+    ensure_upstream_auth_header, gemini_files_transport_unsupported_reason,
+    header_rules_are_locally_supported, header_rules_have_enabled_rules,
+    is_windsurf_provider_transport, local_gemini_transport_unsupported_reason_with_network,
     local_openai_chat_transport_unsupported_reason,
     local_standard_transport_unsupported_reason_with_network,
+    local_windsurf_request_transport_unsupported_reason_with_network,
     openai_image_transport_unsupported_reason, request_conversion_direct_auth,
     request_conversion_enabled_for_transport, request_conversion_transport_supported,
     request_conversion_transport_unsupported_reason, request_pair_allowed_for_transport,
@@ -97,4 +104,5 @@ pub(crate) use aether_provider_transport::{
     StandardPlanFallbackHeadersInput, StandardProviderRequestHeaders,
     StandardProviderRequestHeadersInput, TransportRequestBodySemanticsError,
     TransportRequestUrlParams, GROK_CHAT_PATH, GROK_INTERNAL_HEADER, GROK_RATE_LIMITS_PATH,
+    WINDSURF_ENVELOPE_NAME,
 };

apps/aether-gateway/src/async_task/runtime.rs

@@ -8,7 +8,7 @@ use aether_data_contracts::repository::video_tasks::{
 use aether_usage_runtime::{build_upsert_usage_record_from_event, settle_usage_if_needed};
 use serde_json::{Map, Value};
 use tokio::task::JoinHandle;
-use tracing::{info, warn};
+use tracing::{debug, info, warn};
 
 use crate::log_ids::short_request_id;
 use crate::usage::{UsageEvent, UsageEventData, UsageEventType};
@@ -148,8 +148,20 @@ pub(crate) fn spawn_video_task_poller(state: AppState) -> Option<JoinHandle<()>>
         let mut interval = tokio::time::interval(config.interval);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if state
+                .data
+                .should_defer_maintenance_for_database_pool_pressure(&mut deferred_since)
+            {
+                debug!(
+                    event_name = "video_task_poller_deferred",
+                    log_type = "event",
+                    "gateway video task poller deferred because database pool has no idle reserve"
+                );
+                continue;
+            }
             if let Err(err) = poll_video_tasks_once(&state, config.batch_size).await {
                 warn!(
                     event_name = "video_task_poller_tick_failed",

apps/aether-gateway/src/bark_push.rs

@@ -0,0 +1,173 @@
+use crate::handlers::shared::{
+    decrypt_catalog_secret_with_fallbacks, system_config_bool, system_config_string,
+};
+use crate::{AppState, GatewayError};
+use serde_json::{json, Value};
+
+pub(crate) const BARK_PUSH_ENABLED_KEY: &str = "module.bark_push.enabled";
+pub(crate) const BARK_PUSH_DEVICE_KEY_KEY: &str = "module.bark_push.device_key";
+pub(crate) const BARK_PUSH_SERVER_URL_KEY: &str = "module.bark_push.server_url";
+pub(crate) const BARK_PUSH_TEMPLATE_KEY: &str = "module.bark_push.template";
+
+const DEFAULT_BARK_API_BASE: &str = "https://api.day.app";
+
+#[derive(Debug, Clone)]
+pub(crate) struct BarkPushConfig {
+    pub(crate) enabled: bool,
+    pub(crate) device_key: Option<String>,
+    pub(crate) server_url: String,
+    pub(crate) template: Option<String>,
+}
+
+pub(crate) async fn bark_push_module_enabled(state: &AppState) -> Result<bool, GatewayError> {
+    let value = state
+        .read_system_config_json_value(BARK_PUSH_ENABLED_KEY)
+        .await?;
+    Ok(system_config_bool(value.as_ref(), false))
+}
+
+pub(crate) async fn bark_push_configured(state: &AppState) -> Result<bool, GatewayError> {
+    let config = read_bark_push_config(state).await?;
+    Ok(config.device_key.is_some() && !config.server_url.trim().is_empty())
+}
+
+pub(crate) async fn read_bark_push_config(
+    state: &AppState,
+) -> Result<BarkPushConfig, GatewayError> {
+    let enabled = bark_push_module_enabled(state).await?;
+    let device_key = state
+        .read_system_config_json_value(BARK_PUSH_DEVICE_KEY_KEY)
+        .await?
+        .and_then(|value| system_config_string(Some(&value)))
+        .map(|value| {
+            decrypt_catalog_secret_with_fallbacks(state.encryption_key(), &value).unwrap_or(value)
+        });
+    let server_url = state
+        .read_system_config_json_value(BARK_PUSH_SERVER_URL_KEY)
+        .await?
+        .and_then(|value| system_config_string(Some(&value)))
+        .filter(|value| !value.trim().is_empty())
+        .unwrap_or_else(|| DEFAULT_BARK_API_BASE.to_string());
+    let template = state
+        .read_system_config_json_value(BARK_PUSH_TEMPLATE_KEY)
+        .await?
+        .and_then(|value| system_config_string(Some(&value)));
+
+    Ok(BarkPushConfig {
+        enabled,
+        device_key,
+        server_url,
+        template,
+    })
+}
+
+pub(crate) async fn send_bark_push(
+    state: &AppState,
+    config: &BarkPushConfig,
+    title: &str,
+    markdown_body: &str,
+) -> Result<(), GatewayError> {
+    let Some(device_key) = config.device_key.as_deref() else {
+        return Err(GatewayError::Internal("未配置 Bark Device Key".to_string()));
+    };
+    let device_key = device_key.trim();
+    if device_key.is_empty() {
+        return Err(GatewayError::Internal(
+            "Bark Device Key 不能为空".to_string(),
+        ));
+    }
+    let server_url = normalized_bark_server_url(&config.server_url)?;
+    let body = render_bark_body(config.template.as_deref(), title, markdown_body);
+    let response = state
+        .client
+        .post(format!("{server_url}/push"))
+        .json(&json!({
+            "device_key": device_key,
+            "title": title,
+            "body": body,
+        }))
+        .send()
+        .await
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let status = response.status();
+    let text = response
+        .text()
+        .await
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    if !status.is_success() {
+        return Err(GatewayError::Internal(format!(
+            "Bark 返回 HTTP {status}: {text}"
+        )));
+    }
+    if let Ok(payload) = serde_json::from_str::<Value>(&text) {
+        let code_is_ok = payload
+            .get("code")
+            .and_then(|value| {
+                value
+                    .as_i64()
+                    .map(|code| matches!(code, 0 | 200))
+                    .or_else(|| {
+                        value
+                            .as_str()
+                            .map(|code| matches!(code.trim(), "0" | "200"))
+                    })
+            })
+            .unwrap_or(true);
+        if !code_is_ok {
+            return Err(GatewayError::Internal(format!("Bark 返回失败: {payload}")));
+        }
+    }
+    Ok(())
+}
+
+fn normalized_bark_server_url(server_url: &str) -> Result<String, GatewayError> {
+    let server_url = server_url.trim().trim_end_matches('/');
+    if server_url.is_empty() {
+        return Err(GatewayError::Internal(
+            "Bark 服务器地址不能为空".to_string(),
+        ));
+    }
+    if !server_url.starts_with("https://") && !server_url.starts_with("http://") {
+        return Err(GatewayError::Internal(
+            "Bark 服务器地址必须以 http:// 或 https:// 开头".to_string(),
+        ));
+    }
+    Ok(server_url.to_string())
+}
+
+fn render_bark_body(template: Option<&str>, title: &str, markdown_body: &str) -> String {
+    match template {
+        Some(template) if !template.trim().is_empty() => template
+            .replace("{title}", title)
+            .replace("{body}", markdown_body),
+        _ => markdown_body.to_string(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{normalized_bark_server_url, render_bark_body};
+
+    #[test]
+    fn bark_body_uses_template_when_provided() {
+        let rendered = render_bark_body(Some("{title}\n\n{body}"), "告警", "原始正文");
+        assert_eq!(rendered, "告警\n\n原始正文");
+    }
+
+    #[test]
+    fn bark_body_falls_back_to_markdown_body_for_empty_template() {
+        assert_eq!(render_bark_body(None, "告警", "原始正文"), "原始正文");
+        assert_eq!(
+            render_bark_body(Some("   "), "告警", "原始正文"),
+            "原始正文"
+        );
+    }
+
+    #[test]
+    fn bark_server_url_trims_trailing_slashes() {
+        assert_eq!(
+            normalized_bark_server_url(" https://api.day.app/ ").expect("url should parse"),
+            "https://api.day.app"
+        );
+    }
+}

apps/aether-gateway/src/control/route/admin/endpoints_families.rs

@@ -177,17 +177,6 @@ pub(super) fn classify_admin_endpoints_family_route(
             "admin:endpoints_manage",
             false,
         ))
-    } else if method == http::Method::POST
-        && normalized_path.starts_with("/api/admin/endpoints/providers/")
-        && normalized_path.ends_with("/key-balance")
-    {
-        Some(classified(
-            "admin_proxy",
-            "endpoints_manage",
-            "query_key_balance",
-            "admin:endpoints_manage",
-            false,
-        ))
     } else if method == http::Method::POST
         && normalized_path.starts_with("/api/admin/endpoints/providers/")
         && normalized_path.ends_with("/keys")

apps/aether-gateway/src/control/route/admin/system_families.rs

@@ -162,6 +162,16 @@ pub(super) fn classify_admin_system_family_route(
             "admin:system",
             false,
         ))
+    } else if method == http::Method::POST
+        && normalized_path == "/api/admin/system/important-notification/test"
+    {
+        Some(classified(
+            "admin_proxy",
+            "system_manage",
+            "important_notification_test",
+            "admin:system",
+            false,
+        ))
     } else if method == http::Method::POST && normalized_path == "/api/admin/system/cleanup" {
         Some(classified(
             "admin_proxy",

apps/aether-gateway/src/control/tests/admin_core.rs

@@ -183,6 +183,10 @@ fn classifies_admin_system_maintenance_write_routes_as_admin_proxy_route() {
         ("/api/admin/system/users/import", "users_import"),
         ("/api/admin/system/data/import", "data_import"),
         ("/api/admin/system/smtp/test", "smtp_test"),
+        (
+            "/api/admin/system/important-notification/test",
+            "important_notification_test",
+        ),
         ("/api/admin/system/cleanup", "cleanup"),
         ("/api/admin/system/purge/config", "purge_config"),
         ("/api/admin/system/purge/users", "purge_users"),

apps/aether-gateway/src/control/tests/admin_endpoints.rs

@@ -380,20 +380,6 @@ fn classifies_admin_refresh_provider_quota_as_admin_proxy_route() {
     assert!(!decision.is_execution_runtime_candidate());
 }
 
-#[test]
-fn classifies_admin_query_provider_key_balance_as_admin_proxy_route() {
-    let headers = http::HeaderMap::new();
-    let uri: Uri = "/api/admin/endpoints/providers/provider-newapi/key-balance"
-        .parse()
-        .expect("uri should parse");
-    let decision = classify_control_route(&http::Method::POST, &uri, &headers)
-        .expect("decision should resolve");
-    assert_eq!(decision.route_class.as_deref(), Some("admin_proxy"));
-    assert_eq!(decision.route_family.as_deref(), Some("endpoints_manage"));
-    assert_eq!(decision.route_kind.as_deref(), Some("query_key_balance"));
-    assert!(!decision.is_execution_runtime_candidate());
-}
-
 #[test]
 fn admin_refresh_provider_quota_buffers_request_body_for_key_selection() {
     let headers = headers(&[]);
@@ -413,25 +399,6 @@ fn admin_refresh_provider_quota_buffers_request_body_for_key_selection() {
     assert!(local_proxy_route_requires_buffered_body(&context));
 }
 
-#[test]
-fn admin_query_provider_key_balance_buffers_request_body_for_key_secret() {
-    let headers = headers(&[]);
-    let uri: Uri = "/api/admin/endpoints/providers/provider-newapi/key-balance"
-        .parse()
-        .expect("uri should parse");
-    let decision = classify_control_route(&http::Method::POST, &uri, &headers)
-        .expect("decision should resolve");
-    let context = GatewayPublicRequestContext::from_request_parts(
-        "trace-key-balance",
-        &http::Method::POST,
-        &uri,
-        &headers,
-        Some(decision),
-    );
-
-    assert!(local_proxy_route_requires_buffered_body(&context));
-}
-
 #[test]
 fn classifies_admin_default_body_rules_as_admin_proxy_route() {
     let headers = headers(&[]);

apps/aether-gateway/src/data/state/auth.rs

@@ -2000,7 +2000,7 @@ fn resolve_effective_rate_limit_policy(
     groups: &[aether_data::repository::users::StoredUserGroup],
 ) -> Option<i32> {
     let group_policy = groups.iter().fold(None, |effective, group| {
-        intersect_rate_limit_policies(
+        union_rate_limit_policies(
             effective,
             rate_limit_restriction_from_mode(&group.rate_limit_mode, group.rate_limit),
         )
@@ -2073,6 +2073,22 @@ fn intersect_rate_limit_policies(
     }
 }
 
+fn union_rate_limit_policies(
+    left: Option<RateLimitRestriction>,
+    right: Option<RateLimitRestriction>,
+) -> Option<RateLimitRestriction> {
+    match (left, right) {
+        (None, None) => None,
+        (Some(value), None) | (None, Some(value)) => Some(value),
+        (Some(RateLimitRestriction::Unlimited), _) | (_, Some(RateLimitRestriction::Unlimited)) => {
+            Some(RateLimitRestriction::Unlimited)
+        }
+        (Some(RateLimitRestriction::Limited(left)), Some(RateLimitRestriction::Limited(right))) => {
+            Some(RateLimitRestriction::Limited(left.max(right)))
+        }
+    }
+}
+
 fn rate_limit_policy_value(policy: Option<RateLimitRestriction>) -> Option<i32> {
     match policy {
         None => None,
@@ -2358,35 +2374,44 @@ mod tests {
     }
 
     #[test]
-    fn rate_limit_policy_uses_most_restrictive_custom_limit() {
-        let groups = vec![sample_group(
-            "restricted",
-            10,
-            None,
-            "unrestricted",
-            Some(60),
-            "custom",
-        )];
+    fn rate_limit_policy_uses_highest_group_limit_before_user_restriction() {
+        let groups = vec![
+            sample_group("default", 10, None, "unrestricted", Some(30), "custom"),
+            sample_group("tier-1", 20, None, "unrestricted", Some(100), "custom"),
+        ];
 
         assert_eq!(
             resolve_effective_rate_limit_policy(Some(120), "custom", &groups),
-            Some(60)
+            Some(100)
         );
     }
 
     #[test]
-    fn rate_limit_unlimited_does_not_bypass_limited_group() {
+    fn rate_limit_unlimited_group_overrides_limited_groups() {
+        let groups = vec![
+            sample_group("default", 10, None, "unrestricted", Some(30), "custom"),
+            sample_group("tier-2", 20, None, "unrestricted", Some(0), "custom"),
+        ];
+
+        assert_eq!(
+            resolve_effective_rate_limit_policy(None, "system", &groups),
+            Some(0)
+        );
+    }
+
+    #[test]
+    fn rate_limit_user_policy_still_restricts_group_grants() {
         let groups = vec![sample_group(
-            "restricted",
+            "tier-1",
             10,
             None,
             "unrestricted",
-            Some(60),
+            Some(100),
             "custom",
         )];
 
         assert_eq!(
-            resolve_effective_rate_limit_policy(Some(0), "custom", &groups),
+            resolve_effective_rate_limit_policy(Some(60), "custom", &groups),
             Some(60)
         );
     }

apps/aether-gateway/src/data/state/runtime.rs

@@ -42,8 +42,12 @@ use aether_data_contracts::repository::usage::{
 };
 use aether_runtime_state::RuntimeQueueStore;
 use aether_video_tasks_core::read_data_backed_video_task_response;
+use std::time::{Duration, Instant};
 
 impl GatewayDataState {
+    const MAINTENANCE_POOL_IDLE_RESERVE: usize = 1;
+    const MAINTENANCE_POOL_PRESSURE_MAX_DEFER: Duration = Duration::from_secs(30);
+
     pub(crate) async fn run_database_maintenance(
         &self,
         table_names: &[&str],
@@ -112,6 +116,47 @@ impl GatewayDataState {
             .and_then(|backends| backends.database_pool_summary())
     }
 
+    pub(crate) fn database_pool_under_maintenance_pressure(&self) -> bool {
+        self.database_pool_summary()
+            .as_ref()
+            .is_some_and(Self::database_pool_summary_under_maintenance_pressure)
+    }
+
+    pub(crate) fn database_pool_summary_under_maintenance_pressure(
+        summary: &aether_data::DatabasePoolSummary,
+    ) -> bool {
+        summary.checked_out > 0 && summary.idle <= Self::MAINTENANCE_POOL_IDLE_RESERVE
+    }
+
+    pub(crate) fn should_defer_maintenance_for_database_pool_pressure(
+        &self,
+        deferred_since: &mut Option<Instant>,
+    ) -> bool {
+        Self::should_defer_maintenance_for_pool_pressure_state(
+            self.database_pool_under_maintenance_pressure(),
+            deferred_since,
+        )
+    }
+
+    pub(crate) fn should_defer_maintenance_for_pool_pressure_state(
+        pool_under_pressure: bool,
+        deferred_since: &mut Option<Instant>,
+    ) -> bool {
+        if !pool_under_pressure {
+            *deferred_since = None;
+            return false;
+        }
+
+        let now = Instant::now();
+        let since = deferred_since.get_or_insert(now);
+        if now.duration_since(*since) >= Self::MAINTENANCE_POOL_PRESSURE_MAX_DEFER {
+            *deferred_since = None;
+            return false;
+        }
+
+        true
+    }
+
     pub(crate) async fn aggregate_wallet_daily_usage(
         &self,
         input: &WalletDailyUsageAggregationInput,

apps/aether-gateway/src/data/tests.rs

@@ -1,4 +1,5 @@
 use std::sync::Arc;
+use std::time::{Duration, Instant};
 
 use aether_crypto::{encrypt_python_fernet_plaintext, DEVELOPMENT_ENCRYPTION_KEY};
 use aether_data::repository::auth::{
@@ -51,6 +52,65 @@ fn disabled_gateway_data_state_has_no_backends() {
     assert!(!state.has_video_task_reader());
 }
 
+#[test]
+fn maintenance_pool_pressure_keeps_idle_reserve_for_foreground_work() {
+    let pressured = aether_data::DatabasePoolSummary {
+        driver: DatabaseDriver::Postgres,
+        checked_out: 6,
+        pool_size: 6,
+        idle: 0,
+        max_connections: 20,
+        usage_rate: 30.0,
+    };
+    assert!(GatewayDataState::database_pool_summary_under_maintenance_pressure(&pressured));
+
+    let one_idle_left = aether_data::DatabasePoolSummary {
+        driver: DatabaseDriver::Postgres,
+        checked_out: 5,
+        pool_size: 6,
+        idle: 1,
+        max_connections: 20,
+        usage_rate: 25.0,
+    };
+    assert!(GatewayDataState::database_pool_summary_under_maintenance_pressure(&one_idle_left));
+
+    let idle = aether_data::DatabasePoolSummary {
+        driver: DatabaseDriver::Postgres,
+        checked_out: 0,
+        pool_size: 4,
+        idle: 4,
+        max_connections: 20,
+        usage_rate: 0.0,
+    };
+    assert!(!GatewayDataState::database_pool_summary_under_maintenance_pressure(&idle));
+}
+
+#[test]
+fn maintenance_pool_pressure_deferral_has_timeout() {
+    let mut deferred_since = None;
+    assert!(
+        GatewayDataState::should_defer_maintenance_for_pool_pressure_state(
+            true,
+            &mut deferred_since
+        )
+    );
+    assert!(deferred_since.is_some());
+
+    assert!(
+        !GatewayDataState::should_defer_maintenance_for_pool_pressure_state(
+            false,
+            &mut deferred_since
+        )
+    );
+    assert!(deferred_since.is_none());
+
+    let mut stale_defer = Some(Instant::now() - Duration::from_secs(31));
+    assert!(
+        !GatewayDataState::should_defer_maintenance_for_pool_pressure_state(true, &mut stale_defer)
+    );
+    assert!(stale_defer.is_none());
+}
+
 #[tokio::test]
 async fn postgres_gateway_data_state_builds_video_task_reader() {
     let state = GatewayDataState::from_config(GatewayDataConfig::from_postgres_url(

apps/aether-gateway/src/email_delivery.rs

@@ -0,0 +1,381 @@
+use base64::Engine;
+
+use crate::handlers::shared::{
+    decrypt_catalog_secret_with_fallbacks, system_config_bool, system_config_string,
+};
+use crate::{AppState, GatewayError};
+
+const SMTP_TIMEOUT_SECS: u64 = 30;
+
+#[derive(Debug, Clone)]
+pub(crate) struct SmtpDeliveryConfig {
+    pub(crate) host: String,
+    pub(crate) port: u16,
+    pub(crate) user: Option<String>,
+    pub(crate) password: Option<String>,
+    pub(crate) use_tls: bool,
+    pub(crate) use_ssl: bool,
+    pub(crate) from_email: String,
+    pub(crate) from_name: String,
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct ComposedEmail {
+    pub(crate) to_email: String,
+    pub(crate) subject: String,
+    pub(crate) html_body: String,
+    pub(crate) text_body: String,
+}
+
+pub(crate) async fn read_smtp_delivery_config(
+    state: &AppState,
+) -> Result<Option<SmtpDeliveryConfig>, GatewayError> {
+    let smtp_host = state.read_system_config_json_value("smtp_host").await?;
+    let smtp_from_email = state
+        .read_system_config_json_value("smtp_from_email")
+        .await?;
+    let Some(host) = system_config_string(smtp_host.as_ref()) else {
+        return Ok(None);
+    };
+    let Some(from_email) = system_config_string(smtp_from_email.as_ref()) else {
+        return Ok(None);
+    };
+
+    let smtp_port = state.read_system_config_json_value("smtp_port").await?;
+    let smtp_user = state.read_system_config_json_value("smtp_user").await?;
+    let smtp_password = state.read_system_config_json_value("smtp_password").await?;
+    let smtp_use_tls = state.read_system_config_json_value("smtp_use_tls").await?;
+    let smtp_use_ssl = state.read_system_config_json_value("smtp_use_ssl").await?;
+    let smtp_from_name = state
+        .read_system_config_json_value("smtp_from_name")
+        .await?;
+
+    let password = system_config_string(smtp_password.as_ref()).map(|value| {
+        decrypt_catalog_secret_with_fallbacks(state.encryption_key(), &value).unwrap_or(value)
+    });
+
+    Ok(Some(SmtpDeliveryConfig {
+        host,
+        port: system_config_u16(smtp_port.as_ref(), 587),
+        user: system_config_string(smtp_user.as_ref()),
+        password,
+        use_tls: system_config_bool(smtp_use_tls.as_ref(), true),
+        use_ssl: system_config_bool(smtp_use_ssl.as_ref(), false),
+        from_email,
+        from_name: system_config_string(smtp_from_name.as_ref())
+            .unwrap_or_else(|| "Aether".to_string()),
+    }))
+}
+
+pub(crate) async fn send_smtp_email(
+    config: SmtpDeliveryConfig,
+    email: ComposedEmail,
+) -> Result<(), GatewayError> {
+    tokio::task::spawn_blocking(move || send_smtp_email_blocking(config, email))
+        .await
+        .map_err(|err| GatewayError::Internal(err.to_string()))?
+}
+
+pub(crate) async fn probe_smtp_connection(config: SmtpDeliveryConfig) -> Result<(), GatewayError> {
+    tokio::task::spawn_blocking(move || probe_smtp_connection_blocking(config))
+        .await
+        .map_err(|err| GatewayError::Internal(err.to_string()))?
+}
+
+pub(crate) fn system_config_u16(value: Option<&serde_json::Value>, default: u16) -> u16 {
+    match value {
+        Some(serde_json::Value::Number(value)) => value
+            .as_u64()
+            .and_then(|value| u16::try_from(value).ok())
+            .unwrap_or(default),
+        Some(serde_json::Value::String(value)) => value.trim().parse::<u16>().unwrap_or(default),
+        _ => default,
+    }
+}
+
+fn encode_mime_header(value: &str) -> String {
+    if value.is_ascii() {
+        return value.to_string();
+    }
+    format!(
+        "=?UTF-8?B?{}?=",
+        base64::engine::general_purpose::STANDARD.encode(value.as_bytes())
+    )
+}
+
+fn wrap_base64(value: &str) -> String {
+    let mut wrapped = String::new();
+    for chunk in value.as_bytes().chunks(76) {
+        wrapped.push_str(std::str::from_utf8(chunk).unwrap_or_default());
+        wrapped.push_str("\r\n");
+    }
+    wrapped
+}
+
+fn build_tls_config() -> std::sync::Arc<rustls::ClientConfig> {
+    let _ = rustls::crypto::ring::default_provider().install_default();
+    let root_store =
+        rustls::RootCertStore::from_iter(webpki_roots::TLS_SERVER_ROOTS.iter().cloned());
+    let config = rustls::ClientConfig::builder()
+        .with_root_certificates(root_store)
+        .with_no_client_auth();
+    std::sync::Arc::new(config)
+}
+
+fn resolve_server_name(host: &str) -> Result<rustls::pki_types::ServerName<'static>, GatewayError> {
+    let host = host.trim().trim_start_matches('[').trim_end_matches(']');
+    if let Ok(ip) = host.parse::<std::net::IpAddr>() {
+        return Ok(rustls::pki_types::ServerName::from(ip));
+    }
+    rustls::pki_types::ServerName::try_from(host.to_string())
+        .map_err(|err| GatewayError::Internal(err.to_string()))
+}
+
+fn connect_tcp_stream(config: &SmtpDeliveryConfig) -> Result<std::net::TcpStream, GatewayError> {
+    let stream = std::net::TcpStream::connect((config.host.as_str(), config.port))
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    stream
+        .set_read_timeout(Some(std::time::Duration::from_secs(SMTP_TIMEOUT_SECS)))
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    stream
+        .set_write_timeout(Some(std::time::Duration::from_secs(SMTP_TIMEOUT_SECS)))
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    Ok(stream)
+}
+
+fn wrap_tls_stream(
+    stream: std::net::TcpStream,
+    host: &str,
+) -> Result<rustls::StreamOwned<rustls::ClientConnection, std::net::TcpStream>, GatewayError> {
+    let server_name = resolve_server_name(host)?;
+    let connection = rustls::ClientConnection::new(build_tls_config(), server_name)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    Ok(rustls::StreamOwned::new(connection, stream))
+}
+
+fn smtp_read_response<T: std::io::BufRead>(reader: &mut T) -> Result<(u16, String), GatewayError> {
+    let mut message = String::new();
+    let code = loop {
+        let parsed_code;
+        let continuation;
+        let trimmed;
+        {
+            let mut line = String::new();
+            let bytes = reader
+                .read_line(&mut line)
+                .map_err(|err| GatewayError::Internal(err.to_string()))?;
+            if bytes == 0 {
+                return Err(GatewayError::Internal(
+                    "smtp connection closed unexpectedly".to_string(),
+                ));
+            }
+            trimmed = line.trim_end_matches(['\r', '\n']).to_string();
+            if trimmed.len() < 3 {
+                return Err(GatewayError::Internal("invalid smtp response".to_string()));
+            }
+            parsed_code = trimmed[..3]
+                .parse::<u16>()
+                .map_err(|err| GatewayError::Internal(err.to_string()))?;
+            continuation = trimmed.as_bytes().get(3).copied() == Some(b'-');
+        }
+        if !message.is_empty() {
+            message.push('\n');
+        }
+        message.push_str(&trimmed);
+        if !continuation {
+            break parsed_code;
+        }
+    };
+    Ok((code, message))
+}
+
+fn smtp_expect<T: std::io::BufRead>(
+    reader: &mut T,
+    allowed_codes: &[u16],
+) -> Result<String, GatewayError> {
+    let (code, message) = smtp_read_response(reader)?;
+    if allowed_codes.contains(&code) {
+        return Ok(message);
+    }
+    Err(GatewayError::Internal(format!(
+        "unexpected smtp response {code}: {message}"
+    )))
+}
+
+fn smtp_write_line<T: std::io::Write>(writer: &mut T, line: &str) -> Result<(), GatewayError> {
+    writer
+        .write_all(line.as_bytes())
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    writer
+        .write_all(b"\r\n")
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    writer
+        .flush()
+        .map_err(|err| GatewayError::Internal(err.to_string()))
+}
+
+fn smtp_send_command<S: std::io::Read + std::io::Write>(
+    reader: &mut std::io::BufReader<S>,
+    command: &str,
+    allowed_codes: &[u16],
+) -> Result<String, GatewayError> {
+    smtp_write_line(reader.get_mut(), command)?;
+    smtp_expect(reader, allowed_codes)
+}
+
+fn build_email_message(config: &SmtpDeliveryConfig, email: &ComposedEmail) -> String {
+    let boundary = format!("aether-{}", uuid::Uuid::new_v4().simple());
+    let text_body =
+        wrap_base64(&base64::engine::general_purpose::STANDARD.encode(email.text_body.as_bytes()));
+    let html_body =
+        wrap_base64(&base64::engine::general_purpose::STANDARD.encode(email.html_body.as_bytes()));
+    let from_header = if config.from_name.trim().is_empty() {
+        format!("<{}>", config.from_email)
+    } else {
+        format!(
+            "{} <{}>",
+            encode_mime_header(config.from_name.trim()),
+            config.from_email
+        )
+    };
+    format!(
+        "From: {from_header}\r\nTo: <{to_email}>\r\nSubject: {subject}\r\nMIME-Version: 1.0\r\nContent-Type: multipart/alternative; boundary=\"{boundary}\"\r\n\r\n--{boundary}\r\nContent-Type: text/plain; charset=\"utf-8\"\r\nContent-Transfer-Encoding: base64\r\n\r\n{text_body}--{boundary}\r\nContent-Type: text/html; charset=\"utf-8\"\r\nContent-Transfer-Encoding: base64\r\n\r\n{html_body}--{boundary}--\r\n",
+        to_email = email.to_email,
+        subject = encode_mime_header(&email.subject),
+    )
+}
+
+fn smtp_authenticate<S: std::io::Read + std::io::Write>(
+    reader: &mut std::io::BufReader<S>,
+    config: &SmtpDeliveryConfig,
+) -> Result<(), GatewayError> {
+    let Some(username) = config
+        .user
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+    else {
+        return Ok(());
+    };
+    let password = config.password.as_deref().unwrap_or("");
+    smtp_send_command(reader, "AUTH LOGIN", &[334])?;
+    smtp_send_command(
+        reader,
+        &base64::engine::general_purpose::STANDARD.encode(username.as_bytes()),
+        &[334],
+    )?;
+    smtp_send_command(
+        reader,
+        &base64::engine::general_purpose::STANDARD.encode(password.as_bytes()),
+        &[235],
+    )?;
+    Ok(())
+}
+
+fn smtp_deliver_message<S: std::io::Read + std::io::Write>(
+    reader: &mut std::io::BufReader<S>,
+    config: &SmtpDeliveryConfig,
+    email: &ComposedEmail,
+) -> Result<(), GatewayError> {
+    smtp_send_command(
+        reader,
+        &format!("MAIL FROM:<{}>", config.from_email),
+        &[250],
+    )?;
+    smtp_send_command(
+        reader,
+        &format!("RCPT TO:<{}>", email.to_email),
+        &[250, 251],
+    )?;
+    smtp_send_command(reader, "DATA", &[354])?;
+    let message = build_email_message(config, email);
+    reader
+        .get_mut()
+        .write_all(message.as_bytes())
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    reader
+        .get_mut()
+        .write_all(b"\r\n.\r\n")
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    reader
+        .get_mut()
+        .flush()
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let _ = smtp_expect(reader, &[250])?;
+    let _ = smtp_send_command(reader, "QUIT", &[221]);
+    Ok(())
+}
+
+fn smtp_send_message<S: std::io::Read + std::io::Write>(
+    reader: &mut std::io::BufReader<S>,
+    config: &SmtpDeliveryConfig,
+    email: &ComposedEmail,
+) -> Result<(), GatewayError> {
+    smtp_send_command(reader, "EHLO aether.local", &[250])?;
+    smtp_authenticate(reader, config)?;
+    smtp_deliver_message(reader, config, email)
+}
+
+fn smtp_probe_connection<S: std::io::Read + std::io::Write>(
+    reader: &mut std::io::BufReader<S>,
+    config: &SmtpDeliveryConfig,
+) -> Result<(), GatewayError> {
+    smtp_send_command(reader, "EHLO aether.local", &[250])?;
+    smtp_authenticate(reader, config)?;
+    let _ = smtp_send_command(reader, "QUIT", &[221]);
+    Ok(())
+}
+
+fn send_smtp_email_blocking(
+    config: SmtpDeliveryConfig,
+    email: ComposedEmail,
+) -> Result<(), GatewayError> {
+    if config.use_ssl {
+        let stream = connect_tcp_stream(&config)?;
+        let tls_stream = wrap_tls_stream(stream, &config.host)?;
+        let mut reader = std::io::BufReader::new(tls_stream);
+        let _ = smtp_expect(&mut reader, &[220])?;
+        return smtp_send_message(&mut reader, &config, &email);
+    }
+
+    let stream = connect_tcp_stream(&config)?;
+    let mut reader = std::io::BufReader::new(stream);
+    let _ = smtp_expect(&mut reader, &[220])?;
+    let _ = smtp_send_command(&mut reader, "EHLO aether.local", &[250])?;
+    if config.use_tls {
+        let _ = smtp_send_command(&mut reader, "STARTTLS", &[220])?;
+        let stream = reader.into_inner();
+        let tls_stream = wrap_tls_stream(stream, &config.host)?;
+        let mut reader = std::io::BufReader::new(tls_stream);
+        return smtp_send_message(&mut reader, &config, &email);
+    }
+
+    smtp_authenticate(&mut reader, &config)?;
+    smtp_deliver_message(&mut reader, &config, &email)
+}
+
+fn probe_smtp_connection_blocking(config: SmtpDeliveryConfig) -> Result<(), GatewayError> {
+    if config.use_ssl {
+        let stream = connect_tcp_stream(&config)?;
+        let tls_stream = wrap_tls_stream(stream, &config.host)?;
+        let mut reader = std::io::BufReader::new(tls_stream);
+        let _ = smtp_expect(&mut reader, &[220])?;
+        return smtp_probe_connection(&mut reader, &config);
+    }
+
+    let stream = connect_tcp_stream(&config)?;
+    let mut reader = std::io::BufReader::new(stream);
+    let _ = smtp_expect(&mut reader, &[220])?;
+    let _ = smtp_send_command(&mut reader, "EHLO aether.local", &[250])?;
+    if config.use_tls {
+        let _ = smtp_send_command(&mut reader, "STARTTLS", &[220])?;
+        let stream = reader.into_inner();
+        let tls_stream = wrap_tls_stream(stream, &config.host)?;
+        let mut reader = std::io::BufReader::new(tls_stream);
+        return smtp_probe_connection(&mut reader, &config);
+    }
+
+    smtp_authenticate(&mut reader, &config)?;
+    let _ = smtp_send_command(&mut reader, "QUIT", &[221]);
+    Ok(())
+}

apps/aether-gateway/src/execution_runtime/mod.rs

@@ -20,6 +20,7 @@ mod stream_pump;
 pub(crate) mod submission;
 pub(crate) mod sync;
 pub(crate) mod transport;
+mod windsurf;
 
 pub(crate) use self::chatgpt_web_image::maybe_execute_chatgpt_web_image_sync;
 pub(crate) use self::constants::{

apps/aether-gateway/src/execution_runtime/stream/execution_failures.rs

@@ -126,6 +126,54 @@ pub(super) fn build_stream_failure_from_execution_error(
     }
 }
 
+pub(super) fn build_stream_failure_from_provider_error_body(
+    status_code: u16,
+    body_json: &Value,
+) -> StreamFailureReport {
+    let body_object = body_json.as_object();
+    let error_object = body_object
+        .and_then(|object| object.get("error"))
+        .and_then(Value::as_object);
+    let error_type =
+        first_non_empty_error_text(error_object, body_object, &["type", "code", "status"])
+            .unwrap_or_else(|| "upstream_error".to_string());
+    let error_message = first_non_empty_error_text(
+        error_object,
+        body_object,
+        &["message", "detail", "reason", "status", "type", "code"],
+    )
+    .unwrap_or_else(|| format!("upstream stream returned error status {status_code}"));
+
+    StreamFailureReport {
+        status_code,
+        error_type,
+        error_message,
+        extra_error_fields: Map::new(),
+    }
+}
+
+fn first_non_empty_error_text(
+    error_object: Option<&Map<String, Value>>,
+    body_object: Option<&Map<String, Value>>,
+    keys: &[&str],
+) -> Option<String> {
+    for object in [error_object, body_object].into_iter().flatten() {
+        for key in keys {
+            let Some(value) = object.get(*key) else {
+                continue;
+            };
+            match value {
+                Value::String(text) if !text.trim().is_empty() => {
+                    return Some(text.trim().to_string());
+                }
+                Value::Number(number) => return Some(number.to_string()),
+                _ => {}
+            }
+        }
+    }
+    None
+}
+
 fn build_stream_failure_sync_payload(
     trace_id: &str,
     report_kind: String,
@@ -296,6 +344,49 @@ async fn record_stream_sync_failure(
     .await;
 }
 
+#[allow(clippy::too_many_arguments)] // internal helper for prefetch error handling
+pub(super) async fn handle_prefetch_provider_private_stream_error(
+    state: &AppState,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    plan: &ExecutionPlan,
+    report_context: Option<Value>,
+    request_id: &str,
+    candidate_id: Option<&str>,
+    report_kind: &str,
+    mut headers: std::collections::BTreeMap<String, String>,
+    telemetry: Option<ExecutionTelemetry>,
+    buffered_body: &[u8],
+    status_code: u16,
+    body_json: Value,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    headers.remove("content-encoding");
+    headers.remove("content-length");
+    headers.insert("content-type".to_string(), "application/json".to_string());
+
+    let payload = GatewaySyncReportRequest {
+        trace_id: trace_id.to_string(),
+        report_kind: report_kind.to_string(),
+        report_context,
+        status_code,
+        headers,
+        body_json: Some(body_json),
+        client_body_json: None,
+        body_base64: (!buffered_body.is_empty())
+            .then(|| base64::engine::general_purpose::STANDARD.encode(buffered_body)),
+        telemetry,
+    };
+    record_stream_sync_failure(state, plan, payload.report_context.as_ref(), &payload, None).await;
+
+    let response =
+        submit_local_core_error_or_sync_finalize(state, trace_id, decision, payload).await?;
+    Ok(Some(attach_control_metadata_headers(
+        response,
+        Some(request_id),
+        candidate_id,
+    )?))
+}
+
 #[allow(clippy::too_many_arguments)] // internal helper for prefetch error handling
 pub(super) async fn handle_prefetch_stream_failure(
     state: &AppState,

apps/aether-gateway/src/execution_runtime/submission.rs

@@ -1,9 +1,7 @@
-#[cfg(test)]
-use crate::ai_serving::api::core_success_background_report_kind;
 use crate::ai_serving::api::{
     build_core_error_body_for_client_format, core_error_background_report_kind,
-    core_error_default_client_api_format, is_core_error_finalize_kind,
-    maybe_compile_sync_finalize_response,
+    core_error_default_client_api_format, core_success_background_report_kind,
+    is_core_error_finalize_kind, maybe_compile_sync_finalize_response,
     normalize_provider_private_response_value as unwrap_local_finalize_response_value,
     LocalCoreSyncErrorKind,
 };
@@ -36,6 +34,12 @@ pub(super) fn maybe_build_local_core_error_response(
         return Ok(None);
     };
     let status_source_json = resolve_local_sync_source_body_json(payload)?;
+    if payload.status_code < 400
+        && !has_nested_error(&response_body_json)
+        && !status_source_json.as_ref().is_some_and(has_nested_error)
+    {
+        return Ok(None);
+    }
 
     let mut response_headers = payload.headers.clone();
     response_headers.remove("content-encoding");
@@ -81,8 +85,9 @@ fn build_local_sync_response_from_json(
     payload: &GatewaySyncReportRequest,
     body_json: serde_json::Value,
 ) -> Result<Response<Body>, GatewayError> {
-    let status_code = if is_core_error_finalize_kind(payload.report_kind.as_str())
-        || has_nested_error(&body_json)
+    let body_has_error = has_nested_error(&body_json);
+    let status_code = if body_has_error
+        || (payload.status_code >= 400 && is_core_error_finalize_kind(payload.report_kind.as_str()))
     {
         resolve_local_sync_error_status_code(payload.status_code, &body_json)
     } else {
@@ -367,7 +372,10 @@ pub(crate) fn resolve_core_success_background_report_kind(report_kind: &str) ->
     core_success_background_report_kind(report_kind).map(ToOwned::to_owned)
 }
 
-fn resolve_local_sync_error_status_code(status_code: u16, body_json: &serde_json::Value) -> u16 {
+pub(crate) fn resolve_local_sync_error_status_code(
+    status_code: u16,
+    body_json: &serde_json::Value,
+) -> u16 {
     if (400..600).contains(&status_code) {
         return status_code;
     }
@@ -621,11 +629,30 @@ pub(crate) async fn submit_local_core_error_or_sync_finalize(
         build_local_core_sync_finalize_fallback_response(trace_id, decision, &payload)?
     };
 
-    if let Some(error_report_kind) =
+    let response_status = response.status();
+    if response_status.is_success() {
+        if let Some(success_report_kind) =
+            core_success_background_report_kind(payload.report_kind.as_str())
+        {
+            let mut report_payload = payload.clone();
+            report_payload.report_kind = success_report_kind.to_string();
+            report_payload.status_code = response_status.as_u16();
+            spawn_sync_report(state.clone(), report_payload);
+        } else {
+            warn!(
+                event_name = "local_core_finalize_missing_success_report_mapping",
+                log_type = "event",
+                trace_id = %trace_id,
+                report_kind = %payload.report_kind,
+                "gateway built local core finalize success response without background success report mapping"
+            );
+        }
+    } else if let Some(error_report_kind) =
         resolve_core_error_background_report_kind(payload.report_kind.as_str())
     {
         let mut report_payload = payload.clone();
         report_payload.report_kind = error_report_kind;
+        report_payload.status_code = response_status.as_u16();
         spawn_sync_report(state.clone(), report_payload);
     } else {
         warn!(
@@ -819,4 +846,39 @@ mod tests {
             "unexpected message: {message}"
         );
     }
+
+    #[tokio::test]
+    async fn submit_local_core_finalize_keeps_http_200_for_success_image_body() {
+        let payload = core_finalize_payload(
+            "openai_image_sync_finalize",
+            "openai:image",
+            "openai:image",
+            200,
+            json!({
+                "created": 1779273523,
+                "data": [{
+                    "b64_json": "aGVsbG8="
+                }]
+            }),
+        );
+
+        let state = AppState::new().expect("state should build");
+        let response = submit_local_core_error_or_sync_finalize(
+            &state,
+            "trace-image-success-200",
+            &test_decision(),
+            payload,
+        )
+        .await
+        .expect("response should build");
+
+        assert_eq!(response.status(), http::StatusCode::OK);
+        let body: serde_json::Value = serde_json::from_slice(
+            &to_bytes(response.into_body(), usize::MAX)
+                .await
+                .expect("body should read"),
+        )
+        .expect("body should decode");
+        assert_eq!(body["data"][0]["b64_json"], "aGVsbG8=");
+    }
 }

apps/aether-gateway/src/execution_runtime/sync/execution.rs

@@ -3,6 +3,7 @@ use std::io::Error as IoError;
 use std::sync::Arc;
 use std::time::{Duration, Instant};
 
+use aether_ai_serving::UPSTREAM_IS_STREAM_KEY;
 use aether_contracts::{
     ExecutionError, ExecutionErrorKind, ExecutionPhase, ExecutionPlan, ExecutionResult,
     ExecutionTelemetry,
@@ -20,7 +21,6 @@ use async_stream::stream;
 use axum::body::{to_bytes, Body, Bytes};
 use axum::http::header::{CACHE_CONTROL, CONTENT_ENCODING, CONTENT_LENGTH, CONTENT_TYPE};
 use axum::http::{HeaderName, HeaderValue, Response, StatusCode};
-use base64::Engine as _;
 use futures_util::StreamExt;
 use serde_json::{json, Value};
 use tokio::sync::mpsc;
@@ -29,8 +29,9 @@ use tokio::time::MissedTickBehavior;
 use tracing::{debug, warn};
 
 use crate::ai_serving::api::{
-    build_core_error_body_for_client_format, implicit_sync_finalize_report_kind,
-    maybe_build_sync_finalize_outcome, LocalCoreSyncErrorKind, LocalCoreSyncFinalizeOutcome,
+    build_core_error_body_for_client_format, extract_provider_private_stream_error_body,
+    implicit_sync_finalize_report_kind, maybe_build_sync_finalize_outcome, LocalCoreSyncErrorKind,
+    LocalCoreSyncFinalizeOutcome,
 };
 use crate::api::response::{
     attach_control_metadata_headers, build_client_response, build_client_response_from_parts,
@@ -43,12 +44,16 @@ use crate::execution_runtime::grok::maybe_execute_grok_sync;
 use crate::execution_runtime::oauth_retry::refresh_oauth_plan_auth_for_retry;
 #[cfg(test)]
 use crate::execution_runtime::remote_compat::post_sync_plan_to_remote_execution_runtime;
-use crate::execution_runtime::submission::submit_local_core_error_or_sync_finalize;
-use crate::execution_runtime::transport::{
-    build_request_body, collect_response_headers, decode_response_body_bytes,
-    format_upstream_request_error, format_wreq_upstream_request_error, response_body_is_json,
-    send_request, DirectHttpResponse, DirectSyncExecutionRuntime, ExecutionRuntimeTransportError,
+use crate::execution_runtime::submission::{
+    resolve_local_sync_error_status_code, submit_local_core_error_or_sync_finalize,
 };
+use crate::execution_runtime::transport::{
+    build_execution_response_body, build_request_body, collect_response_headers,
+    decode_response_body_bytes, format_upstream_request_error, format_wreq_upstream_request_error,
+    response_body_is_json, send_request, DirectHttpResponse, DirectSyncExecutionRuntime,
+    ExecutionRuntimeTransportError,
+};
+use crate::execution_runtime::windsurf::maybe_execute_windsurf_sync;
 use crate::execution_runtime::{
     analyze_local_candidate_failover_sync, apply_endpoint_response_header_rules,
     attach_provider_response_headers_to_report_context, local_failover_response_text,
@@ -398,6 +403,43 @@ fn build_invalid_provider_success_body(
     )
 }
 
+fn provider_private_error_details(body_json: &Value) -> (Option<String>, Option<String>) {
+    let body_object = body_json.as_object();
+    let error_object = body_object
+        .and_then(|object| object.get("error"))
+        .and_then(Value::as_object);
+    let error_type =
+        first_non_empty_error_text(error_object, body_object, &["type", "code", "status"]);
+    let error_message = first_non_empty_error_text(
+        error_object,
+        body_object,
+        &["message", "detail", "reason", "status", "type", "code"],
+    );
+    (error_type, error_message)
+}
+
+fn first_non_empty_error_text(
+    error_object: Option<&serde_json::Map<String, Value>>,
+    body_object: Option<&serde_json::Map<String, Value>>,
+    keys: &[&str],
+) -> Option<String> {
+    for object in [error_object, body_object].into_iter().flatten() {
+        for key in keys {
+            let Some(value) = object.get(*key) else {
+                continue;
+            };
+            match value {
+                Value::String(text) if !text.trim().is_empty() => {
+                    return Some(text.trim().to_string());
+                }
+                Value::Number(number) => return Some(number.to_string()),
+                _ => {}
+            }
+        }
+    }
+    None
+}
+
 #[derive(Debug, Clone)]
 struct OpenAiImageSyncProgressSnapshot {
     phase: &'static str,
@@ -788,6 +830,12 @@ async fn execute_direct_sync_runtime_candidate(
     candidate_index: &str,
     progress_snapshot: Option<Arc<Mutex<OpenAiImageSyncProgressSnapshot>>>,
 ) -> Result<ExecutionResult, SyncExecutionFailure> {
+    if let Some(result) = maybe_execute_windsurf_sync(state, plan, report_context)
+        .await
+        .map_err(SyncExecutionFailure::from_transport)?
+    {
+        return Ok(result);
+    }
     if !should_track_openai_image_sync_upstream_sse(plan_kind, plan, report_context) {
         return DirectSyncExecutionRuntime::new()
             .execute_sync(plan)
@@ -945,27 +993,9 @@ async fn execute_openai_image_sync_upstream_sse_candidate(
     let upstream_bytes = body_bytes.len() as u64;
     progress.finish(status_code, elapsed_ms).await;
 
-    let body = if body_bytes.is_empty() {
-        None
-    } else if plan.stream {
-        Some(aether_contracts::ResponseBody {
-            json_body: None,
-            body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(&body_bytes)),
-        })
-    } else if response_body_is_json(&headers, &decoded_body_bytes) {
-        let body_json: Value = serde_json::from_slice(&decoded_body_bytes)
-            .map_err(ExecutionRuntimeTransportError::InvalidJson)
+    let body =
+        build_execution_response_body(&headers, &body_bytes, &decoded_body_bytes, plan.stream)
             .map_err(SyncExecutionFailure::from_transport)?;
-        Some(aether_contracts::ResponseBody {
-            json_body: Some(body_json),
-            body_bytes_b64: None,
-        })
-    } else {
-        Some(aether_contracts::ResponseBody {
-            json_body: None,
-            body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(&body_bytes)),
-        })
-    };
 
     Ok(ExecutionResult {
         request_id: plan.request_id.clone(),
@@ -1005,7 +1035,7 @@ fn resolve_openai_image_sync_total_timeout_ms(plan: &ExecutionPlan) -> u64 {
 
 fn report_context_upstream_is_stream(report_context: Option<&Value>) -> bool {
     report_context
-        .and_then(|value| value.get("upstream_is_stream"))
+        .and_then(|value| value.get(UPSTREAM_IS_STREAM_KEY))
         .and_then(Value::as_bool)
         .unwrap_or(false)
 }
@@ -1696,8 +1726,21 @@ async fn execute_execution_runtime_sync_impl(
                 headers.insert("content-type".to_string(), "application/json".to_string());
             }
         }
-        let (result_error_type, result_error_message) =
+        let (mut result_error_type, mut result_error_message) =
             execution_error_details(result.error.as_ref(), body_json.as_ref());
+        if result.status_code < 400 && body_json.is_none() {
+            if let Some(error_body_json) =
+                extract_provider_private_stream_error_body(report_context.as_ref(), &body_bytes)
+            {
+                result.status_code =
+                    resolve_local_sync_error_status_code(result.status_code, &error_body_json);
+                let (private_error_type, private_error_message) =
+                    provider_private_error_details(&error_body_json);
+                result_error_type = private_error_type.or(result_error_type);
+                result_error_message = private_error_message.or(result_error_message);
+                body_json = Some(error_body_json);
+            }
+        }
         let local_failover_response_text = local_failover_response_text(
             body_json.as_ref(),
             &body_bytes,

apps/aether-gateway/src/execution_runtime/transport.rs

@@ -25,8 +25,10 @@ use serde_json::json;
 use serde_json::Value;
 use thiserror::Error;
 
+use crate::ai_serving::api::extract_provider_private_stream_error_body;
 #[cfg(test)]
 use crate::execution_runtime::remote_compat::execute_sync_plan_via_remote_execution_runtime;
+use crate::execution_runtime::windsurf::maybe_execute_windsurf_sync;
 use crate::frontdoor_loop_guard::{
     configured_gateway_frontdoor_base_url, gateway_frontdoor_self_loop_guard_error,
 };
@@ -232,26 +234,8 @@ impl DirectSyncExecutionRuntime {
         let elapsed_ms = started_at.elapsed().as_millis() as u64;
         let upstream_bytes = body_bytes.len() as u64;
 
-        let body = if body_bytes.is_empty() {
-            None
-        } else if plan.stream {
-            Some(ResponseBody {
-                json_body: None,
-                body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(&body_bytes)),
-            })
-        } else if response_body_is_json(&headers, &decoded_body_bytes) {
-            let body_json: Value = serde_json::from_slice(&decoded_body_bytes)
-                .map_err(ExecutionRuntimeTransportError::InvalidJson)?;
-            Some(ResponseBody {
-                json_body: Some(body_json),
-                body_bytes_b64: None,
-            })
-        } else {
-            Some(ResponseBody {
-                json_body: None,
-                body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(&body_bytes)),
-            })
-        };
+        let body =
+            build_execution_response_body(&headers, &body_bytes, &decoded_body_bytes, plan.stream)?;
 
         Ok(ExecutionResult {
             request_id: plan.request_id.clone(),
@@ -347,6 +331,11 @@ pub(crate) async fn execute_sync_plan_with_report_context(
     }
 
     let _ = trace_id;
+    match maybe_execute_windsurf_sync(state, plan, None).await {
+        Ok(Some(result)) => return Ok(result),
+        Ok(None) => {}
+        Err(err) => return Err(GatewayError::Internal(err.to_string())),
+    }
     match DirectSyncExecutionRuntime::new().execute_sync(plan).await {
         Ok(result) => {
             record_manual_proxy_request_outcome(state, plan, result.status_code).await;
@@ -567,26 +556,8 @@ async fn execute_sync_plan_via_local_tunnel(
         );
     }
 
-    let body = if body_bytes.is_empty() {
-        None
-    } else if plan.stream {
-        Some(ResponseBody {
-            json_body: None,
-            body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(&body_bytes)),
-        })
-    } else if response_body_is_json(&headers, &decoded_body_bytes) {
-        let body_json: Value = serde_json::from_slice(&decoded_body_bytes)
-            .map_err(ExecutionRuntimeTransportError::InvalidJson)?;
-        Some(ResponseBody {
-            json_body: Some(body_json),
-            body_bytes_b64: None,
-        })
-    } else {
-        Some(ResponseBody {
-            json_body: None,
-            body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(&body_bytes)),
-        })
-    };
+    let body =
+        build_execution_response_body(&headers, &body_bytes, &decoded_body_bytes, plan.stream)?;
 
     Ok(ExecutionResult {
         request_id: plan.request_id.clone(),
@@ -1470,17 +1441,63 @@ pub(crate) fn decode_response_body_bytes(
 }
 
 pub(crate) fn response_body_is_json(headers: &BTreeMap<String, String>, body_bytes: &[u8]) -> bool {
-    if headers
+    let content_type = headers
         .get("content-type")
         .map(|value| value.to_ascii_lowercase())
-        .is_some_and(|value| value.contains("json"))
+        .unwrap_or_default();
+    if content_type.contains("application/connect+json")
+        || content_type.contains("application/connect+proto")
     {
+        return false;
+    }
+    if content_type.contains("json") {
         return true;
     }
 
     serde_json::from_slice::<Value>(body_bytes).is_ok()
 }
 
+pub(crate) fn build_execution_response_body(
+    headers: &BTreeMap<String, String>,
+    body_bytes: &[u8],
+    decoded_body_bytes: &[u8],
+    stream: bool,
+) -> Result<Option<ResponseBody>, ExecutionRuntimeTransportError> {
+    if body_bytes.is_empty() {
+        return Ok(None);
+    }
+
+    if let Some(body_json) = extract_provider_private_stream_error_body(None, decoded_body_bytes)
+        .or_else(|| extract_provider_private_stream_error_body(None, body_bytes))
+    {
+        return Ok(Some(ResponseBody {
+            json_body: Some(body_json),
+            body_bytes_b64: None,
+        }));
+    }
+
+    if stream {
+        return Ok(Some(ResponseBody {
+            json_body: None,
+            body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(body_bytes)),
+        }));
+    }
+
+    if response_body_is_json(headers, decoded_body_bytes) {
+        let body_json: Value = serde_json::from_slice(decoded_body_bytes)
+            .map_err(ExecutionRuntimeTransportError::InvalidJson)?;
+        return Ok(Some(ResponseBody {
+            json_body: Some(body_json),
+            body_bytes_b64: None,
+        }));
+    }
+
+    Ok(Some(ResponseBody {
+        json_body: None,
+        body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(body_bytes)),
+    }))
+}
+
 #[cfg(test)]
 mod tests {
     use std::collections::BTreeMap;
@@ -1505,11 +1522,12 @@ mod tests {
     use tokio::sync::watch;
 
     use super::{
-        build_browser_wreq_client, build_client, build_request_headers, execute_sync_plan,
-        record_manual_proxy_request_failure, record_manual_proxy_request_outcome,
-        record_manual_proxy_request_success, record_manual_proxy_stream_error,
-        resolve_execution_transport_controls, DirectSyncExecutionRuntime,
-        ExecutionRuntimeTransportError, ExecutionTransportControls,
+        build_browser_wreq_client, build_client, build_execution_response_body,
+        build_request_headers, execute_sync_plan, record_manual_proxy_request_failure,
+        record_manual_proxy_request_outcome, record_manual_proxy_request_success,
+        record_manual_proxy_stream_error, resolve_execution_transport_controls,
+        response_body_is_json, DirectSyncExecutionRuntime, ExecutionRuntimeTransportError,
+        ExecutionTransportControls,
     };
     use crate::constants::{
         EXECUTION_RUNTIME_LOOP_GUARD_HEADER, EXECUTION_RUNTIME_LOOP_GUARD_VIA_TOKEN,
@@ -2764,6 +2782,41 @@ mod tests {
         ));
     }
 
+    #[test]
+    fn connect_json_response_is_not_treated_as_plain_json() {
+        let headers = BTreeMap::from([(
+            "content-type".to_string(),
+            "application/connect+json".to_string(),
+        )]);
+        let body = [2, 0, 0, 0, 2, b'{', b'}'];
+
+        assert!(!response_body_is_json(&headers, &body));
+    }
+
+    #[test]
+    fn connect_json_error_response_is_decoded_for_stream_sync_body() {
+        let headers = BTreeMap::from([(
+            "content-type".to_string(),
+            "application/connect+json".to_string(),
+        )]);
+        let payload = br#"{"error":{"code":"resource_exhausted","message":"quota exhausted"}}"#;
+        let mut body_bytes = vec![2];
+        body_bytes.extend_from_slice(&(payload.len() as u32).to_be_bytes());
+        body_bytes.extend_from_slice(payload);
+
+        let body = build_execution_response_body(&headers, &body_bytes, &body_bytes, true)
+            .expect("body should build")
+            .expect("body should be present");
+
+        assert_eq!(
+            body.json_body
+                .as_ref()
+                .and_then(|value| value.pointer("/error/code")),
+            Some(&json!("resource_exhausted"))
+        );
+        assert!(body.body_bytes_b64.is_none());
+    }
+
     #[tokio::test]
     async fn direct_sync_execution_runtime_compresses_json_body_when_requested() {
         let listener = crate::test_support::bind_loopback_listener()

apps/aether-gateway/src/executor/candidate_loop.rs

@@ -1,5 +1,6 @@
 use aether_ai_serving::{
     run_ai_attempt_loop, AiAttemptLoopOutcome, AiAttemptLoopPort, AiExecutionAttempt,
+    UPSTREAM_IS_STREAM_KEY,
 };
 use aether_data_contracts::repository::candidates::RequestCandidateStatus;
 use aether_scheduler_core::{
@@ -471,11 +472,24 @@ fn should_skip_unused_persistence_from_metadata(
     metadata.candidate_group_id.is_some() && metadata.pool_key_index.is_some()
 }
 
-fn resolve_stream_candidate_watchdog_timeout(plan: &aether_contracts::ExecutionPlan) -> Duration {
+fn resolve_stream_candidate_watchdog_timeout(
+    plan: &aether_contracts::ExecutionPlan,
+    report_context: Option<&serde_json::Value>,
+) -> Duration {
+    let upstream_is_stream = report_context
+        .and_then(|context| context.get(UPSTREAM_IS_STREAM_KEY))
+        .and_then(serde_json::Value::as_bool)
+        .unwrap_or(true);
     let timeout_ms = plan
         .timeouts
         .as_ref()
-        .and_then(|timeouts| timeouts.first_byte_ms.or(timeouts.total_ms))
+        .and_then(|timeouts| {
+            if upstream_is_stream {
+                timeouts.first_byte_ms.or(timeouts.total_ms)
+            } else {
+                timeouts.total_ms.or(timeouts.first_byte_ms)
+            }
+        })
         .unwrap_or(DEFAULT_STREAM_CANDIDATE_WATCHDOG_TIMEOUT_MS)
         .max(1);
     Duration::from_millis(timeout_ms)
@@ -493,7 +507,7 @@ where
     Fut:
         std::future::Future<Output = Result<Option<Response<Body>>, GatewayError>> + Send + 'static,
 {
-    let timeout_duration = resolve_stream_candidate_watchdog_timeout(plan);
+    let timeout_duration = resolve_stream_candidate_watchdog_timeout(plan, report_context);
     let candidate_started_unix_ms = current_unix_ms();
     let mut join_handle = tokio::spawn(execute());
     match timeout(timeout_duration, &mut join_handle).await {
@@ -655,19 +669,22 @@ mod tests {
 
     #[test]
     fn stream_candidate_watchdog_prefers_first_byte_timeout() {
-        let timeout =
-            resolve_stream_candidate_watchdog_timeout(&test_plan(Some(ExecutionTimeouts {
+        let report_context = json!({"upstream_is_stream": true});
+        let timeout = resolve_stream_candidate_watchdog_timeout(
+            &test_plan(Some(ExecutionTimeouts {
                 first_byte_ms: Some(12_345),
                 total_ms: Some(90_000),
                 ..ExecutionTimeouts::default()
-            })));
+            })),
+            Some(&report_context),
+        );
 
         assert_eq!(timeout, Duration::from_millis(12_345));
     }
 
     #[test]
     fn stream_candidate_watchdog_uses_default_when_timeouts_missing() {
-        let timeout = resolve_stream_candidate_watchdog_timeout(&test_plan(None));
+        let timeout = resolve_stream_candidate_watchdog_timeout(&test_plan(None), None);
 
         assert_eq!(
             timeout,
@@ -675,6 +692,50 @@ mod tests {
         );
     }
 
+    #[test]
+    fn stream_candidate_watchdog_prefers_total_timeout_when_upstream_non_stream() {
+        let report_context = json!({"upstream_is_stream": false});
+        let timeout = resolve_stream_candidate_watchdog_timeout(
+            &test_plan(Some(ExecutionTimeouts {
+                first_byte_ms: Some(300_000),
+                total_ms: Some(599_000),
+                ..ExecutionTimeouts::default()
+            })),
+            Some(&report_context),
+        );
+
+        assert_eq!(timeout, Duration::from_millis(599_000));
+    }
+
+    #[test]
+    fn stream_candidate_watchdog_falls_back_to_first_byte_when_upstream_non_stream_lacks_total() {
+        let report_context = json!({"upstream_is_stream": false});
+        let timeout = resolve_stream_candidate_watchdog_timeout(
+            &test_plan(Some(ExecutionTimeouts {
+                first_byte_ms: Some(300_000),
+                ..ExecutionTimeouts::default()
+            })),
+            Some(&report_context),
+        );
+
+        assert_eq!(timeout, Duration::from_millis(300_000));
+    }
+
+    #[test]
+    fn stream_candidate_watchdog_defaults_to_streaming_when_flag_missing() {
+        let report_context = json!({});
+        let timeout = resolve_stream_candidate_watchdog_timeout(
+            &test_plan(Some(ExecutionTimeouts {
+                first_byte_ms: Some(12_345),
+                total_ms: Some(90_000),
+                ..ExecutionTimeouts::default()
+            })),
+            Some(&report_context),
+        );
+
+        assert_eq!(timeout, Duration::from_millis(12_345));
+    }
+
     #[test]
     fn unused_persistence_skips_pool_internal_candidates() {
         assert!(should_skip_unused_persistence(Some(&json!({

apps/aether-gateway/src/handlers/admin/mod.rs

@@ -37,6 +37,7 @@ pub(crate) use self::provider::oauth::runtime::{
     refresh_provider_oauth_account_state_after_update,
 };
 pub(crate) use self::provider::ops::providers::actions::admin_provider_ops_local_action_response;
+pub(crate) use self::provider::ops::providers::store_admin_provider_ops_balance_cache;
 pub(crate) use self::provider::pool::config::admin_provider_pool_config;
 pub(crate) use self::provider::pool_admin::maybe_build_local_admin_pool_response;
 pub(crate) use self::provider::shared::payloads::{

apps/aether-gateway/src/handlers/admin/observability/monitoring/cache_model_mapping.rs

@@ -59,6 +59,11 @@ pub(super) async fn build_admin_monitoring_redis_cache_categories_response(
 ) -> Result<Response<Body>, GatewayError> {
     let mut categories = Vec::with_capacity(ADMIN_MONITORING_REDIS_CACHE_CATEGORIES.len());
     let mut total_keys = 0usize;
+    let diagnostics = state
+        .runtime_state()
+        .redis_diagnostics()
+        .await
+        .map_err(|err| GatewayError::Internal(format!("redis diagnostics failed: {err}")))?;
 
     for (key, name, pattern, description) in ADMIN_MONITORING_REDIS_CACHE_CATEGORIES {
         let count = list_admin_monitoring_namespaced_keys(state, pattern)
@@ -81,6 +86,7 @@ pub(super) async fn build_admin_monitoring_redis_cache_categories_response(
             "backend": state.runtime_state().backend_kind().as_str(),
             "categories": categories,
             "total_keys": total_keys,
+            "diagnostics": diagnostics,
         }
     }))
     .into_response())

apps/aether-gateway/src/handlers/admin/observability/monitoring/tests/mod.rs

@@ -1194,6 +1194,7 @@ async fn admin_monitoring_redis_keys_returns_local_payload_without_redis() {
     assert_eq!(payload["data"]["available"], json!(true));
     assert_eq!(payload["data"]["backend"], json!("memory"));
     assert_eq!(payload["data"]["total_keys"], json!(0));
+    assert_eq!(payload["data"]["diagnostics"], serde_json::Value::Null);
 }
 
 #[tokio::test]

apps/aether-gateway/src/handlers/admin/observability/monitoring/tests/trace.rs

@@ -5,8 +5,11 @@ use super::local_monitoring_response;
 use crate::data::GatewayDataState;
 use crate::AppState;
 use aether_crypto::{encrypt_python_fernet_plaintext, DEVELOPMENT_ENCRYPTION_KEY};
-use aether_data_contracts::repository::candidates::RequestCandidateStatus;
+use aether_data_contracts::repository::{
+    candidates::RequestCandidateStatus, usage::UsageBodyCaptureState,
+};
 use axum::body::to_bytes;
+use base64::{engine::general_purpose::STANDARD as BASE64_STANDARD, Engine as _};
 use serde_json::json;
 use std::sync::Arc;
 
@@ -530,6 +533,89 @@ async fn admin_monitoring_trace_request_exposes_failed_candidate_upstream_respon
     assert!(extra.get("provider_response").is_none());
 }
 
+#[tokio::test]
+async fn admin_monitoring_trace_request_decodes_connect_json_response_body_refs() {
+    let mut candidate = sample_candidate(
+        "cand-used",
+        "request-connect",
+        0,
+        RequestCandidateStatus::Failed,
+        Some(101),
+        Some(33),
+        Some(429),
+    );
+    candidate.extra_data = Some(json!({"cache_1h": true}));
+
+    let request_candidates = Arc::new(InMemoryRequestCandidateRepository::seed(vec![candidate]));
+    let provider_catalog = Arc::new(InMemoryProviderCatalogReadRepository::seed(
+        vec![sample_provider()],
+        vec![sample_endpoint()],
+        vec![sample_key()],
+    ));
+    let mut usage = sample_usage(
+        "request-connect",
+        "provider-1",
+        "Windsurf",
+        0,
+        0.0,
+        "failed",
+        Some(429),
+        100,
+    );
+    usage.candidate_id = Some("cand-used".to_string());
+    usage.response_headers = Some(json!({
+        "content-type": "application/connect+json"
+    }));
+    let mut framed = Vec::new();
+    framed.push(2);
+    let payload = br#"{"error":{"code":"resource_exhausted","message":"quota exhausted"}}"#;
+    framed.extend_from_slice(&(payload.len() as u32).to_be_bytes());
+    framed.extend_from_slice(payload);
+    usage.response_body = Some(json!(BASE64_STANDARD.encode(framed)));
+    usage.response_body_ref = Some("usage://request/request-connect/response_body".to_string());
+    usage.response_body_state = Some(UsageBodyCaptureState::Inline);
+    let usage_repository = Arc::new(InMemoryUsageReadRepository::seed(vec![usage]));
+    let data_state =
+        crate::data::GatewayDataState::with_request_candidate_and_usage_repository_for_tests(
+            request_candidates,
+            usage_repository,
+        )
+        .with_provider_catalog_reader(provider_catalog);
+    let state = AppState::new()
+        .expect("state should build")
+        .with_data_state_for_tests(data_state);
+    let context = request_context(
+        http::Method::GET,
+        "/api/admin/monitoring/trace/request-connect",
+    );
+
+    let response = local_monitoring_response(&state, &context)
+        .await
+        .expect("handler should not error")
+        .expect("route should be handled locally");
+
+    assert_eq!(response.status(), http::StatusCode::OK);
+    let body = to_bytes(response.into_body(), usize::MAX)
+        .await
+        .expect("body should read");
+    let payload: serde_json::Value = serde_json::from_slice(&body).expect("json body should parse");
+    let upstream_response = &payload["candidates"][0]["extra_data"]["upstream_response"];
+    assert_eq!(upstream_response["status_code"], json!(429));
+    assert_eq!(
+        upstream_response["body"]["error"]["code"],
+        json!("resource_exhausted")
+    );
+    assert_eq!(
+        upstream_response["body"]["error"]["message"],
+        json!("quota exhausted")
+    );
+    assert_eq!(
+        upstream_response["body_ref"],
+        json!("usage://request/request-connect/response_body")
+    );
+    assert_eq!(upstream_response["body_state"], json!("inline"));
+}
+
 #[tokio::test]
 async fn admin_monitoring_trace_request_exposes_structured_ranking_metadata() {
     let mut candidate = sample_candidate(

apps/aether-gateway/src/handlers/admin/provider/endpoint_keys.rs

@@ -1,4 +1,3 @@
-mod balance;
 mod mutations;
 mod quota;
 mod reads;
@@ -19,10 +18,6 @@ pub(crate) async fn maybe_build_local_admin_endpoints_keys_response(
         return Ok(Some(response));
     }
 
-    if let Some(response) = balance::maybe_handle(state, request_context, request_body).await? {
-        return Ok(Some(response));
-    }
-
     if let Some(response) = mutations::maybe_handle(state, request_context, request_body).await? {
         return Ok(Some(response));
     }

apps/aether-gateway/src/handlers/admin/provider/endpoints_admin/reads.rs

@@ -1,6 +1,6 @@
 use crate::handlers::admin::request::AdminAppState;
 use aether_data_contracts::repository::provider_catalog::{
-    StoredProviderCatalogEndpoint, StoredProviderCatalogProvider,
+    StoredProviderCatalogEndpoint, StoredProviderCatalogKey, StoredProviderCatalogProvider,
 };
 use std::time::{SystemTime, UNIX_EPOCH};
 

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/batch/execution.rs

@@ -25,10 +25,15 @@ use crate::handlers::admin::provider::oauth::runtime::{
 use crate::handlers::admin::provider::oauth::state::{
     admin_provider_oauth_template, exchange_admin_provider_oauth_refresh_token,
 };
+use crate::handlers::admin::provider::shared::support::ADMIN_PROVIDER_OAUTH_DATA_UNAVAILABLE_DETAIL;
 use crate::handlers::admin::request::{AdminAppState, AdminProviderOAuthTemplate};
 use crate::GatewayError;
 use aether_admin::provider::oauth::parse_admin_provider_oauth_kiro_batch_import_entries;
 use aether_contracts::ProxySnapshot;
+use aether_oauth::core::OAuthError;
+use aether_oauth::provider::{
+    ProviderOAuthImportInput, ProviderOAuthService, ProviderOAuthTransportContext,
+};
 use serde_json::{json, Map, Value};
 
 struct AdminProviderOAuthResolvedBatchImport {
@@ -37,6 +42,16 @@ struct AdminProviderOAuthResolvedBatchImport {
     expires_at: Option<u64>,
 }
 
+fn sanitize_windsurf_batch_import_error(error: &OAuthError) -> String {
+    match error {
+        OAuthError::InvalidRequest(_) => "Windsurf 凭据验证失败: 请求参数无效".to_string(),
+        OAuthError::HttpStatus { status_code, .. } => {
+            format!("Windsurf 凭据验证失败: HTTP {status_code}")
+        }
+        _ => "Windsurf 凭据验证失败".to_string(),
+    }
+}
+
 pub(super) fn estimate_admin_provider_oauth_batch_import_total(
     provider_type: &str,
     raw_credentials: &str,
@@ -98,6 +113,61 @@ async fn resolve_admin_provider_oauth_batch_import_tokens(
         .map(str::trim)
         .filter(|value| !value.is_empty());
 
+    if provider_type.eq_ignore_ascii_case("windsurf") {
+        let token_for_import = refresh_token.or(access_token);
+        let ctx = ProviderOAuthTransportContext {
+            provider_id: String::new(),
+            provider_type: provider_type.to_string(),
+            endpoint_id: None,
+            key_id: None,
+            auth_type: Some("oauth".to_string()),
+            decrypted_api_key: None,
+            decrypted_auth_config: None,
+            provider_config: None,
+            endpoint_config: None,
+            key_config: None,
+            network: aether_oauth::network::OAuthNetworkContext::provider_operation(
+                request_proxy.clone(),
+            ),
+        };
+        let executor = crate::oauth::GatewayOAuthHttpExecutor::new(*state);
+        let result = ProviderOAuthService::with_builtin_adapters()
+            .import_credentials(
+                &executor,
+                &ctx,
+                ProviderOAuthImportInput {
+                    provider_type: provider_type.to_string(),
+                    name: entry
+                        .raw_credentials
+                        .as_ref()
+                        .and_then(|raw| raw.get("name"))
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .filter(|value| !value.is_empty())
+                        .map(ToOwned::to_owned),
+                    refresh_token: token_for_import.map(ToOwned::to_owned),
+                    raw_credentials: entry.raw_credentials.clone(),
+                    network: ctx.network.clone(),
+                },
+            )
+            .await
+            .map_err(|error| sanitize_windsurf_batch_import_error(&error))?;
+        let access_token = result.token_set.access_token.trim().to_string();
+        if access_token.is_empty() {
+            return Err("Windsurf 凭据验证返回缺少 apiKey/sessionToken".to_string());
+        }
+        let auth_config = result
+            .auth_config
+            .as_object()
+            .cloned()
+            .ok_or_else(|| "Windsurf 凭据验证返回缺少 auth_config".to_string())?;
+        return Ok(AdminProviderOAuthResolvedBatchImport {
+            access_token,
+            auth_config,
+            expires_at: result.token_set.expires_at_unix_secs,
+        });
+    }
+
     if let Some(refresh_token) = refresh_token {
         let Some(template) = template else {
             if provider_type_supports_access_token_import(provider_type) {
@@ -228,6 +298,28 @@ pub(super) async fn execute_admin_provider_oauth_batch_import(
     };
 
     let template = admin_provider_oauth_template(provider_type);
+    if template.is_none()
+        && !provider_type.eq_ignore_ascii_case("windsurf")
+        && !provider_type_supports_access_token_import(provider_type)
+    {
+        return Ok(AdminProviderOAuthBatchImportOutcome {
+            total: entries.len(),
+            success: 0,
+            failed: entries.len(),
+            results: entries
+                .iter()
+                .enumerate()
+                .map(|(index, _)| {
+                    json!({
+                        "index": index,
+                        "status": "error",
+                        "error": ADMIN_PROVIDER_OAUTH_DATA_UNAVAILABLE_DETAIL,
+                        "replaced": false,
+                    })
+                })
+                .collect(),
+        });
+    }
 
     let endpoint_resolution =
         resolve_provider_oauth_runtime_endpoints(state, &provider, provider_type).await?;
@@ -251,6 +343,25 @@ pub(super) async fn execute_admin_provider_oauth_batch_import(
     let mut failed = 0usize;
 
     for (index, entry) in entries.iter().enumerate() {
+        if let Some(error) = entry.parse_error.as_ref() {
+            failed += 1;
+            results.push(json!({
+                "index": index,
+                "status": "error",
+                "error": error,
+                "replaced": false,
+            }));
+            maybe_report_admin_provider_oauth_batch_import_progress(
+                &mut progress,
+                entries.len(),
+                success,
+                failed,
+                &results,
+            )
+            .await;
+            continue;
+        }
+
         let resolved_import = match resolve_admin_provider_oauth_batch_import_tokens(
             state,
             template,
@@ -418,3 +529,33 @@ pub(super) async fn execute_admin_provider_oauth_batch_import(
         results,
     })
 }
+
+#[cfg(test)]
+mod tests {
+    use super::sanitize_windsurf_batch_import_error;
+    use aether_oauth::core::OAuthError;
+
+    #[test]
+    fn windsurf_batch_import_error_redacts_http_body() {
+        let error = OAuthError::HttpStatus {
+            status_code: 401,
+            body_excerpt: "sessionToken=devin-session-token$secret".to_string(),
+        };
+
+        let detail = sanitize_windsurf_batch_import_error(&error);
+
+        assert_eq!(detail, "Windsurf 凭据验证失败: HTTP 401");
+        assert!(!detail.contains("devin-session-token$secret"));
+    }
+
+    #[test]
+    fn windsurf_batch_import_error_redacts_provider_detail() {
+        let error = OAuthError::invalid_response("apiKey=sk-secret token=secret-token");
+
+        let detail = sanitize_windsurf_batch_import_error(&error);
+
+        assert_eq!(detail, "Windsurf 凭据验证失败");
+        assert!(!detail.contains("sk-secret"));
+        assert!(!detail.contains("secret-token"));
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/batch/orchestration.rs

@@ -8,7 +8,7 @@ use super::parse::{
 };
 use crate::handlers::admin::provider::oauth::errors::build_internal_control_error_response;
 use crate::handlers::admin::provider::oauth::state::{
-    build_admin_provider_oauth_backend_unavailable_response,
+    admin_provider_oauth_template, build_admin_provider_oauth_backend_unavailable_response,
     is_fixed_provider_type_for_provider_oauth,
 };
 use crate::handlers::admin::provider::shared::paths::admin_provider_oauth_batch_import_provider_id;
@@ -60,6 +60,13 @@ pub(in super::super) async fn handle_admin_provider_oauth_batch_import(
             "该 Provider 不是固定类型，无法使用 provider-oauth",
         ));
     }
+    if provider_type != "kiro"
+        && provider_type != "windsurf"
+        && admin_provider_oauth_template(&provider_type).is_none()
+    {
+        return Ok(build_admin_provider_oauth_backend_unavailable_response());
+    }
+
     let total = estimate_admin_provider_oauth_batch_import_total(
         &provider_type,
         payload.credentials.as_str(),

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/batch/parse.rs

@@ -19,8 +19,10 @@ pub(super) struct AdminProviderOAuthBatchImportRequest {
 
 #[derive(Debug, Clone)]
 pub(super) struct AdminProviderOAuthBatchImportEntry {
+    pub parse_error: Option<String>,
     pub refresh_token: Option<String>,
     pub access_token: Option<String>,
+    pub raw_credentials: Option<serde_json::Value>,
     pub expires_at: Option<u64>,
     pub account_id: Option<String>,
     pub account_user_id: Option<String>,
@@ -141,8 +143,10 @@ fn extract_admin_provider_oauth_batch_import_entry(
                     access_token.as_deref(),
                 );
                 Some(AdminProviderOAuthBatchImportEntry {
+                    parse_error: None,
                     refresh_token,
                     access_token,
+                    raw_credentials: None,
                     expires_at: None,
                     account_id: None,
                     account_user_id: None,
@@ -160,6 +164,8 @@ fn extract_admin_provider_oauth_batch_import_entry(
             }
         }
         serde_json::Value::Object(object) => {
+            let is_grok = provider_type.trim().eq_ignore_ascii_case("grok");
+            let is_windsurf = provider_type.trim().eq_ignore_ascii_case("windsurf");
             let refresh_token = coerce_admin_provider_oauth_import_str(
                 object
                     .get("refresh_token")
@@ -170,12 +176,8 @@ fn extract_admin_provider_oauth_batch_import_entry(
                     .get("access_token")
                     .or_else(|| object.get("accessToken")),
             );
-            let grok_token_alias = if provider_type.trim().eq_ignore_ascii_case("grok") {
-                object.get("token")
-            } else {
-                None
-            };
-            let grok_cookie = if provider_type.trim().eq_ignore_ascii_case("grok") {
+            let grok_token_alias = if is_grok { object.get("token") } else { None };
+            let grok_cookie = if is_grok {
                 coerce_admin_provider_oauth_import_str(
                     object.get("cookie").or_else(|| object.get("cookieHeader")),
                 )
@@ -198,9 +200,43 @@ fn extract_admin_provider_oauth_batch_import_entry(
                 refresh_token.as_deref(),
                 access_token.as_deref().or(session_token.as_deref()),
             );
-            if refresh_token.is_none() && access_token.is_none() {
+            let windsurf_api_key = is_windsurf
+                .then(|| {
+                    coerce_admin_provider_oauth_import_str(
+                        object.get("api_key").or_else(|| object.get("apiKey")),
+                    )
+                })
+                .flatten();
+            let windsurf_token = is_windsurf
+                .then(|| {
+                    coerce_admin_provider_oauth_import_str(
+                        object
+                            .get("token")
+                            .or_else(|| object.get("auth_token"))
+                            .or_else(|| object.get("authToken")),
+                    )
+                })
+                .flatten();
+            let windsurf_password = is_windsurf
+                .then(|| coerce_admin_provider_oauth_import_str(object.get("password")))
+                .flatten();
+            let raw_credentials = if is_windsurf
+                && (windsurf_api_key.is_some()
+                    || windsurf_token.is_some()
+                    || windsurf_password.is_some())
+            {
+                Some(item.clone())
+            } else {
+                None
+            };
+            if refresh_token.is_none() && access_token.is_none() && raw_credentials.is_none() {
                 return None;
             }
+            let refresh_token = if is_windsurf {
+                refresh_token.or(windsurf_api_key).or(windsurf_token)
+            } else {
+                refresh_token
+            };
             let expires_at =
                 json_u64_value(object.get("expires_at").or_else(|| object.get("expiresAt")));
             let account_id = coerce_admin_provider_oauth_import_str(
@@ -285,8 +321,10 @@ fn extract_admin_provider_oauth_batch_import_entry(
                     .or_else(|| object.get("impersonate")),
             );
             Some(AdminProviderOAuthBatchImportEntry {
+                parse_error: None,
                 refresh_token,
                 access_token,
+                raw_credentials,
                 expires_at,
                 account_id,
                 account_user_id,
@@ -316,14 +354,17 @@ pub(super) fn parse_admin_provider_oauth_batch_import_entries(
     }
 
     if raw.starts_with('[') {
-        if let Ok(serde_json::Value::Array(items)) = serde_json::from_str::<serde_json::Value>(raw)
-        {
-            return items
-                .iter()
-                .filter_map(|item| {
-                    extract_admin_provider_oauth_batch_import_entry(provider_type, item)
-                })
-                .collect();
+        match serde_json::from_str::<serde_json::Value>(raw) {
+            Ok(serde_json::Value::Array(items)) => {
+                return items
+                    .iter()
+                    .filter_map(|item| {
+                        extract_admin_provider_oauth_batch_import_entry(provider_type, item)
+                    })
+                    .collect();
+            }
+            Ok(_) => {}
+            Err(error) => return vec![parse_error_entry(format!("JSON 数组解析失败: {error}"))],
         }
     }
 
@@ -340,23 +381,61 @@ pub(super) fn parse_admin_provider_oauth_batch_import_entries(
     raw.lines()
         .map(str::trim)
         .filter(|line| !line.is_empty() && !line.starts_with('#'))
-        .filter_map(|line| {
-            if line.starts_with('{') {
-                return serde_json::from_str::<serde_json::Value>(line)
-                    .ok()
-                    .and_then(|value| {
-                        extract_admin_provider_oauth_batch_import_entry(provider_type, &value)
-                    });
+        .filter_map(|token| {
+            if is_json_like_batch_line(token) {
+                match serde_json::from_str::<serde_json::Value>(token) {
+                    Ok(value @ serde_json::Value::Object(_)) => {
+                        return extract_admin_provider_oauth_batch_import_entry(
+                            provider_type,
+                            &value,
+                        );
+                    }
+                    Ok(_) => {
+                        return Some(parse_error_entry(
+                            "JSON 行必须是账号对象，不能作为 raw token 导入".to_string(),
+                        ));
+                    }
+                    Err(error) => {
+                        return Some(parse_error_entry(format!("JSON 行解析失败: {error}")));
+                    }
+                }
             }
 
             extract_admin_provider_oauth_batch_import_entry(
                 provider_type,
-                &serde_json::Value::String(line.to_string()),
+                &serde_json::Value::String(token.to_string()),
             )
         })
         .collect()
 }
 
+fn parse_error_entry(error: String) -> AdminProviderOAuthBatchImportEntry {
+    AdminProviderOAuthBatchImportEntry {
+        parse_error: Some(error),
+        refresh_token: None,
+        access_token: None,
+        raw_credentials: None,
+        expires_at: None,
+        account_id: None,
+        account_user_id: None,
+        plan_type: None,
+        pool_tier: None,
+        user_id: None,
+        email: None,
+        account_name: None,
+        sso_rw_token: None,
+        cf_cookies: None,
+        cf_clearance: None,
+        user_agent: None,
+        browser_profile: None,
+    }
+}
+
+fn is_json_like_batch_line(line: &str) -> bool {
+    let line = line.trim_start();
+    line.starts_with('{') || line.starts_with('[')
+}
+
 pub(super) fn apply_admin_provider_oauth_batch_import_hints(
     provider_type: &str,
     entry: &AdminProviderOAuthBatchImportEntry,
@@ -633,4 +712,115 @@ mod tests {
         assert_eq!(entries[0].user_id.as_deref(), Some("user-1"));
         assert_eq!(entries[0].pool_tier.as_deref(), Some("heavy"));
     }
+
+    #[test]
+    fn parses_windsurf_json_credentials_for_native_import() {
+        let entries = parse_admin_provider_oauth_batch_import_entries(
+            "windsurf",
+            r#"[
+                {"api_key":"devin-session-token$abc","email":"a@example.com"},
+                {"token":"firebase-id-token","name":"Browser Login"},
+                {"email":"b@example.com","password":"secret"},
+                {"access_token":"devin-session-token$alias","email":"c@example.com"}
+            ]"#,
+        );
+
+        assert_eq!(entries.len(), 4);
+        assert_eq!(
+            entries[0].refresh_token.as_deref(),
+            Some("devin-session-token$abc")
+        );
+        assert_eq!(entries[0].email.as_deref(), Some("a@example.com"));
+        assert_eq!(
+            entries[0]
+                .raw_credentials
+                .as_ref()
+                .and_then(|value| value.get("api_key")),
+            Some(&json!("devin-session-token$abc"))
+        );
+        assert_eq!(
+            entries[1]
+                .raw_credentials
+                .as_ref()
+                .and_then(|value| value.get("token")),
+            Some(&json!("firebase-id-token"))
+        );
+        assert_eq!(
+            entries[2]
+                .raw_credentials
+                .as_ref()
+                .and_then(|value| value.get("password")),
+            Some(&json!("secret"))
+        );
+        assert_eq!(
+            entries[3].access_token.as_deref(),
+            Some("devin-session-token$alias")
+        );
+    }
+
+    #[test]
+    fn parses_windsurf_json_lines_credentials_for_native_import() {
+        let entries = parse_admin_provider_oauth_batch_import_entries(
+            "windsurf",
+            r#"{"api_key":"devin-session-token$abc","email":"a@example.com"}
+{"token":"firebase-id-token","name":"Browser Login"}
+{"email":"b@example.com","password":"secret"}"#,
+        );
+
+        assert_eq!(entries.len(), 3);
+        assert_eq!(
+            entries[0]
+                .raw_credentials
+                .as_ref()
+                .and_then(|value| value.get("api_key")),
+            Some(&json!("devin-session-token$abc"))
+        );
+        assert_eq!(
+            entries[1]
+                .raw_credentials
+                .as_ref()
+                .and_then(|value| value.get("token")),
+            Some(&json!("firebase-id-token"))
+        );
+        assert_eq!(
+            entries[2]
+                .raw_credentials
+                .as_ref()
+                .and_then(|value| value.get("password")),
+            Some(&json!("secret"))
+        );
+    }
+
+    #[test]
+    fn invalid_json_line_is_parse_error_not_token() {
+        let entries = parse_admin_provider_oauth_batch_import_entries(
+            "windsurf",
+            r#"{"email":"b@example.com","password":"secret""#,
+        );
+
+        assert_eq!(entries.len(), 1);
+        assert!(entries[0].parse_error.is_some());
+        assert!(entries[0].refresh_token.is_none());
+        assert!(entries[0].access_token.is_none());
+        assert!(entries[0].raw_credentials.is_none());
+    }
+
+    #[test]
+    fn json_like_line_after_token_is_parse_error_not_token() {
+        let entries = parse_admin_provider_oauth_batch_import_entries(
+            "windsurf",
+            "devin-session-token$abc\n[not-json",
+        );
+
+        assert_eq!(entries.len(), 2);
+        assert!(entries[0].parse_error.is_none());
+        assert_eq!(
+            entries[0].refresh_token.as_deref(),
+            Some("devin-session-token$abc")
+        );
+        assert!(entries[1].parse_error.is_some());
+        assert!(entries[1].refresh_token.is_none());
+        assert!(entries[1].access_token.is_none());
+        assert!(entries[1].raw_credentials.is_none());
+    }
 }

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/batch/task.rs

@@ -10,7 +10,7 @@ use super::progress::{
 };
 use crate::handlers::admin::provider::oauth::errors::build_internal_control_error_response;
 use crate::handlers::admin::provider::oauth::state::{
-    build_admin_provider_oauth_backend_unavailable_response,
+    admin_provider_oauth_template, build_admin_provider_oauth_backend_unavailable_response,
     is_fixed_provider_type_for_provider_oauth,
 };
 use crate::handlers::admin::provider::shared::paths::admin_provider_oauth_batch_import_task_provider_id;
@@ -124,6 +124,13 @@ pub(in super::super) async fn handle_admin_provider_oauth_start_batch_import_tas
             "该 Provider 不是固定类型，无法使用 provider-oauth",
         ));
     }
+    if provider_type != "kiro"
+        && provider_type != "windsurf"
+        && admin_provider_oauth_template(&provider_type).is_none()
+    {
+        return Ok(build_admin_provider_oauth_backend_unavailable_response());
+    }
+
     let total = estimate_admin_provider_oauth_batch_import_total(
         &provider_type,
         payload.credentials.as_str(),

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/device/authorize.rs

@@ -12,6 +12,7 @@ use crate::GatewayError;
 use aether_data::repository::provider_oauth::{
     StoredAdminProviderOAuthDeviceSession, KIRO_DEVICE_AUTH_SESSION_TTL_BUFFER_SECS,
 };
+use aether_oauth::provider::{ProviderOAuthService, ProviderOAuthTransportContext};
 use axum::{
     body::{Body, Bytes},
     http,
@@ -28,6 +29,8 @@ const KIRO_SOCIAL_MANUAL_CALLBACK_PORT: u16 = 49153;
 const KIRO_SOCIAL_ALLOWED_CALLBACK_PORTS: &[u16] = &[
     3128, 4649, 6588, 8008, 9091, 49153, 50153, 51153, 52153, 53153,
 ];
+const WINDSURF_BROWSER_AUTH_EXPIRES_IN_SECS: u64 = 600;
+const WINDSURF_BROWSER_AUTH_POLL_INTERVAL_SECS: u64 = 5;
 
 fn normalize_kiro_device_auth_type(raw: Option<&str>) -> String {
     match raw
@@ -119,6 +122,26 @@ fn build_kiro_social_authorization_url(
     )
 }
 
+fn build_windsurf_authorization_url(authorize_url: &str, login_option: &str) -> String {
+    let login_option = login_option.trim();
+    if login_option.is_empty() {
+        return authorize_url.to_string();
+    }
+    if let Ok(mut url) = Url::parse(authorize_url) {
+        url.query_pairs_mut()
+            .append_pair("login_option", login_option);
+        return url.to_string();
+    }
+    let separator = if authorize_url.contains('?') {
+        '&'
+    } else {
+        '?'
+    };
+    let mut serializer = form_urlencoded::Serializer::new(String::new());
+    serializer.append_pair("login_option", login_option);
+    format!("{authorize_url}{separator}{}", serializer.finish())
+}
+
 pub(super) async fn handle_admin_provider_oauth_device_authorize(
     state: &AdminAppState<'_>,
     request_context: &AdminRequestContext<'_>,
@@ -163,14 +186,14 @@ pub(super) async fn handle_admin_provider_oauth_device_authorize(
         ));
     };
     let provider_type = provider.provider_type.trim().to_ascii_lowercase();
-    if provider_type != "kiro" {
+    if provider_type != "kiro" && provider_type != "windsurf" {
         return Ok(build_internal_control_error_response(
             http::StatusCode::BAD_REQUEST,
-            "设备授权仅支持 Kiro provider",
+            "设备授权仅支持 Kiro / Windsurf provider",
         ));
     }
     let endpoint_resolution =
-        resolve_provider_oauth_runtime_endpoints(state, &provider, "kiro").await?;
+        resolve_provider_oauth_runtime_endpoints(state, &provider, &provider_type).await?;
     let runtime_endpoint = endpoint_resolution.runtime_endpoint;
     let request_proxy = state
         .resolve_admin_provider_oauth_operation_proxy_snapshot(
@@ -184,6 +207,103 @@ pub(super) async fn handle_admin_provider_oauth_device_authorize(
         )
         .await;
 
+    if provider_type == "windsurf" {
+        let session_id = generate_provider_oauth_nonce();
+        let login_option = payload
+            .login_option
+            .as_deref()
+            .or(payload.auth_type.as_deref())
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+            .unwrap_or("default")
+            .to_ascii_lowercase();
+        let ctx = ProviderOAuthTransportContext {
+            provider_id: provider_id.clone(),
+            provider_type: provider_type.clone(),
+            endpoint_id: runtime_endpoint
+                .as_ref()
+                .map(|endpoint| endpoint.id.clone()),
+            key_id: None,
+            auth_type: Some("oauth".to_string()),
+            decrypted_api_key: None,
+            decrypted_auth_config: None,
+            provider_config: provider.config.clone(),
+            endpoint_config: runtime_endpoint
+                .as_ref()
+                .and_then(|endpoint| endpoint.config.clone()),
+            key_config: None,
+            network: aether_oauth::network::OAuthNetworkContext::provider_operation(
+                request_proxy.clone(),
+            ),
+        };
+        let mut authorization = match ProviderOAuthService::with_builtin_adapters()
+            .build_authorize_url(&ctx, &session_id, None)
+        {
+            Ok(authorization) => authorization,
+            Err(error) => {
+                return Ok(build_internal_control_error_response(
+                    http::StatusCode::BAD_REQUEST,
+                    format!("Windsurf 授权 URL 构建失败: {error}"),
+                ));
+            }
+        };
+        authorization.authorize_url =
+            build_windsurf_authorization_url(&authorization.authorize_url, &login_option);
+        let now_unix_secs = current_unix_secs();
+        let session = StoredAdminProviderOAuthDeviceSession {
+            provider_id: provider_id.clone(),
+            region: String::new(),
+            client_id: String::new(),
+            client_secret: String::new(),
+            device_code: String::new(),
+            auth_type: Some("browser".to_string()),
+            social_provider: Some(login_option.clone()),
+            code_verifier: None,
+            redirect_uri: Some("show-auth-token".to_string()),
+            machine_id: Some(uuid::Uuid::new_v4().to_string().to_ascii_lowercase()),
+            interval: WINDSURF_BROWSER_AUTH_POLL_INTERVAL_SECS,
+            expires_at_unix_secs: now_unix_secs
+                .saturating_add(WINDSURF_BROWSER_AUTH_EXPIRES_IN_SECS),
+            status: "pending".to_string(),
+            proxy_node_id: payload
+                .proxy_node_id
+                .as_deref()
+                .map(str::trim)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned),
+            created_at_unix_ms: now_unix_secs,
+            key_id: None,
+            email: None,
+            replaced: false,
+            error_msg: None,
+        };
+        if let Err(response) = state
+            .save_provider_oauth_device_session(
+                &session_id,
+                &session,
+                WINDSURF_BROWSER_AUTH_EXPIRES_IN_SECS
+                    .saturating_add(KIRO_DEVICE_AUTH_SESSION_TTL_BUFFER_SECS),
+            )
+            .await
+        {
+            return Ok(response);
+        }
+
+        return Ok(Json(json!({
+            "session_id": session_id,
+            "user_code": "",
+            "verification_uri": "https://windsurf.com/windsurf/signin",
+            "verification_uri_complete": authorization.authorize_url,
+            "expires_in": WINDSURF_BROWSER_AUTH_EXPIRES_IN_SECS,
+            "interval": WINDSURF_BROWSER_AUTH_POLL_INTERVAL_SECS,
+            "auth_type": "browser",
+            "login_option": login_option,
+            "redirect_uri": "show-auth-token",
+            "callback_required": true,
+        }))
+        .into_response());
+    }
+
     let auth_type = normalize_kiro_device_auth_type(payload.auth_type.as_deref());
     if let Some(social_provider) = kiro_social_provider_id(&auth_type) {
         let redirect_uri = match normalize_kiro_social_redirect_uri(payload.redirect_uri.as_deref())
@@ -394,3 +514,27 @@ pub(super) async fn handle_admin_provider_oauth_device_authorize(
     }))
     .into_response())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::build_windsurf_authorization_url;
+
+    #[test]
+    fn windsurf_authorization_url_includes_login_option() {
+        let url = build_windsurf_authorization_url(
+            "https://windsurf.com/windsurf/signin?state=session-1",
+            "github",
+        );
+
+        let parsed = url::Url::parse(&url).expect("url should parse");
+        let params = parsed
+            .query_pairs()
+            .map(|(key, value)| (key.to_string(), value.to_string()))
+            .collect::<std::collections::BTreeMap<_, _>>();
+        assert_eq!(params.get("state").map(String::as_str), Some("session-1"));
+        assert_eq!(
+            params.get("login_option").map(String::as_str),
+            Some("github")
+        );
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/device/poll.rs

@@ -26,6 +26,9 @@ use aether_data::repository::provider_oauth::StoredAdminProviderOAuthDeviceSessi
 use aether_data_contracts::repository::provider_catalog::{
     StoredProviderCatalogEndpoint, StoredProviderCatalogProvider,
 };
+use aether_oauth::provider::{
+    ProviderOAuthImportInput, ProviderOAuthService, ProviderOAuthTransportContext,
+};
 use axum::{
     body::{Body, Bytes},
     http,
@@ -86,6 +89,99 @@ fn kiro_social_poll_error_response(error: impl Into<String>) -> Response<Body> {
     .into_response()
 }
 
+fn windsurf_browser_poll_error_response(error: impl Into<String>) -> Response<Body> {
+    Json(json!({
+        "status": "error",
+        "error": error.into(),
+        "replaced": false,
+    }))
+    .into_response()
+}
+
+fn sanitize_windsurf_browser_poll_detail(detail: impl AsRef<str>) -> String {
+    let detail = detail.as_ref().trim();
+    if detail.is_empty() {
+        return "-".to_string();
+    }
+    if contains_windsurf_sensitive_marker(detail) {
+        "[REDACTED upstream error body]".to_string()
+    } else {
+        detail.chars().take(500).collect()
+    }
+}
+
+fn sanitize_windsurf_browser_poll_callback_error(error: &str, description: &str) -> String {
+    let error = sanitize_windsurf_browser_poll_error_code(error);
+    let description = sanitize_windsurf_browser_poll_detail(description);
+    format!("{error}: {description}")
+}
+
+fn sanitize_windsurf_browser_poll_error_code(error: &str) -> String {
+    let error = error.trim();
+    if !error.is_empty()
+        && error.len() <= 80
+        && error
+            .chars()
+            .all(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '_' | '-' | '.'))
+    {
+        return error.to_string();
+    }
+    sanitize_windsurf_browser_poll_detail(error)
+}
+
+fn sanitize_windsurf_browser_poll_oauth_error(error: &aether_oauth::core::OAuthError) -> String {
+    match error {
+        aether_oauth::core::OAuthError::InvalidRequest(_) => {
+            "Windsurf token 验证失败: 请求参数无效".to_string()
+        }
+        aether_oauth::core::OAuthError::HttpStatus { status_code, .. } => {
+            format!("Windsurf token 验证失败: HTTP {status_code}")
+        }
+        _ => "Windsurf token 验证失败".to_string(),
+    }
+}
+
+fn contains_windsurf_sensitive_marker(value: &str) -> bool {
+    let lowered = value.to_ascii_lowercase();
+    [
+        "token",
+        "api_key",
+        "apikey",
+        "sessiontoken",
+        "firebase_id_token",
+        "idtoken",
+        "authorization",
+        "password",
+        "secret",
+        "devin-session-token$",
+    ]
+    .iter()
+    .any(|marker| lowered.contains(marker))
+        || value.contains("sk-")
+}
+
+fn secret_fingerprint(value: &str) -> Option<String> {
+    let value = value.trim();
+    if value.is_empty() {
+        return None;
+    }
+
+    use sha2::{Digest, Sha256};
+    let digest = Sha256::digest(value.as_bytes());
+    Some(
+        digest[..8]
+            .iter()
+            .map(|byte| format!("{byte:02x}"))
+            .collect::<String>(),
+    )
+}
+
+fn insert_secret_fingerprint(target: &mut serde_json::Map<String, Value>, key: &str, secret: &str) {
+    if let Some(fingerprint) = secret_fingerprint(secret) {
+        target.insert(key.to_string(), json!(fingerprint));
+    }
+}
+
 fn kiro_social_provider_from_login_option(login_option: Option<&str>) -> Option<&'static str> {
     match login_option
         .map(str::trim)
@@ -322,8 +418,9 @@ pub(super) async fn handle_admin_provider_oauth_device_poll(
             "Provider 不存在",
         ));
     };
+    let provider_type = provider.provider_type.trim().to_ascii_lowercase();
     let endpoint_resolution =
-        resolve_provider_oauth_runtime_endpoints(state, &provider, "kiro").await?;
+        resolve_provider_oauth_runtime_endpoints(state, &provider, &provider_type).await?;
     let endpoints = endpoint_resolution.endpoints;
     let runtime_endpoint = endpoint_resolution.runtime_endpoint;
     let request_proxy = state
@@ -338,6 +435,20 @@ pub(super) async fn handle_admin_provider_oauth_device_poll(
         )
         .await;
 
+    if provider_type == "windsurf" {
+        return handle_admin_provider_oauth_windsurf_browser_device_poll(
+            state,
+            &provider,
+            &endpoints,
+            request_proxy,
+            session_id,
+            session,
+            payload.callback_url.as_deref(),
+            payload.token.as_deref(),
+        )
+        .await;
+    }
+
     if kiro_device_session_is_social(&session) {
         return handle_admin_provider_oauth_kiro_social_device_poll(
             state,
@@ -630,6 +741,276 @@ pub(super) async fn handle_admin_provider_oauth_device_poll(
     ))
 }
 
+fn windsurf_raw_api_key(value: &str) -> Option<&str> {
+    let value = value.trim();
+    if value.starts_with("devin-session-token$") || value.starts_with("sk-") {
+        Some(value)
+    } else {
+        None
+    }
+}
+
+async fn handle_admin_provider_oauth_windsurf_browser_device_poll(
+    state: &AdminAppState<'_>,
+    provider: &StoredProviderCatalogProvider,
+    endpoints: &[StoredProviderCatalogEndpoint],
+    request_proxy: Option<ProxySnapshot>,
+    session_id: &str,
+    mut session: StoredAdminProviderOAuthDeviceSession,
+    callback_url: Option<&str>,
+    token: Option<&str>,
+) -> Result<Response<Body>, GatewayError> {
+    let callback_url = callback_url
+        .map(str::trim)
+        .filter(|value| !value.is_empty());
+    let token = token.map(str::trim).filter(|value| !value.is_empty());
+    if callback_url.is_none() && token.is_none() {
+        return Ok(Json(json!({"status": "pending", "replaced": false})).into_response());
+    }
+
+    let mut social_provider = session
+        .social_provider
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned);
+    let imported_token = if let Some(callback_url) = callback_url {
+        let callback_params = parse_provider_oauth_callback_params(callback_url);
+        if let Some(error) = callback_params.get("error").map(String::as_str) {
+            let error_description = callback_params
+                .get("error_description")
+                .map(String::as_str)
+                .unwrap_or("用户拒绝授权");
+            let sanitized_error =
+                sanitize_windsurf_browser_poll_callback_error(error, error_description);
+            session.status = "error".to_string();
+            session.error_msg = Some(sanitized_error.clone());
+            let _ = state
+                .save_provider_oauth_device_session(session_id, &session, 30)
+                .await;
+            return Ok(attach_admin_provider_oauth_device_poll_terminal_response(
+                session_id,
+                "error",
+                windsurf_browser_poll_error_response(sanitized_error),
+            ));
+        }
+        let Some(callback_state) = callback_params
+            .get("state")
+            .map(String::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+        else {
+            return Ok(windsurf_browser_poll_error_response("回调 URL 缺少 state"));
+        };
+        if callback_state != session_id {
+            return Ok(windsurf_browser_poll_error_response(
+                "回调 state 与会话不匹配",
+            ));
+        }
+        if let Some(provider) = callback_params
+            .get("provider")
+            .or_else(|| callback_params.get("login_option"))
+            .map(String::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+        {
+            social_provider = Some(provider.to_string());
+        }
+        let Some(callback_token) = callback_params
+            .get("token")
+            .or_else(|| callback_params.get("auth_token"))
+            .or_else(|| callback_params.get("access_token"))
+            .map(String::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+        else {
+            return Ok(windsurf_browser_poll_error_response("回调 URL 缺少 token"));
+        };
+        callback_token.to_string()
+    } else {
+        let token = token.unwrap_or_default();
+        if windsurf_raw_api_key(token).is_none() {
+            return Ok(windsurf_browser_poll_error_response(
+                "浏览器授权请提交包含 state 的回调 URL；纯 token 请使用导入授权",
+            ));
+        }
+        token.to_string()
+    };
+
+    let mut raw_credentials = serde_json::Map::new();
+    if windsurf_raw_api_key(&imported_token).is_some() {
+        raw_credentials.insert("api_key".to_string(), json!(imported_token));
+    } else {
+        raw_credentials.insert("token".to_string(), json!(imported_token));
+    }
+    if let Some(social_provider) = social_provider.as_ref() {
+        raw_credentials.insert("social_provider".to_string(), json!(social_provider));
+    }
+
+    let ctx = ProviderOAuthTransportContext {
+        provider_id: provider.id.clone(),
+        provider_type: provider.provider_type.clone(),
+        endpoint_id: None,
+        key_id: None,
+        auth_type: Some("oauth".to_string()),
+        decrypted_api_key: None,
+        decrypted_auth_config: None,
+        provider_config: provider.config.clone(),
+        endpoint_config: None,
+        key_config: None,
+        network: aether_oauth::network::OAuthNetworkContext::provider_operation(
+            request_proxy.clone(),
+        ),
+    };
+    let executor = crate::oauth::GatewayOAuthHttpExecutor::new(*state);
+    let result = match ProviderOAuthService::with_builtin_adapters()
+        .import_credentials(
+            &executor,
+            &ctx,
+            ProviderOAuthImportInput {
+                provider_type: provider.provider_type.clone(),
+                name: None,
+                refresh_token: None,
+                raw_credentials: Some(Value::Object(raw_credentials)),
+                network: ctx.network.clone(),
+            },
+        )
+        .await
+    {
+        Ok(result) => result,
+        Err(error) => {
+            let sanitized_error = sanitize_windsurf_browser_poll_oauth_error(&error);
+            session.status = "error".to_string();
+            session.error_msg = Some(sanitized_error.clone());
+            let _ = state
+                .save_provider_oauth_device_session(session_id, &session, 30)
+                .await;
+            return Ok(attach_admin_provider_oauth_device_poll_terminal_response(
+                session_id,
+                "error",
+                windsurf_browser_poll_error_response(sanitized_error),
+            ));
+        }
+    };
+    let access_token = result.token_set.access_token.trim().to_string();
+    if access_token.is_empty() {
+        return Ok(windsurf_browser_poll_error_response(
+            "Windsurf token 验证返回缺少 apiKey/sessionToken",
+        ));
+    }
+    let mut auth_config = result.auth_config.as_object().cloned().unwrap_or_default();
+    auth_config.insert("provider_type".to_string(), json!("windsurf"));
+    auth_config.insert("auth_method".to_string(), json!("browser"));
+    if let Some(social_provider) = social_provider.as_ref() {
+        auth_config
+            .entry("social_provider".to_string())
+            .or_insert_with(|| json!(social_provider));
+    }
+
+    let duplicate = match state
+        .find_duplicate_provider_oauth_key(&provider.id, &auth_config, None)
+        .await
+    {
+        Ok(duplicate) => duplicate,
+        Err(detail) => {
+            return Ok(Json(json!({
+                "status": "error",
+                "error": detail,
+                "replaced": false,
+            }))
+            .into_response());
+        }
+    };
+
+    let api_formats = provider_oauth_active_api_formats(endpoints);
+    let key_proxy = provider_oauth_key_proxy_value(session.proxy_node_id.as_deref());
+    let expires_at = result.token_set.expires_at_unix_secs;
+    let email = auth_config
+        .get("email")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned);
+    let mut replaced = false;
+    let persisted_key = if let Some(existing_key) = duplicate {
+        replaced = true;
+        match state
+            .update_existing_provider_oauth_catalog_key(
+                &existing_key,
+                &provider.provider_type,
+                &access_token,
+                &auth_config,
+                &api_formats,
+                key_proxy.clone(),
+                expires_at,
+            )
+            .await?
+        {
+            Some(key) => key,
+            None => {
+                return Ok(build_internal_control_error_response(
+                    http::StatusCode::SERVICE_UNAVAILABLE,
+                    "provider oauth write unavailable",
+                ));
+            }
+        }
+    } else {
+        let key_name = email
+            .as_deref()
+            .map(|email| format!("windsurf_{email}"))
+            .unwrap_or_else(|| format!("windsurf_{}", current_unix_secs()));
+        match state
+            .create_provider_oauth_catalog_key(
+                &provider.id,
+                &provider.provider_type,
+                &key_name,
+                &access_token,
+                &auth_config,
+                &api_formats,
+                key_proxy,
+                expires_at,
+            )
+            .await?
+        {
+            Some(key) => key,
+            None => {
+                return Ok(build_internal_control_error_response(
+                    http::StatusCode::SERVICE_UNAVAILABLE,
+                    "provider oauth write unavailable",
+                ));
+            }
+        }
+    };
+
+    spawn_provider_oauth_account_state_refresh_after_update(
+        state.cloned_app(),
+        provider.clone(),
+        persisted_key.id.clone(),
+        request_proxy.clone(),
+    );
+
+    session.status = "authorized".to_string();
+    session.key_id = Some(persisted_key.id.clone());
+    session.email = email.clone();
+    session.replaced = replaced;
+    session.error_msg = None;
+    let _ = state
+        .save_provider_oauth_device_session(session_id, &session, 60)
+        .await;
+
+    Ok(attach_admin_provider_oauth_device_poll_terminal_response(
+        session_id,
+        "authorized",
+        Json(json!({
+            "status": "authorized",
+            "key_id": persisted_key.id,
+            "email": email,
+            "replaced": replaced,
+        }))
+        .into_response(),
+    ))
+}
+
 async fn handle_admin_provider_oauth_kiro_social_device_poll(
     state: &AdminAppState<'_>,
     provider: &StoredProviderCatalogProvider,
@@ -814,7 +1195,7 @@ async fn handle_admin_provider_oauth_kiro_social_device_poll(
             .get("idToken")
             .or_else(|| token_result.get("id_token")),
     ) {
-        auth_config_object.insert("id_token".to_string(), json!(id_token));
+        insert_secret_fingerprint(&mut auth_config_object, "id_token_fingerprint", &id_token);
     }
     if let Some(token_type) = json_non_empty_string(
         token_result
@@ -921,3 +1302,18 @@ async fn handle_admin_provider_oauth_kiro_social_device_poll(
         .into_response(),
     ))
 }
+
+#[cfg(test)]
+mod tests {
+    #[test]
+    fn windsurf_browser_poll_callback_error_redacts_sensitive_values() {
+        let detail = super::sanitize_windsurf_browser_poll_callback_error(
+            "access_denied",
+            "bad token devin-session-token$secret and apiKey sk-secret",
+        );
+
+        assert_eq!(detail, "access_denied: [REDACTED upstream error body]");
+        assert!(!detail.contains("devin-session-token$secret"));
+        assert!(!detail.contains("sk-secret"));
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/device/session.rs

@@ -12,6 +12,7 @@ pub(super) struct AdminProviderOAuthDeviceAuthorizePayload {
     #[serde(default = "default_kiro_device_region")]
     pub(super) region: String,
     pub(super) auth_type: Option<String>,
+    pub(super) login_option: Option<String>,
     pub(super) redirect_uri: Option<String>,
     pub(super) proxy_node_id: Option<String>,
 }
@@ -20,6 +21,7 @@ pub(super) struct AdminProviderOAuthDeviceAuthorizePayload {
 pub(super) struct AdminProviderOAuthDevicePollPayload {
     pub(super) session_id: String,
     pub(super) callback_url: Option<String>,
+    pub(super) token: Option<String>,
 }
 
 pub(super) fn attach_admin_provider_oauth_device_poll_terminal_response(

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/import.rs

@@ -25,6 +25,10 @@ use crate::handlers::admin::request::{
 };
 use crate::GatewayError;
 use aether_contracts::ProxySnapshot;
+use aether_oauth::core::OAuthError;
+use aether_oauth::provider::{
+    ProviderOAuthImportInput, ProviderOAuthService, ProviderOAuthTransportContext,
+};
 use axum::{
     body::Body,
     http,
@@ -39,6 +43,49 @@ struct AdminProviderOAuthSingleImportTokens {
     expires_at: Option<u64>,
 }
 
+fn sanitize_windsurf_import_error(error: &OAuthError) -> String {
+    match error {
+        OAuthError::InvalidRequest(_) => "Windsurf 凭据验证失败: 请求参数无效".to_string(),
+        OAuthError::HttpStatus { status_code, .. } => {
+            format!("Windsurf 凭据验证失败: HTTP {status_code}")
+        }
+        OAuthError::InvalidResponse(detail) => sanitize_windsurf_invalid_response_detail(detail)
+            .unwrap_or_else(|| "Windsurf 凭据验证失败".to_string()),
+        _ => "Windsurf 凭据验证失败".to_string(),
+    }
+}
+
+fn sanitize_windsurf_invalid_response_detail(detail: &str) -> Option<String> {
+    let detail = detail.trim();
+    if detail.eq_ignore_ascii_case("Auth1 response is not json") {
+        return Some("Windsurf 凭据验证失败: Auth1 响应无法解析".to_string());
+    }
+    if detail.eq_ignore_ascii_case("Auth1 response missing token") {
+        return Some("Windsurf 凭据验证失败: Auth1 响应缺少 token".to_string());
+    }
+    if detail.contains("WindsurfPostAuth response missing sessionToken")
+        || detail.contains("WindsurfPostAuth response is not json")
+        || (detail.contains("WindsurfPostAuth failed") && detail.contains("missing sessionToken"))
+    {
+        return Some("Windsurf 凭据验证失败: PostAuth 未返回 sessionToken".to_string());
+    }
+    if detail.contains("WindsurfPostAuth failed") {
+        return Some("Windsurf 凭据验证失败: PostAuth 失败".to_string());
+    }
+    None
+}
+
+fn import_payload_has_windsurf_credentials(
+    payload: &serde_json::Map<String, serde_json::Value>,
+) -> bool {
+    import_payload_string(payload, "api_key", "apiKey").is_some()
+        || import_payload_string_any(payload, &["token", "auth_token", "authToken"]).is_some()
+        || import_payload_string(payload, "refresh_token", "refreshToken").is_some()
+        || import_payload_string(payload, "access_token", "accessToken").is_some()
+        || (import_payload_string_any(payload, &["email"]).is_some()
+            && import_payload_string_any(payload, &["password"]).is_some())
+}
+
 fn import_payload_string(
     payload: &serde_json::Map<String, serde_json::Value>,
     snake_case: &str,
@@ -266,6 +313,70 @@ async fn resolve_admin_provider_oauth_single_import_tokens(
     })
 }
 
+async fn resolve_admin_provider_oauth_windsurf_single_import_tokens(
+    state: &AdminAppState<'_>,
+    provider_type: &str,
+    name: Option<String>,
+    raw_payload: &serde_json::Map<String, serde_json::Value>,
+    refresh_token: Option<&str>,
+    request_proxy: Option<ProxySnapshot>,
+) -> Result<AdminProviderOAuthSingleImportTokens, Response<Body>> {
+    let ctx = ProviderOAuthTransportContext {
+        provider_id: String::new(),
+        provider_type: provider_type.to_string(),
+        endpoint_id: None,
+        key_id: None,
+        auth_type: Some("oauth".to_string()),
+        decrypted_api_key: None,
+        decrypted_auth_config: None,
+        provider_config: None,
+        endpoint_config: None,
+        key_config: None,
+        network: aether_oauth::network::OAuthNetworkContext::provider_operation(
+            request_proxy.clone(),
+        ),
+    };
+    let executor = crate::oauth::GatewayOAuthHttpExecutor::new(*state);
+    let service = ProviderOAuthService::with_builtin_adapters();
+    let result = service
+        .import_credentials(
+            &executor,
+            &ctx,
+            ProviderOAuthImportInput {
+                provider_type: provider_type.to_string(),
+                name,
+                refresh_token: refresh_token.map(ToOwned::to_owned),
+                raw_credentials: Some(serde_json::Value::Object(raw_payload.clone())),
+                network: ctx.network.clone(),
+            },
+        )
+        .await
+        .map_err(|error| {
+            build_internal_control_error_response(
+                http::StatusCode::BAD_REQUEST,
+                sanitize_windsurf_import_error(&error),
+            )
+        })?;
+    let access_token = result.token_set.access_token.trim().to_string();
+    if access_token.is_empty() {
+        return Err(build_internal_control_error_response(
+            http::StatusCode::BAD_REQUEST,
+            "Windsurf 凭据验证返回缺少 apiKey/sessionToken",
+        ));
+    }
+    let auth_config = result.auth_config.as_object().cloned().ok_or_else(|| {
+        build_internal_control_error_response(
+            http::StatusCode::BAD_REQUEST,
+            "Windsurf 凭据验证返回缺少 auth_config",
+        )
+    })?;
+    Ok(AdminProviderOAuthSingleImportTokens {
+        access_token,
+        auth_config,
+        expires_at: result.token_set.expires_at_unix_secs,
+    })
+}
+
 pub(super) async fn handle_admin_provider_oauth_import_refresh_token(
     state: &AdminAppState<'_>,
     request_context: &AdminRequestContext<'_>,
@@ -370,19 +481,47 @@ pub(super) async fn handle_admin_provider_oauth_import_refresh_token(
         .await;
     let key_proxy = provider_oauth_key_proxy_value(proxy_node_id.as_deref());
 
-    let resolved_import = match resolve_admin_provider_oauth_single_import_tokens(
-        state,
-        template,
-        &provider_type,
-        refresh_token_input.as_deref(),
-        access_token_input.as_deref(),
-        imported_expires_at,
-        request_proxy.clone(),
-    )
-    .await
-    {
-        Ok(value) => value,
-        Err(response) => return Ok(response),
+    let resolved_import = if provider_type == "windsurf" {
+        if !import_payload_has_windsurf_credentials(&raw_payload) {
+            return Ok(build_internal_control_error_response(
+                http::StatusCode::BAD_REQUEST,
+                "Windsurf 凭据不能为空",
+            ));
+        }
+        match resolve_admin_provider_oauth_windsurf_single_import_tokens(
+            state,
+            &provider_type,
+            name.clone(),
+            &raw_payload,
+            refresh_token_input.as_deref(),
+            request_proxy.clone(),
+        )
+        .await
+        {
+            Ok(value) => value,
+            Err(response) => return Ok(response),
+        }
+    } else {
+        if refresh_token_input.is_none() && access_token_input.is_none() {
+            return Ok(build_internal_control_error_response(
+                http::StatusCode::BAD_REQUEST,
+                "Refresh Token 或 Access Token 不能为空",
+            ));
+        }
+        match resolve_admin_provider_oauth_single_import_tokens(
+            state,
+            template,
+            &provider_type,
+            refresh_token_input.as_deref(),
+            access_token_input.as_deref(),
+            imported_expires_at,
+            request_proxy.clone(),
+        )
+        .await
+        {
+            Ok(value) => value,
+            Err(response) => return Ok(response),
+        }
     };
     let AdminProviderOAuthSingleImportTokens {
         access_token,
@@ -480,3 +619,47 @@ pub(super) async fn handle_admin_provider_oauth_import_refresh_token(
     }))
     .into_response())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::sanitize_windsurf_import_error;
+    use aether_oauth::core::OAuthError;
+
+    #[test]
+    fn windsurf_import_error_redacts_http_body() {
+        let error = OAuthError::HttpStatus {
+            status_code: 400,
+            body_excerpt: "token=secret-token password=secret-password".to_string(),
+        };
+
+        let detail = sanitize_windsurf_import_error(&error);
+
+        assert_eq!(detail, "Windsurf 凭据验证失败: HTTP 400");
+        assert!(!detail.contains("secret-token"));
+        assert!(!detail.contains("secret-password"));
+    }
+
+    #[test]
+    fn windsurf_import_error_redacts_invalid_response_detail() {
+        let error =
+            OAuthError::invalid_response("RegisterUser failed with firebase_id_token=secret-token");
+
+        let detail = sanitize_windsurf_import_error(&error);
+
+        assert_eq!(detail, "Windsurf 凭据验证失败");
+        assert!(!detail.contains("secret-token"));
+        assert!(!detail.contains("firebase_id_token"));
+    }
+
+    #[test]
+    fn windsurf_import_error_keeps_safe_post_auth_stage() {
+        let error = OAuthError::invalid_response("WindsurfPostAuth response missing sessionToken");
+
+        let detail = sanitize_windsurf_import_error(&error);
+
+        assert_eq!(
+            detail,
+            "Windsurf 凭据验证失败: PostAuth 未返回 sessionToken"
+        );
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/oauth/dispatch/start.rs

@@ -63,6 +63,12 @@ pub(super) async fn handle_admin_provider_oauth_start_key(
             "该 Provider 不是固定类型，无法使用 provider-oauth",
         ));
     }
+    if provider_type == "windsurf" {
+        return Ok(build_internal_control_error_response(
+            http::StatusCode::BAD_REQUEST,
+            "Windsurf 请使用浏览器登录或导入凭据。",
+        ));
+    }
     let Some(template) = admin_provider_oauth_template(&provider_type) else {
         return Ok(build_internal_control_error_response(
             http::StatusCode::BAD_REQUEST,
@@ -134,6 +140,12 @@ pub(super) async fn handle_admin_provider_oauth_start_provider(
             "Kiro 不支持 OAuth 授权，请使用导入授权。",
         ));
     }
+    if provider_type == "windsurf" {
+        return Ok(build_internal_control_error_response(
+            http::StatusCode::BAD_REQUEST,
+            "Windsurf 请使用浏览器登录或导入凭据。",
+        ));
+    }
     let Some(template) = admin_provider_oauth_template(&provider_type) else {
         return Ok(build_internal_control_error_response(
             http::StatusCode::BAD_REQUEST,

apps/aether-gateway/src/handlers/admin/provider/oauth/duplicates.rs

@@ -36,6 +36,13 @@ fn is_openai_provider_oauth_provider_type(value: Option<&serde_json::Value>) ->
         })
 }
 
+fn is_windsurf_provider_oauth_provider_type(value: Option<&serde_json::Value>) -> bool {
+    value
+        .and_then(serde_json::Value::as_str)
+        .map(str::trim)
+        .is_some_and(|provider_type| provider_type.eq_ignore_ascii_case("windsurf"))
+}
+
 fn match_codex_provider_oauth_identity(
     new_auth_config: &serde_json::Map<String, serde_json::Value>,
     existing_auth_config: &serde_json::Map<String, serde_json::Value>,
@@ -106,6 +113,41 @@ fn match_codex_provider_oauth_identity(
     None
 }
 
+fn match_windsurf_provider_oauth_identity(
+    new_auth_config: &serde_json::Map<String, serde_json::Value>,
+    existing_auth_config: &serde_json::Map<String, serde_json::Value>,
+) -> Option<bool> {
+    let new_provider_type = new_auth_config.get("provider_type");
+    let existing_provider_type = existing_auth_config.get("provider_type");
+    if !is_windsurf_provider_oauth_provider_type(new_provider_type)
+        && !is_windsurf_provider_oauth_provider_type(existing_provider_type)
+    {
+        return None;
+    }
+
+    let new_account_id = normalize_provider_oauth_identity_value(new_auth_config.get("account_id"));
+    let existing_account_id =
+        normalize_provider_oauth_identity_value(existing_auth_config.get("account_id"));
+    if let (Some(new_account_id), Some(existing_account_id)) =
+        (new_account_id.as_deref(), existing_account_id.as_deref())
+    {
+        return Some(new_account_id == existing_account_id);
+    }
+
+    let new_credential_fingerprint =
+        normalize_provider_oauth_identity_value(new_auth_config.get("credential_fingerprint"));
+    let existing_credential_fingerprint =
+        normalize_provider_oauth_identity_value(existing_auth_config.get("credential_fingerprint"));
+    if let (Some(new_fingerprint), Some(existing_fingerprint)) = (
+        new_credential_fingerprint.as_deref(),
+        existing_credential_fingerprint.as_deref(),
+    ) {
+        return Some(new_fingerprint == existing_fingerprint);
+    }
+
+    None
+}
+
 fn is_codex_cross_plan_group_non_duplicate(
     new_auth_config: &serde_json::Map<String, serde_json::Value>,
     existing_auth_config: &serde_json::Map<String, serde_json::Value>,
@@ -169,10 +211,17 @@ pub(crate) async fn find_duplicate_provider_oauth_key(
 ) -> Result<Option<StoredProviderCatalogKey>, String> {
     let new_email = normalize_provider_oauth_identity_value(auth_config.get("email"));
     let new_user_id = normalize_provider_oauth_identity_value(auth_config.get("user_id"));
+    let new_account_id = normalize_provider_oauth_identity_value(auth_config.get("account_id"));
+    let new_credential_fingerprint =
+        normalize_provider_oauth_identity_value(auth_config.get("credential_fingerprint"));
     let new_auth_method = normalize_provider_oauth_identity_value(auth_config.get("auth_method"));
     let new_kiro_provider = normalize_provider_oauth_identity_value(auth_config.get("provider"));
 
-    if new_email.is_none() && new_user_id.is_none() {
+    if new_email.is_none()
+        && new_user_id.is_none()
+        && new_account_id.is_none()
+        && new_credential_fingerprint.is_none()
+    {
         return Ok(None);
     }
 
@@ -201,15 +250,28 @@ pub(crate) async fn find_duplicate_provider_oauth_key(
             normalize_provider_oauth_identity_value(existing_auth_config.get("auth_method"));
         let existing_kiro_provider =
             normalize_provider_oauth_identity_value(existing_auth_config.get("provider"));
+        let is_windsurf = auth_config
+            .get("provider_type")
+            .and_then(serde_json::Value::as_str)
+            .is_some_and(|value| value.eq_ignore_ascii_case("windsurf"))
+            || existing_auth_config
+                .get("provider_type")
+                .and_then(serde_json::Value::as_str)
+                .is_some_and(|value| value.eq_ignore_ascii_case("windsurf"));
 
         let mut is_duplicate = false;
         let codex_identity_match =
             match_codex_provider_oauth_identity(auth_config, &existing_auth_config);
+        let windsurf_identity_match =
+            match_windsurf_provider_oauth_identity(auth_config, &existing_auth_config);
         if let Some(codex_identity_match) = codex_identity_match {
             is_duplicate = codex_identity_match;
+        } else if let Some(windsurf_identity_match) = windsurf_identity_match {
+            is_duplicate = windsurf_identity_match;
         }
 
         if codex_identity_match.is_none()
+            && windsurf_identity_match.is_none()
             && !is_duplicate
             && new_user_id.is_some()
             && existing_user_id.is_some()
@@ -220,7 +282,9 @@ pub(crate) async fn find_duplicate_provider_oauth_key(
         }
 
         if codex_identity_match.is_none()
+            && windsurf_identity_match.is_none()
             && !is_duplicate
+            && !is_windsurf
             && new_email.is_some()
             && existing_email.is_some()
             && new_email == existing_email
@@ -261,6 +325,12 @@ pub(crate) async fn find_duplicate_provider_oauth_key(
         let identifier =
             normalize_provider_oauth_identity_value(auth_config.get("account_user_id"))
                 .or_else(|| normalize_provider_oauth_identity_value(auth_config.get("account_id")))
+                .or_else(|| {
+                    normalize_provider_oauth_identity_value(
+                        auth_config.get("credential_fingerprint"),
+                    )
+                    .map(|value| format!("fingerprint:{value}"))
+                })
                 .or_else(|| new_email.clone())
                 .or_else(|| new_user_id.clone())
                 .unwrap_or_default();
@@ -272,3 +342,91 @@ pub(crate) async fn find_duplicate_provider_oauth_key(
 
     Ok(None)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::match_windsurf_provider_oauth_identity;
+    use serde_json::{json, Map, Value};
+
+    fn auth_config(value: Value) -> Map<String, Value> {
+        value.as_object().cloned().expect("auth config object")
+    }
+
+    #[test]
+    fn windsurf_identity_matches_account_id_without_email() {
+        let new_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "auth_method": "api_key",
+            "account_id": "acct-ws-1"
+        }));
+        let existing_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "auth_method": "browser",
+            "account_id": "acct-ws-1"
+        }));
+
+        assert_eq!(
+            match_windsurf_provider_oauth_identity(&new_auth_config, &existing_auth_config),
+            Some(true)
+        );
+    }
+
+    #[test]
+    fn windsurf_identity_rejects_different_account_id() {
+        let new_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "account_id": "acct-ws-1",
+            "email": "same@example.com"
+        }));
+        let existing_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "account_id": "acct-ws-2",
+            "email": "same@example.com"
+        }));
+
+        assert_eq!(
+            match_windsurf_provider_oauth_identity(&new_auth_config, &existing_auth_config),
+            Some(false)
+        );
+    }
+
+    #[test]
+    fn windsurf_identity_matches_credential_fingerprint_without_profile() {
+        let new_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "auth_method": "api_key",
+            "credential_fingerprint": "abcdef0123456789"
+        }));
+        let existing_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "auth_method": "browser",
+            "credential_fingerprint": "abcdef0123456789"
+        }));
+
+        assert_eq!(
+            match_windsurf_provider_oauth_identity(&new_auth_config, &existing_auth_config),
+            Some(true)
+        );
+    }
+
+    #[test]
+    fn windsurf_identity_does_not_match_user_supplied_email_only() {
+        let new_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "auth_method": "api_key",
+            "email": "same@example.com",
+            "email_verified": false
+        }));
+        let existing_auth_config = auth_config(json!({
+            "provider_type": "windsurf",
+            "auth_method": "api_key",
+            "email": "same@example.com",
+            "email_verified": false
+        }));
+
+        assert_eq!(
+            match_windsurf_provider_oauth_identity(&new_auth_config, &existing_auth_config),
+            None
+        );
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/oauth/quota/dispatch.rs

@@ -6,6 +6,7 @@ use super::chatgpt_web::refresh_chatgpt_web_provider_quota_locally;
 use super::codex::refresh_codex_provider_quota_locally;
 use super::grok::refresh_grok_provider_quota_locally;
 use super::kiro::refresh_kiro_provider_quota_locally;
+use super::windsurf::refresh_windsurf_provider_quota_locally;
 use crate::handlers::admin::request::AdminAppState;
 use crate::GatewayError;
 use aether_contracts::ProxySnapshot;
@@ -36,6 +37,7 @@ const PROVIDER_QUOTA_REFRESH_HANDLERS: &[(&str, ProviderQuotaRefreshHandler)] =
     ("codex", refresh_codex_provider_quota_locally_boxed),
     ("grok", refresh_grok_provider_quota_locally_boxed),
     ("kiro", refresh_kiro_provider_quota_locally_boxed),
+    ("windsurf", refresh_windsurf_provider_quota_locally_boxed),
 ];
 
 pub(crate) async fn refresh_provider_pool_quota_locally(
@@ -135,3 +137,19 @@ fn refresh_grok_provider_quota_locally_boxed<'a>(
         proxy_override,
     ))
 }
+
+fn refresh_windsurf_provider_quota_locally_boxed<'a>(
+    state: &'a AdminAppState<'a>,
+    provider: &'a StoredProviderCatalogProvider,
+    endpoint: &'a StoredProviderCatalogEndpoint,
+    keys: Vec<StoredProviderCatalogKey>,
+    proxy_override: Option<ProxySnapshot>,
+) -> ProviderQuotaRefreshFuture<'a> {
+    Box::pin(refresh_windsurf_provider_quota_locally(
+        state,
+        provider,
+        endpoint,
+        keys,
+        proxy_override,
+    ))
+}

apps/aether-gateway/src/handlers/admin/provider/oauth/quota/mod.rs

@@ -5,3 +5,4 @@ pub(crate) mod dispatch;
 pub(crate) mod grok;
 pub(crate) mod kiro;
 pub(crate) mod shared;
+pub(crate) mod windsurf;

apps/aether-gateway/src/handlers/admin/provider/oauth/quota/windsurf.rs

@@ -0,0 +1,632 @@
+use super::shared::{
+    build_provider_quota_execution_plan, build_quota_snapshot_payload,
+    default_provider_quota_execution_timeouts, execute_provider_quota_plan,
+    extract_execution_error_message, persist_provider_quota_refresh_state,
+    quota_refresh_success_invalid_state, ProviderQuotaExecutionOutcome,
+};
+use crate::handlers::admin::request::{AdminAppState, AdminGatewayProviderTransportSnapshot};
+use crate::GatewayError;
+use aether_contracts::ProxySnapshot;
+use aether_data_contracts::repository::provider_catalog::{
+    StoredProviderCatalogEndpoint, StoredProviderCatalogKey, StoredProviderCatalogProvider,
+};
+use aether_provider_pool::{
+    build_windsurf_pool_model_configs_request_with_base_url,
+    build_windsurf_pool_quota_request_with_base_url,
+    build_windsurf_pool_rate_limit_request_with_base_url, ProviderPoolQuotaRequestSpec,
+};
+use serde_json::json;
+use std::time::{SystemTime, UNIX_EPOCH};
+
+async fn execute_windsurf_probe_plan(
+    state: &AdminAppState<'_>,
+    transport: &AdminGatewayProviderTransportSnapshot,
+    spec: ProviderPoolQuotaRequestSpec,
+    proxy_override: Option<&ProxySnapshot>,
+    quota_kind: &str,
+) -> Result<ProviderQuotaExecutionOutcome, GatewayError> {
+    let proxy = match proxy_override {
+        Some(proxy) => Some(proxy.clone()),
+        None => {
+            state
+                .resolve_transport_proxy_snapshot_with_tunnel_affinity(transport)
+                .await
+        }
+    };
+    let timeouts = state
+        .resolve_transport_execution_timeouts(transport)
+        .or(Some(default_provider_quota_execution_timeouts(
+            proxy.as_ref(),
+        )));
+    let plan = build_provider_quota_execution_plan(
+        transport,
+        spec,
+        proxy,
+        state.resolve_transport_profile(transport),
+        timeouts,
+    );
+
+    execute_provider_quota_plan(state, transport, plan, quota_kind).await
+}
+
+async fn execute_windsurf_user_status_plan(
+    state: &AdminAppState<'_>,
+    transport: &AdminGatewayProviderTransportSnapshot,
+    api_key: &str,
+    proxy_override: Option<&ProxySnapshot>,
+) -> Result<ProviderQuotaExecutionOutcome, GatewayError> {
+    let spec = build_windsurf_pool_quota_request_with_base_url(
+        &transport.key.id,
+        &transport.endpoint.base_url,
+        api_key,
+    );
+    execute_windsurf_probe_plan(
+        state,
+        transport,
+        spec,
+        proxy_override,
+        "windsurf:user_status",
+    )
+    .await
+}
+
+async fn execute_windsurf_model_configs_plan(
+    state: &AdminAppState<'_>,
+    transport: &AdminGatewayProviderTransportSnapshot,
+    api_key: &str,
+    proxy_override: Option<&ProxySnapshot>,
+) -> Result<ProviderQuotaExecutionOutcome, GatewayError> {
+    let spec = build_windsurf_pool_model_configs_request_with_base_url(
+        &transport.key.id,
+        &transport.endpoint.base_url,
+        api_key,
+    );
+    execute_windsurf_probe_plan(
+        state,
+        transport,
+        spec,
+        proxy_override,
+        "windsurf:model_configs",
+    )
+    .await
+}
+
+async fn execute_windsurf_rate_limit_plan(
+    state: &AdminAppState<'_>,
+    transport: &AdminGatewayProviderTransportSnapshot,
+    api_key: &str,
+    proxy_override: Option<&ProxySnapshot>,
+) -> Result<ProviderQuotaExecutionOutcome, GatewayError> {
+    let spec = build_windsurf_pool_rate_limit_request_with_base_url(
+        &transport.key.id,
+        &transport.endpoint.base_url,
+        api_key,
+    );
+    execute_windsurf_probe_plan(
+        state,
+        transport,
+        spec,
+        proxy_override,
+        "windsurf:rate_limit",
+    )
+    .await
+}
+
+fn merge_windsurf_probe_metadata(
+    mut user_status_metadata: serde_json::Value,
+    model_configs_metadata: Option<serde_json::Value>,
+    rate_limit_metadata: Option<serde_json::Value>,
+) -> serde_json::Value {
+    let Some(target) = user_status_metadata.as_object_mut() else {
+        return user_status_metadata;
+    };
+    for metadata in [model_configs_metadata, rate_limit_metadata]
+        .into_iter()
+        .flatten()
+    {
+        if let Some(source) = metadata.as_object() {
+            for (key, value) in source {
+                target.insert(key.clone(), value.clone());
+            }
+        }
+    }
+    user_status_metadata
+}
+
+fn append_windsurf_probe_warning(metadata: &mut serde_json::Value, probe: &str, message: String) {
+    let Some(target) = metadata.as_object_mut() else {
+        return;
+    };
+    let warnings = target
+        .entry("probe_warnings".to_string())
+        .or_insert_with(|| serde_json::Value::Array(Vec::new()));
+    if let Some(items) = warnings.as_array_mut() {
+        items.push(json!({
+            "probe": probe,
+            "message": message,
+        }));
+    }
+}
+
+fn build_windsurf_metadata_update(
+    current_upstream_metadata: Option<&serde_json::Value>,
+    patch: serde_json::Value,
+) -> serde_json::Value {
+    let Some(patch_object) = patch.as_object() else {
+        return json!({ "windsurf": patch });
+    };
+    let mut merged_bucket = current_upstream_metadata
+        .and_then(|value| value.get("windsurf"))
+        .and_then(serde_json::Value::as_object)
+        .cloned()
+        .unwrap_or_default();
+    for (key, value) in patch_object {
+        merged_bucket.insert(key.clone(), value.clone());
+    }
+    json!({ "windsurf": merged_bucket })
+}
+
+fn sanitize_windsurf_probe_detail(detail: impl AsRef<str>) -> String {
+    let detail = detail.as_ref().trim();
+    if detail.is_empty() {
+        return "-".to_string();
+    }
+    if let Ok(mut value) = serde_json::from_str::<serde_json::Value>(detail) {
+        redact_windsurf_sensitive_json(&mut value);
+        return value.to_string().chars().take(500).collect();
+    }
+    if contains_windsurf_sensitive_marker(detail) {
+        "[REDACTED upstream error body]".to_string()
+    } else {
+        detail.chars().take(500).collect()
+    }
+}
+
+fn redact_windsurf_sensitive_json(value: &mut serde_json::Value) {
+    match value {
+        serde_json::Value::Object(object) => {
+            for (key, value) in object {
+                if is_windsurf_sensitive_key(key) {
+                    *value = json!("[REDACTED]");
+                } else {
+                    redact_windsurf_sensitive_json(value);
+                }
+            }
+        }
+        serde_json::Value::Array(items) => {
+            for item in items {
+                redact_windsurf_sensitive_json(item);
+            }
+        }
+        serde_json::Value::String(text) if looks_like_windsurf_secret(text) => {
+            *text = "[REDACTED]".to_string();
+        }
+        _ => {}
+    }
+}
+
+fn is_windsurf_sensitive_key(key: &str) -> bool {
+    let normalized = key
+        .chars()
+        .filter(|ch| ch.is_ascii_alphanumeric())
+        .collect::<String>()
+        .to_ascii_lowercase();
+    normalized.contains("token")
+        || normalized.contains("apikey")
+        || normalized.contains("password")
+        || normalized.contains("authorization")
+        || normalized.contains("secret")
+}
+
+fn looks_like_windsurf_secret(value: &str) -> bool {
+    let value = value.trim();
+    value.starts_with("devin-session-token$")
+        || value.starts_with("sk-")
+        || (value.len() > 80 && value.split('.').count() == 3)
+}
+
+fn contains_windsurf_sensitive_marker(value: &str) -> bool {
+    let lowered = value.to_ascii_lowercase();
+    [
+        "token",
+        "api_key",
+        "apikey",
+        "sessiontoken",
+        "firebase_id_token",
+        "idtoken",
+        "authorization",
+        "password",
+        "secret",
+        "devin-session-token$",
+    ]
+    .iter()
+    .any(|marker| lowered.contains(marker))
+        || value.contains("sk-")
+}
+
+pub(crate) async fn refresh_windsurf_provider_quota_locally(
+    state: &AdminAppState<'_>,
+    provider: &StoredProviderCatalogProvider,
+    endpoint: &StoredProviderCatalogEndpoint,
+    keys: Vec<StoredProviderCatalogKey>,
+    proxy_override: Option<ProxySnapshot>,
+) -> Result<Option<serde_json::Value>, GatewayError> {
+    let mut results = Vec::new();
+    let mut success_count = 0usize;
+    let mut failed_count = 0usize;
+
+    for key in keys {
+        let transport = match state
+            .read_provider_transport_snapshot(&provider.id, &endpoint.id, &key.id)
+            .await?
+        {
+            Some(transport) => transport,
+            None => {
+                failed_count += 1;
+                results.push(json!({
+                    "key_id": key.id,
+                    "key_name": key.name,
+                    "status": "error",
+                    "message": "Provider transport snapshot unavailable",
+                }));
+                continue;
+            }
+        };
+
+        let api_key = transport.key.decrypted_api_key.trim();
+        if api_key.is_empty() {
+            failed_count += 1;
+            results.push(json!({
+                "key_id": key.id,
+                "key_name": key.name,
+                "status": "error",
+                "message": "缺少 Windsurf apiKey/sessionToken",
+            }));
+            continue;
+        }
+
+        let result = match execute_windsurf_user_status_plan(
+            state,
+            &transport,
+            api_key,
+            proxy_override.as_ref(),
+        )
+        .await?
+        {
+            ProviderQuotaExecutionOutcome::Response(result) => result,
+            ProviderQuotaExecutionOutcome::Failure(detail) => {
+                failed_count += 1;
+                let detail = sanitize_windsurf_probe_detail(detail);
+                results.push(json!({
+                    "key_id": key.id,
+                    "key_name": key.name,
+                    "status": "error",
+                    "message": format!("GetUserStatus 请求执行失败: {detail}"),
+                    "status_code": 502,
+                }));
+                continue;
+            }
+        };
+
+        let now_unix_secs = SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .ok()
+            .map(|duration| duration.as_secs())
+            .unwrap_or(0);
+        let mut metadata_update = None::<serde_json::Value>;
+        let (mut oauth_invalid_at_unix_secs, mut oauth_invalid_reason) =
+            quota_refresh_success_invalid_state(&key);
+        let mut status = "error".to_string();
+        let mut message = None::<String>;
+
+        if result.status_code == 200 {
+            if let Some(body_json) = result
+                .body
+                .as_ref()
+                .and_then(|body| body.json_body.as_ref())
+            {
+                let mut windsurf_metadata =
+                    aether_admin::provider::quota::parse_windsurf_user_status_response(
+                        body_json,
+                        now_unix_secs,
+                    );
+                if let Some(mut metadata) = windsurf_metadata.take() {
+                    let model_metadata = match execute_windsurf_model_configs_plan(
+                        state,
+                        &transport,
+                        api_key,
+                        proxy_override.as_ref(),
+                    )
+                    .await?
+                    {
+                        ProviderQuotaExecutionOutcome::Response(model_result)
+                            if model_result.status_code == 200 =>
+                        {
+                            model_result
+                                .body
+                                .as_ref()
+                                .and_then(|body| body.json_body.as_ref())
+                                .and_then(|body_json| {
+                                    aether_admin::provider::quota::parse_windsurf_model_configs_response(
+                                        body_json,
+                                        now_unix_secs,
+                                    )
+                                })
+                        }
+                        ProviderQuotaExecutionOutcome::Response(model_result) => {
+                            let detail = extract_execution_error_message(&model_result)
+                                .unwrap_or_else(|| format!("HTTP {}", model_result.status_code));
+                            let detail = sanitize_windsurf_probe_detail(detail);
+                            append_windsurf_probe_warning(
+                                &mut metadata,
+                                "model_configs",
+                                format!("GetCascadeModelConfigs 返回: {detail}"),
+                            );
+                            None
+                        }
+                        ProviderQuotaExecutionOutcome::Failure(detail) => {
+                            let detail = sanitize_windsurf_probe_detail(detail);
+                            append_windsurf_probe_warning(
+                                &mut metadata,
+                                "model_configs",
+                                format!("GetCascadeModelConfigs 执行失败: {detail}"),
+                            );
+                            None
+                        }
+                    };
+                    let rate_limit_metadata = match execute_windsurf_rate_limit_plan(
+                        state,
+                        &transport,
+                        api_key,
+                        proxy_override.as_ref(),
+                    )
+                    .await?
+                    {
+                        ProviderQuotaExecutionOutcome::Response(rate_limit_result)
+                            if rate_limit_result.status_code == 200 =>
+                        {
+                            rate_limit_result
+                                .body
+                                .as_ref()
+                                .and_then(|body| body.json_body.as_ref())
+                                .and_then(|body_json| {
+                                    aether_admin::provider::quota::parse_windsurf_rate_limit_response(
+                                        body_json,
+                                        now_unix_secs,
+                                    )
+                                })
+                        }
+                        ProviderQuotaExecutionOutcome::Response(rate_limit_result) => {
+                            let detail = extract_execution_error_message(&rate_limit_result)
+                                .unwrap_or_else(|| format!("HTTP {}", rate_limit_result.status_code));
+                            let detail = sanitize_windsurf_probe_detail(detail);
+                            append_windsurf_probe_warning(
+                                &mut metadata,
+                                "rate_limit",
+                                format!("CheckUserMessageRateLimit 返回: {detail}"),
+                            );
+                            None
+                        }
+                        ProviderQuotaExecutionOutcome::Failure(detail) => {
+                            let detail = sanitize_windsurf_probe_detail(detail);
+                            append_windsurf_probe_warning(
+                                &mut metadata,
+                                "rate_limit",
+                                format!("CheckUserMessageRateLimit 执行失败: {detail}"),
+                            );
+                            None
+                        }
+                    };
+                    metadata = merge_windsurf_probe_metadata(
+                        metadata,
+                        model_metadata,
+                        rate_limit_metadata,
+                    );
+                    metadata_update = Some(build_windsurf_metadata_update(
+                        key.upstream_metadata.as_ref(),
+                        metadata,
+                    ));
+                    status = "success".to_string();
+                } else {
+                    status = "no_metadata".to_string();
+                    message = Some("响应中未包含 Windsurf 限额信息".to_string());
+                }
+            } else {
+                status = "no_metadata".to_string();
+                message = Some("无法解析 GetUserStatus 响应".to_string());
+            }
+        } else {
+            let err_msg =
+                extract_execution_error_message(&result).map(sanitize_windsurf_probe_detail);
+            message = Some(match err_msg.as_deref() {
+                Some(detail) if !detail.is_empty() => {
+                    format!(
+                        "GetUserStatus 返回状态码 {}: {}",
+                        result.status_code, detail
+                    )
+                }
+                _ => format!("GetUserStatus 返回状态码 {}", result.status_code),
+            });
+            let detail = err_msg
+                .clone()
+                .filter(|value| !value.trim().is_empty())
+                .unwrap_or_else(|| format!("HTTP {}", result.status_code));
+            let mut metadata = serde_json::Map::new();
+            metadata.insert("updated_at".to_string(), json!(now_unix_secs));
+            metadata.insert("last_error".to_string(), json!(detail));
+            match result.status_code {
+                401 | 403 => {
+                    oauth_invalid_at_unix_secs = Some(now_unix_secs);
+                    oauth_invalid_reason =
+                        Some(format!("Windsurf token 无效或已被拒绝: {}", detail));
+                    metadata.insert("banned".to_string(), json!(result.status_code == 403));
+                    status = if result.status_code == 401 {
+                        "auth_invalid".to_string()
+                    } else {
+                        "forbidden".to_string()
+                    };
+                }
+                429 => {
+                    metadata.insert(
+                        "rate_limit".to_string(),
+                        json!({
+                            "limited": true,
+                            "message": metadata
+                                .get("last_error")
+                                .cloned()
+                                .unwrap_or_else(|| json!("rate limited")),
+                        }),
+                    );
+                    status = "rate_limited".to_string();
+                }
+                _ => {}
+            }
+            metadata_update = Some(build_windsurf_metadata_update(
+                key.upstream_metadata.as_ref(),
+                serde_json::Value::Object(metadata),
+            ));
+        }
+
+        if !persist_provider_quota_refresh_state(
+            state,
+            &key.id,
+            metadata_update.as_ref(),
+            oauth_invalid_at_unix_secs,
+            oauth_invalid_reason,
+            None,
+        )
+        .await?
+        {
+            failed_count += 1;
+            results.push(json!({
+                "key_id": key.id,
+                "key_name": key.name,
+                "status": "error",
+                "message": "Key 状态写入失败",
+            }));
+            continue;
+        }
+
+        if status == "success" {
+            success_count += 1;
+        } else {
+            failed_count += 1;
+        }
+
+        let mut payload = serde_json::Map::new();
+        payload.insert("key_id".to_string(), json!(key.id));
+        payload.insert("key_name".to_string(), json!(key.name));
+        payload.insert("status".to_string(), json!(status));
+        if let Some(message) = message {
+            payload.insert("message".to_string(), json!(message));
+        }
+        if result.status_code != 200 {
+            payload.insert("status_code".to_string(), json!(result.status_code));
+        }
+        if let Some(metadata) = metadata_update
+            .as_ref()
+            .and_then(|value| value.get("windsurf"))
+            .cloned()
+        {
+            payload.insert("metadata".to_string(), metadata);
+        }
+        if let Some(quota_snapshot) = build_quota_snapshot_payload(
+            "windsurf",
+            key.status_snapshot.as_ref(),
+            metadata_update.as_ref(),
+        ) {
+            payload.insert("quota_snapshot".to_string(), quota_snapshot);
+        }
+        results.push(serde_json::Value::Object(payload));
+    }
+
+    Ok(Some(json!({
+        "success": success_count,
+        "failed": failed_count,
+        "total": success_count + failed_count,
+        "results": results,
+        "message": format!("已处理 {} 个 Key", success_count + failed_count),
+        "auto_removed": 0,
+    })))
+}
+
+#[cfg(test)]
+mod tests {
+    use serde_json::json;
+
+    #[test]
+    fn windsurf_probe_metadata_merges_user_status_models_and_rate_limit() {
+        let metadata = super::merge_windsurf_probe_metadata(
+            json!({
+                "plan_name": "Pro",
+                "daily_remaining_percent": 42.0,
+                "updated_at": 1_770_000_000u64,
+            }),
+            Some(json!({
+                "allowed_models_count": 2u64,
+                "models": [
+                    {"model_uid": "claude-sonnet-4-5"},
+                    {"model_uid": "gpt-5-mini"}
+                ],
+                "updated_at": 1_770_000_010u64,
+            })),
+            Some(json!({
+                "rate_limit": {
+                    "limited": true,
+                    "messages_remaining": 0.0,
+                    "retry_after_ms": 60_000u64
+                },
+                "updated_at": 1_770_000_020u64,
+            })),
+        );
+
+        assert_eq!(metadata["plan_name"], json!("Pro"));
+        assert_eq!(metadata["daily_remaining_percent"], json!(42.0));
+        assert_eq!(metadata["allowed_models_count"], json!(2u64));
+        assert_eq!(metadata["rate_limit"]["limited"], json!(true));
+        assert_eq!(metadata["updated_at"], json!(1_770_000_020u64));
+    }
+
+    #[test]
+    fn windsurf_probe_detail_redacts_sensitive_values() {
+        let detail = super::sanitize_windsurf_probe_detail(
+            r#"{"error":{"message":"bad"},"apiKey":"sk-secret","sessionToken":"devin-session-token$secret"}"#,
+        );
+
+        assert!(detail.contains("[REDACTED]"));
+        assert!(!detail.contains("sk-secret"));
+        assert!(!detail.contains("devin-session-token$secret"));
+    }
+
+    #[test]
+    fn windsurf_metadata_update_preserves_existing_bucket_fields() {
+        let update = super::build_windsurf_metadata_update(
+            Some(&json!({
+                "windsurf": {
+                    "daily_remaining_percent": 0.0,
+                    "allowed_models_count": 3,
+                    "updated_at": 1u64
+                }
+            })),
+            json!({
+                "last_error": "HTTP 429",
+                "rate_limit": {"limited": true},
+                "updated_at": 2u64
+            }),
+        );
+
+        assert_eq!(
+            update.pointer("/windsurf/daily_remaining_percent"),
+            Some(&json!(0.0))
+        );
+        assert_eq!(
+            update.pointer("/windsurf/allowed_models_count"),
+            Some(&json!(3))
+        );
+        assert_eq!(update.pointer("/windsurf/updated_at"), Some(&json!(2u64)));
+        assert_eq!(
+            update.pointer("/windsurf/rate_limit/limited"),
+            Some(&json!(true))
+        );
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/ops/providers/actions/mod.rs

@@ -32,68 +32,6 @@ pub(super) fn admin_provider_ops_is_valid_action_type(action_type: &str) -> bool
     )
 }
 
-pub(crate) fn admin_provider_ops_saved_connector_credentials(
-    state: &AdminAppState<'_>,
-    provider: &StoredProviderCatalogProvider,
-) -> serde_json::Map<String, serde_json::Value> {
-    admin_provider_ops_decrypted_credentials(
-        state,
-        admin_provider_ops_config_object(provider)
-            .and_then(admin_provider_ops_connector_object)
-            .and_then(|connector| connector.get("credentials")),
-    )
-}
-
-pub(crate) async fn admin_provider_ops_query_balance_response_for_credentials(
-    state: &AdminAppState<'_>,
-    provider_id: &str,
-    provider: &StoredProviderCatalogProvider,
-    architecture_id: &str,
-    base_url: &str,
-    provider_ops_config: &serde_json::Map<String, serde_json::Value>,
-    connector_config: &serde_json::Map<String, serde_json::Value>,
-    credentials: &serde_json::Map<String, serde_json::Value>,
-    request_config: Option<&serde_json::Map<String, serde_json::Value>>,
-) -> serde_json::Value {
-    let architecture_id = normalize_architecture_id(architecture_id);
-    let Some(architecture) = get_architecture(architecture_id) else {
-        return responses::admin_provider_ops_action_not_supported(
-            "query_balance",
-            ADMIN_PROVIDER_OPS_ACTION_RUST_ONLY_MESSAGE,
-        );
-    };
-    let headers = match build_headers(architecture.architecture_id, connector_config, credentials) {
-        Ok(headers) => headers,
-        Err(message) => {
-            return responses::admin_provider_ops_action_not_configured("query_balance", message);
-        }
-    };
-    let Some(action_config) = resolve_action_config(
-        architecture_id,
-        provider_ops_config,
-        "query_balance",
-        request_config,
-    ) else {
-        return responses::admin_provider_ops_action_not_supported(
-            "query_balance",
-            ADMIN_PROVIDER_OPS_ACTION_RUST_ONLY_MESSAGE,
-        );
-    };
-
-    query_balance::admin_provider_ops_run_query_balance_action(
-        state,
-        provider_id,
-        provider,
-        &architecture,
-        base_url,
-        &action_config,
-        &headers,
-        credentials,
-        None,
-    )
-    .await
-}
-
 pub(crate) async fn admin_provider_ops_local_action_response(
     state: &AdminAppState<'_>,
     provider_id: &str,

apps/aether-gateway/src/handlers/admin/provider/ops/providers/actions/query_balance/mod.rs

@@ -111,14 +111,11 @@ pub(super) async fn admin_provider_ops_run_query_balance_action(
 
     if status != http::StatusCode::OK {
         let cookie_auth = architecture.query_balance_cookie_auth_errors;
-        let new_api_token_auth = architecture.architecture_id == "new_api";
         return match status {
             http::StatusCode::UNAUTHORIZED => admin_provider_ops_action_error(
                 "auth_failed",
                 "query_balance",
-                if new_api_token_auth {
-                    "访问令牌无效，请使用 New API 个人安全设置里的访问令牌"
-                } else if cookie_auth {
+                if cookie_auth {
                     "Cookie 已失效，请重新配置"
                 } else {
                     "认证失败"
@@ -128,9 +125,7 @@ pub(super) async fn admin_provider_ops_run_query_balance_action(
             http::StatusCode::FORBIDDEN => admin_provider_ops_action_error(
                 "auth_failed",
                 "query_balance",
-                if new_api_token_auth {
-                    "访问令牌无效或无权限，请使用 New API 个人安全设置里的访问令牌"
-                } else if cookie_auth {
+                if cookie_auth {
                     "Cookie 已失效或无权限"
                 } else {
                     "无权限访问"

apps/aether-gateway/src/handlers/admin/provider/ops/providers/actions/query_balance/sub2api.rs

@@ -8,9 +8,7 @@ use super::super::responses::{
 };
 use super::super::support::admin_provider_ops_json_object_map;
 use crate::handlers::admin::request::AdminAppState;
-use aether_admin::provider::ops::{
-    parse_sub2api_api_key_usage_payload, parse_sub2api_balance_payload,
-};
+use aether_admin::provider::ops::parse_sub2api_balance_payload;
 use aether_contracts::ProxySnapshot;
 use aether_data_contracts::repository::provider_catalog::StoredProviderCatalogProvider;
 use serde_json::{json, Value};
@@ -26,23 +24,6 @@ pub(super) async fn admin_provider_ops_sub2api_balance_payload(
     proxy_snapshot: Option<&ProxySnapshot>,
 ) -> serde_json::Value {
     let start = std::time::Instant::now();
-    if let Some(api_key) = credentials
-        .get("api_key")
-        .and_then(Value::as_str)
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-    {
-        return admin_provider_ops_sub2api_api_key_balance_payload(
-            state,
-            provider_id,
-            base_url,
-            action_config,
-            api_key,
-            proxy_snapshot,
-            start,
-        )
-        .await;
-    }
     let (access_token, updated_credentials, _frontend_updated_credentials) =
         match admin_provider_ops_sub2api_exchange_token(
             state,
@@ -210,117 +191,6 @@ pub(super) async fn admin_provider_ops_sub2api_balance_payload(
     )
 }
 
-async fn admin_provider_ops_sub2api_api_key_balance_payload(
-    state: &AdminAppState<'_>,
-    provider_id: &str,
-    base_url: &str,
-    action_config: &serde_json::Map<String, serde_json::Value>,
-    api_key: &str,
-    proxy_snapshot: Option<&ProxySnapshot>,
-    start: std::time::Instant,
-) -> serde_json::Value {
-    let usage_endpoint = action_config
-        .get("api_key_usage_endpoint")
-        .or_else(|| action_config.get("usage_endpoint"))
-        .and_then(serde_json::Value::as_str)
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-        .unwrap_or("/v1/usage");
-    let usage_url = admin_provider_ops_sub2api_request_url(base_url, usage_endpoint);
-    let auth_value = match reqwest::header::HeaderValue::from_str(&format!("Bearer {api_key}")) {
-        Ok(value) => value,
-        Err(_) => {
-            return admin_provider_ops_action_error(
-                "parse_error",
-                "query_balance",
-                "API Key 格式无效",
-                Some(start.elapsed().as_millis() as u64),
-            );
-        }
-    };
-    let auth_headers = reqwest::header::HeaderMap::from_iter([
-        (reqwest::header::AUTHORIZATION, auth_value),
-        (
-            reqwest::header::ACCEPT,
-            reqwest::header::HeaderValue::from_static("application/json"),
-        ),
-    ]);
-    let request_id = format!("provider-ops-action:sub2api:usage:{provider_id}");
-    let result = admin_provider_ops_execute_json_request(
-        state,
-        &request_id,
-        reqwest::Method::GET,
-        &usage_url,
-        &auth_headers,
-        None,
-        proxy_snapshot,
-    )
-    .await;
-    let response_time_ms = Some(start.elapsed().as_millis() as u64);
-    let (status, response_json) = match result {
-        Ok(result) => result,
-        Err(AdminProviderOpsExecuteJsonError::InvalidJson(message))
-        | Err(AdminProviderOpsExecuteJsonError::Transport(message)) => {
-            return admin_provider_ops_action_error(
-                "network_error",
-                "query_balance",
-                network_error_message(&message),
-                response_time_ms,
-            );
-        }
-    };
-    if matches!(
-        status,
-        http::StatusCode::UNAUTHORIZED | http::StatusCode::FORBIDDEN
-    ) {
-        return admin_provider_ops_action_error(
-            "auth_failed",
-            "query_balance",
-            "认证失败，请检查 API Key",
-            response_time_ms,
-        );
-    }
-    if status != http::StatusCode::OK {
-        return admin_provider_ops_action_error(
-            "unknown_error",
-            "query_balance",
-            format!(
-                "HTTP {}: {}",
-                status.as_u16(),
-                status.canonical_reason().unwrap_or("Unknown")
-            ),
-            response_time_ms,
-        );
-    }
-
-    let data = match parse_sub2api_api_key_usage_payload(action_config, &response_json) {
-        Ok(payload) => payload,
-        Err(message) => {
-            return admin_provider_ops_action_error(
-                if message.contains("无效") {
-                    "auth_failed"
-                } else if message == "响应格式无效" {
-                    "parse_error"
-                } else {
-                    "unknown_error"
-                },
-                "query_balance",
-                message,
-                response_time_ms,
-            );
-        }
-    };
-
-    admin_provider_ops_action_response(
-        "success",
-        "query_balance",
-        data,
-        None,
-        response_time_ms,
-        86400,
-    )
-}
-
 fn network_error_message(error: &str) -> String {
     let normalized = error.trim();
     let lower = normalized.to_ascii_lowercase();

apps/aether-gateway/src/handlers/admin/provider/ops/providers/balance_cache.rs

@@ -70,7 +70,7 @@ pub(super) async fn read_admin_provider_ops_balance_cache(
     }
 }
 
-pub(super) async fn store_admin_provider_ops_balance_cache(
+pub(crate) async fn store_admin_provider_ops_balance_cache(
     state: &AdminAppState<'_>,
     provider_id: &str,
     payload: &Value,

apps/aether-gateway/src/handlers/admin/provider/ops/providers/config.rs

@@ -1,4 +1,7 @@
-use super::support::{AdminProviderOpsSaveConfigRequest, ADMIN_PROVIDER_OPS_SENSITIVE_FIELDS};
+use super::support::{
+    AdminProviderOpsQuotaAlertConfigRequest, AdminProviderOpsSaveConfigRequest,
+    ADMIN_PROVIDER_OPS_SENSITIVE_FIELDS,
+};
 use crate::handlers::admin::request::AdminAppState;
 use crate::GatewayError;
 use aether_admin::provider::ops as admin_provider_ops_pure;
@@ -8,6 +11,10 @@ use aether_data_contracts::repository::provider_catalog::{
 use serde_json::json;
 use std::time::{SystemTime, UNIX_EPOCH};
 
+const PROVIDER_OPS_QUOTA_ALERT_DEFAULT_FETCH_INTERVAL_SECS: u64 = 30;
+const PROVIDER_OPS_QUOTA_ALERT_MIN_FETCH_INTERVAL_SECS: u64 = 30;
+const PROVIDER_OPS_QUOTA_ALERT_MAX_FETCH_INTERVAL_SECS: u64 = 86_400;
+
 pub(super) fn admin_provider_ops_config_object(
     provider: &StoredProviderCatalogProvider,
 ) -> Option<&serde_json::Map<String, serde_json::Value>> {
@@ -250,9 +257,6 @@ pub(super) fn build_admin_provider_ops_saved_config_value(
     provider: &StoredProviderCatalogProvider,
     payload: AdminProviderOpsSaveConfigRequest,
 ) -> Result<serde_json::Value, String> {
-    let architecture_id =
-        admin_provider_ops_pure::normalize_architecture_id(payload.architecture_id.as_str())
-            .to_string();
     let auth_type = payload.connector.auth_type.trim().to_string();
     if auth_type.is_empty() || !admin_provider_ops_is_supported_auth_type(auth_type.as_str()) {
         return Err("connector.auth_type 必须是合法的认证类型".to_string());
@@ -260,7 +264,7 @@ pub(super) fn build_admin_provider_ops_saved_config_value(
 
     let merged_credentials = admin_provider_ops_merge_credentials(
         state,
-        architecture_id.as_str(),
+        payload.architecture_id.as_str(),
         provider,
         payload.connector.credentials,
     );
@@ -279,9 +283,10 @@ pub(super) fn build_admin_provider_ops_saved_config_value(
             )
         })
         .collect::<serde_json::Map<String, serde_json::Value>>();
+    let quota_alert = normalize_admin_provider_ops_quota_alert(payload.quota_alert)?;
 
     Ok(json!({
-        "architecture_id": architecture_id,
+        "architecture_id": payload.architecture_id,
         "base_url": payload.base_url,
         "connector": {
             "auth_type": auth_type,
@@ -290,9 +295,48 @@ pub(super) fn build_admin_provider_ops_saved_config_value(
         },
         "actions": actions,
         "schedule": payload.schedule,
+        "quota_alert": quota_alert,
     }))
 }
 
+fn normalize_admin_provider_ops_quota_alert(
+    request: Option<AdminProviderOpsQuotaAlertConfigRequest>,
+) -> Result<serde_json::Value, String> {
+    let Some(request) = request else {
+        return Ok(default_admin_provider_ops_quota_alert());
+    };
+    let threshold_amount = request.threshold_amount.unwrap_or(0.0);
+    if threshold_amount < 0.0 {
+        return Err("quota_alert.threshold_amount 必须大于等于 0".to_string());
+    }
+    let fetch_interval_seconds = request
+        .fetch_interval_seconds
+        .unwrap_or(PROVIDER_OPS_QUOTA_ALERT_DEFAULT_FETCH_INTERVAL_SECS);
+    if !(PROVIDER_OPS_QUOTA_ALERT_MIN_FETCH_INTERVAL_SECS
+        ..=PROVIDER_OPS_QUOTA_ALERT_MAX_FETCH_INTERVAL_SECS)
+        .contains(&fetch_interval_seconds)
+    {
+        return Err(format!(
+            "quota_alert.fetch_interval_seconds 必须在 {} 到 {} 秒之间",
+            PROVIDER_OPS_QUOTA_ALERT_MIN_FETCH_INTERVAL_SECS,
+            PROVIDER_OPS_QUOTA_ALERT_MAX_FETCH_INTERVAL_SECS
+        ));
+    }
+    Ok(json!({
+        "enabled": request.enabled,
+        "threshold_amount": threshold_amount,
+        "fetch_interval_seconds": fetch_interval_seconds,
+    }))
+}
+
+fn default_admin_provider_ops_quota_alert() -> serde_json::Value {
+    json!({
+        "enabled": false,
+        "threshold_amount": 0.0,
+        "fetch_interval_seconds": PROVIDER_OPS_QUOTA_ALERT_DEFAULT_FETCH_INTERVAL_SECS,
+    })
+}
+
 pub(super) fn resolve_admin_provider_ops_base_url(
     provider: &StoredProviderCatalogProvider,
     endpoints: &[StoredProviderCatalogEndpoint],
@@ -331,16 +375,14 @@ pub(super) fn build_admin_provider_ops_config_payload(
         });
     };
     let connector = admin_provider_ops_connector_object(provider_ops_config);
-    let architecture_id = provider_ops_config
-        .get("architecture_id")
-        .and_then(serde_json::Value::as_str)
-        .map(admin_provider_ops_pure::normalize_architecture_id)
-        .unwrap_or("generic_api");
 
     json!({
         "provider_id": provider_id,
         "is_configured": true,
-        "architecture_id": architecture_id,
+        "architecture_id": provider_ops_config
+            .get("architecture_id")
+            .and_then(serde_json::Value::as_str)
+            .unwrap_or("generic_api"),
         "base_url": resolve_admin_provider_ops_base_url(
             provider,
             endpoints,
@@ -361,5 +403,10 @@ pub(super) fn build_admin_provider_ops_config_payload(
                 connector.and_then(|connector| connector.get("credentials")),
             ),
         },
+        "quota_alert": provider_ops_config
+            .get("quota_alert")
+            .filter(|value| value.is_object())
+            .cloned()
+            .unwrap_or_else(default_admin_provider_ops_quota_alert),
     })
 }

apps/aether-gateway/src/handlers/admin/provider/ops/providers/mod.rs

@@ -4,4 +4,5 @@ mod config;
 mod routes;
 mod support;
 mod verify;
+pub(crate) use self::balance_cache::store_admin_provider_ops_balance_cache;
 pub(super) use self::routes::maybe_build_local_admin_provider_ops_providers_response;

apps/aether-gateway/src/handlers/admin/provider/ops/providers/support.rs

@@ -33,6 +33,8 @@ pub(super) struct AdminProviderOpsSaveConfigRequest {
     pub(crate) actions: BTreeMap<String, AdminProviderOpsActionConfigRequest>,
     #[serde(default)]
     pub(crate) schedule: BTreeMap<String, String>,
+    #[serde(default)]
+    pub(crate) quota_alert: Option<AdminProviderOpsQuotaAlertConfigRequest>,
 }
 
 #[derive(Debug, Deserialize)]
@@ -52,6 +54,16 @@ pub(super) struct AdminProviderOpsActionConfigRequest {
     pub(crate) config: serde_json::Map<String, serde_json::Value>,
 }
 
+#[derive(Debug, Deserialize)]
+pub(super) struct AdminProviderOpsQuotaAlertConfigRequest {
+    #[serde(default)]
+    pub(crate) enabled: bool,
+    #[serde(default, deserialize_with = "deserialize_optional_f64_from_number")]
+    pub(crate) threshold_amount: Option<f64>,
+    #[serde(default)]
+    pub(crate) fetch_interval_seconds: Option<u64>,
+}
+
 #[derive(Debug, Deserialize)]
 pub(super) struct AdminProviderOpsConnectRequest {
     #[serde(default)]
@@ -71,3 +83,28 @@ fn default_admin_provider_ops_architecture_id() -> String {
 fn default_admin_provider_ops_action_enabled() -> bool {
     true
 }
+
+fn deserialize_optional_f64_from_number<'de, D>(deserializer: D) -> Result<Option<f64>, D::Error>
+where
+    D: serde::Deserializer<'de>,
+{
+    let value = Option::<serde_json::Value>::deserialize(deserializer)?;
+    match value {
+        None | Some(serde_json::Value::Null) => Ok(None),
+        Some(serde_json::Value::Number(number)) => number
+            .as_f64()
+            .filter(|value| value.is_finite())
+            .map(Some)
+            .ok_or_else(|| serde::de::Error::custom("expected a finite number")),
+        Some(serde_json::Value::String(raw)) => raw
+            .trim()
+            .parse::<f64>()
+            .ok()
+            .filter(|value| value.is_finite())
+            .map(Some)
+            .ok_or_else(|| serde::de::Error::custom("expected a finite number or numeric string")),
+        Some(_) => Err(serde::de::Error::custom(
+            "expected a finite number or numeric string",
+        )),
+    }
+}

apps/aether-gateway/src/handlers/admin/provider/ops/providers/verify/sub2api.rs

@@ -6,8 +6,7 @@ use crate::handlers::admin::provider::ops::providers::config::persist_admin_prov
 use crate::handlers::admin::request::AdminAppState;
 use aether_admin::provider::ops::{
     admin_provider_ops_frontend_updated_credentials, admin_provider_ops_verify_failure,
-    admin_provider_ops_verify_success, admin_provider_ops_verify_user_payload,
-    parse_sub2api_api_key_usage_payload, parse_verify_payload, ADMIN_PROVIDER_OPS_USER_AGENT,
+    parse_verify_payload, ADMIN_PROVIDER_OPS_USER_AGENT,
 };
 use aether_contracts::ProxySnapshot;
 use aether_data_contracts::repository::provider_catalog::StoredProviderCatalogProvider;
@@ -22,21 +21,6 @@ pub(super) async fn admin_provider_ops_local_sub2api_verify_response(
     credentials: &Map<String, Value>,
     proxy_snapshot: Option<&ProxySnapshot>,
 ) -> Value {
-    if let Some(api_key) = credentials
-        .get("api_key")
-        .and_then(Value::as_str)
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-    {
-        return admin_provider_ops_local_sub2api_api_key_verify_response(
-            state,
-            base_url,
-            api_key,
-            proxy_snapshot,
-        )
-        .await;
-    }
-
     let (access_token, updated_credentials, frontend_updated_credentials) =
         match admin_provider_ops_sub2api_exchange_token(
             state,
@@ -109,78 +93,6 @@ pub(super) async fn admin_provider_ops_local_sub2api_verify_response(
     )
 }
 
-async fn admin_provider_ops_local_sub2api_api_key_verify_response(
-    state: &AdminAppState<'_>,
-    base_url: &str,
-    api_key: &str,
-    proxy_snapshot: Option<&ProxySnapshot>,
-) -> Value {
-    let usage_url = admin_provider_ops_sub2api_request_url(base_url, "/v1/usage");
-    let auth_value = match reqwest::header::HeaderValue::from_str(&format!("Bearer {api_key}")) {
-        Ok(value) => value,
-        Err(_) => return admin_provider_ops_verify_failure("API Key 格式无效"),
-    };
-    let auth_headers = reqwest::header::HeaderMap::from_iter([
-        (reqwest::header::AUTHORIZATION, auth_value),
-        (
-            reqwest::header::ACCEPT,
-            reqwest::header::HeaderValue::from_static("application/json"),
-        ),
-    ]);
-    let auth_headers =
-        admin_provider_ops_headers_with_transport_controls(&auth_headers, None, true);
-    let (status, response_json) = match admin_provider_ops_execute_json_request(
-        state,
-        "provider-ops-verify:sub2api:api-key",
-        reqwest::Method::GET,
-        &usage_url,
-        &auth_headers,
-        None,
-        proxy_snapshot,
-    )
-    .await
-    {
-        Ok(result) => result,
-        Err(AdminProviderOpsExecuteJsonError::InvalidJson(message))
-        | Err(AdminProviderOpsExecuteJsonError::Transport(message)) => {
-            return admin_provider_ops_verify_failure(
-                admin_provider_ops_verify_execution_error_message(&message),
-            );
-        }
-    };
-    if matches!(
-        status,
-        http::StatusCode::UNAUTHORIZED | http::StatusCode::FORBIDDEN
-    ) {
-        return admin_provider_ops_verify_failure("认证失败：API Key 无效或已过期");
-    }
-    if status != http::StatusCode::OK {
-        return admin_provider_ops_verify_failure(format!("验证失败：HTTP {}", status.as_u16()));
-    }
-
-    let payload = match parse_sub2api_api_key_usage_payload(&Map::new(), &response_json) {
-        Ok(payload) => payload,
-        Err(message) => return admin_provider_ops_verify_failure(message),
-    };
-    let quota = payload.get("total_available").and_then(Value::as_f64);
-    let extra = payload
-        .get("extra")
-        .and_then(Value::as_object)
-        .cloned()
-        .unwrap_or_default();
-
-    admin_provider_ops_verify_success(
-        admin_provider_ops_verify_user_payload(
-            Some("Sub2API API Key".to_string()),
-            Some("Sub2API API Key".to_string()),
-            None,
-            quota,
-            Some(extra),
-        ),
-        None,
-    )
-}
-
 // 对齐 Python httpx.AsyncClient(base_url=...) 的行为:
 // 以 "/" 开头的端点始终相对站点根路径解析，而不是简单字符串拼接。
 pub(in super::super) fn admin_provider_ops_sub2api_request_url(

apps/aether-gateway/src/handlers/admin/provider/query/models/model_test.rs

@@ -14,7 +14,7 @@ use super::{provider_query_key_display_name, provider_query_provider_payload};
 use crate::ai_serving::{
     maybe_build_sync_finalize_outcome, GatewayControlDecision,
     ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME, GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND,
-    OPENAI_IMAGE_SYNC_FINALIZE_REPORT_KIND,
+    OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND, OPENAI_IMAGE_SYNC_FINALIZE_REPORT_KIND,
 };
 use crate::clock::current_unix_ms;
 use crate::execution_runtime;
@@ -1486,6 +1486,17 @@ fn provider_query_decode_execution_body(
         .and_then(|value| base64::engine::general_purpose::STANDARD.decode(value).ok())
 }
 
+fn provider_query_execution_json_body(result: &aether_contracts::ExecutionResult) -> Option<Value> {
+    result
+        .body
+        .as_ref()
+        .and_then(|body| body.json_body.clone())
+        .or_else(|| {
+            provider_query_decode_execution_body(result)
+                .and_then(|body| serde_json::from_slice::<Value>(&body).ok())
+        })
+}
+
 fn provider_query_aggregate_standard_stream_sync_response(
     provider_api_format: &str,
     body: &[u8],
@@ -1505,15 +1516,11 @@ fn provider_query_standard_execution_response_body(
     provider_api_format: &str,
     result: &aether_contracts::ExecutionResult,
 ) -> Option<Value> {
-    let body = result
-        .body
-        .as_ref()
-        .and_then(|body| body.json_body.clone())
-        .or_else(|| {
-            provider_query_decode_execution_body(result).and_then(|body| {
-                provider_query_aggregate_standard_stream_sync_response(provider_api_format, &body)
-            })
-        })?;
+    let body = provider_query_execution_json_body(result).or_else(|| {
+        provider_query_decode_execution_body(result).and_then(|body| {
+            provider_query_aggregate_standard_stream_sync_response(provider_api_format, &body)
+        })
+    })?;
     if result.status_code < 400
         && provider_query_normalize_api_format_alias(provider_api_format)
             == "gemini:generate_content"
@@ -1527,10 +1534,8 @@ fn provider_query_standard_execution_response_body(
 fn provider_query_extract_error_message(
     result: &aether_contracts::ExecutionResult,
 ) -> Option<String> {
-    result
-        .body
+    provider_query_execution_json_body(result)
         .as_ref()
-        .and_then(|body| body.json_body.as_ref())
         .and_then(Value::as_object)
         .and_then(|value| {
             value
@@ -1589,7 +1594,7 @@ async fn provider_query_finalize_kiro_result(
         })),
         status_code: result.status_code,
         headers: result.headers.clone(),
-        body_json: result.body.as_ref().and_then(|body| body.json_body.clone()),
+        body_json: provider_query_execution_json_body(result),
         client_body_json: None,
         body_base64: result
             .body
@@ -1781,6 +1786,56 @@ async fn provider_query_execute_kiro_test_candidate(
     })
 }
 
+async fn provider_query_finalize_windsurf_result(
+    route_path: &str,
+    trace_id: &str,
+    requested_model: &str,
+    mapped_model: &str,
+    original_request_body: &Value,
+    result: &aether_contracts::ExecutionResult,
+) -> Result<Option<Value>, GatewayError> {
+    let decision = GatewayControlDecision::synthetic(
+        route_path,
+        Some("admin_proxy".to_string()),
+        Some("provider_query_manage".to_string()),
+        Some("test_model_failover".to_string()),
+        Some("openai:chat".to_string()),
+    );
+    let payload = GatewaySyncReportRequest {
+        trace_id: trace_id.to_string(),
+        report_kind: OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND.to_string(),
+        report_context: Some(json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "openai:chat",
+            "model": requested_model,
+            "mapped_model": mapped_model,
+            "needs_conversion": false,
+            "has_envelope": true,
+            "envelope_name": crate::provider_transport::windsurf::WINDSURF_ENVELOPE_NAME,
+            "original_request_body": original_request_body,
+        })),
+        status_code: result.status_code,
+        headers: result.headers.clone(),
+        body_json: result.body.as_ref().and_then(|body| body.json_body.clone()),
+        client_body_json: None,
+        body_base64: result
+            .body
+            .as_ref()
+            .and_then(|body| body.body_bytes_b64.clone()),
+        telemetry: result.telemetry.clone(),
+    };
+
+    let Some(outcome) = maybe_build_sync_finalize_outcome(trace_id, &decision, &payload)? else {
+        return Ok(None);
+    };
+    let bytes = to_bytes(outcome.response.into_body(), usize::MAX)
+        .await
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    serde_json::from_slice::<Value>(&bytes)
+        .map(Some)
+        .map_err(|err| GatewayError::Internal(err.to_string()))
+}
+
 fn provider_query_build_openai_image_test_request_body_for_route(
     payload: &Value,
     model: &str,
@@ -1821,6 +1876,7 @@ fn provider_query_chatgpt_web_image_internal_url(base_url: &str) -> String {
 
 fn provider_query_openai_image_test_upstream_url(
     transport: &AdminGatewayProviderTransportSnapshot,
+    request_path: Option<&str>,
     request_query: Option<&str>,
 ) -> String {
     if transport
@@ -1841,7 +1897,11 @@ fn provider_query_openai_image_test_upstream_url(
             crate::provider_transport::GROK_CHAT_PATH,
         )
     } else {
-        crate::provider_transport::build_openai_image_upstream_url(transport, request_query)
+        crate::provider_transport::build_openai_image_upstream_url(
+            transport,
+            request_path,
+            request_query,
+        )
     }
 }
 
@@ -1874,7 +1934,7 @@ async fn provider_query_finalize_openai_image_result(
         })),
         status_code: result.status_code,
         headers: result.headers.clone(),
-        body_json: result.body.as_ref().and_then(|body| body.json_body.clone()),
+        body_json: provider_query_execution_json_body(result),
         client_body_json: None,
         body_base64: result
             .body
@@ -1967,13 +2027,23 @@ async fn provider_query_execute_openai_image_test_candidate(
         .provider_type
         .trim()
         .eq_ignore_ascii_case("grok");
+    let is_codex = transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("codex");
     let mut provider_request_body = if is_chatgpt_web {
         match crate::ai_serving::build_chatgpt_web_image_request_body(&parts, &request_body, None) {
             Ok(body) => body,
             Err(err) => err.to_error_json(),
         }
-    } else {
+    } else if is_codex || is_grok {
         crate::ai_serving::build_openai_image_provider_request_body(&normalized_request)
+    } else {
+        crate::ai_serving::build_openai_image_api_provider_request_body(
+            &normalized_request,
+            Some(candidate.effective_model.as_str()),
+        )
     };
     if !is_chatgpt_web {
         crate::ai_serving::apply_codex_openai_responses_special_body_edits(
@@ -2075,7 +2145,11 @@ async fn provider_query_execute_openai_image_test_candidate(
     } else {
         normalized_request.summary_json.clone()
     };
-    let request_url = provider_query_openai_image_test_upstream_url(&transport, parts.uri.query());
+    let request_url = provider_query_openai_image_test_upstream_url(
+        &transport,
+        Some(parts.uri.path()),
+        parts.uri.query(),
+    );
     let upstream_is_stream = provider_request_body
         .get("stream")
         .and_then(Value::as_bool)
@@ -2158,9 +2232,9 @@ async fn provider_query_execute_openai_image_test_candidate(
             &result,
         )
         .await?
-        .or_else(|| result.body.as_ref().and_then(|body| body.json_body.clone()))
+        .or_else(|| provider_query_execution_json_body(&result))
     } else {
-        result.body.as_ref().and_then(|body| body.json_body.clone())
+        provider_query_execution_json_body(&result)
     };
     let did_fail = result.status_code >= 400;
     let error_message = if did_fail {
@@ -2635,6 +2709,22 @@ async fn provider_query_execute_standard_test_candidate(
         route_path,
         client_api_format,
     );
+    if crate::provider_transport::is_windsurf_provider_transport(&transport)
+        && provider_query_normalize_api_format_alias(candidate.endpoint.api_format.as_str())
+            == "openai:chat"
+    {
+        return provider_query_execute_windsurf_test_candidate(
+            state,
+            provider,
+            candidate,
+            payload,
+            route_path,
+            trace_id,
+            transport,
+            original_request_body,
+        )
+        .await;
+    }
     if !provider_query_transport_supports_model_test_execution(
         state,
         &transport,
@@ -3042,6 +3132,182 @@ async fn provider_query_execute_standard_test_candidate(
     })
 }
 
+#[allow(clippy::too_many_arguments)]
+async fn provider_query_execute_windsurf_test_candidate(
+    state: &AdminAppState<'_>,
+    provider: &StoredProviderCatalogProvider,
+    candidate: &ProviderQueryTestCandidate,
+    payload: &Value,
+    route_path: &str,
+    trace_id: &str,
+    transport: AdminGatewayProviderTransportSnapshot,
+    original_request_body: Value,
+) -> Result<ProviderQueryExecutionOutcome, GatewayError> {
+    if let Some(_reason) =
+        crate::provider_transport::local_windsurf_request_transport_unsupported_reason_with_network(
+            &transport,
+        )
+    {
+        return Ok(provider_query_skipped_execution_outcome(
+            original_request_body,
+            provider_query_standard_test_unsupported_reason(
+                &transport,
+                candidate.endpoint.api_format.as_str(),
+            ),
+        ));
+    }
+
+    let incoming_request_headers = provider_query_extract_request_headers(payload);
+    let request_body = original_request_body.clone();
+    let request_model =
+        provider_query_request_body_model(&request_body, &candidate.effective_model);
+    let client_is_stream = request_body
+        .get("stream")
+        .and_then(Value::as_bool)
+        .unwrap_or(false);
+    let hard_requires_streaming = crate::ai_serving::force_upstream_streaming_for_provider(
+        transport.provider.provider_type.as_str(),
+        candidate.endpoint.api_format.as_str(),
+    );
+    let upstream_is_stream = crate::ai_serving::resolve_upstream_is_stream_from_endpoint_config(
+        transport.endpoint.config.as_ref(),
+        client_is_stream,
+        hard_requires_streaming,
+    );
+    let Some((auth_header, auth_value)) =
+        crate::provider_transport::windsurf::resolve_windsurf_cascade_auth(&transport).or_else(
+            || crate::provider_transport::auth::resolve_local_openai_bearer_auth(&transport),
+        )
+    else {
+        return Ok(provider_query_skipped_execution_outcome(
+            request_body,
+            "Provider auth is unavailable for windsurf".to_string(),
+        ));
+    };
+
+    let mut synthetic_request = http::Request::builder()
+        .uri(route_path)
+        .body(())
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    *synthetic_request.headers_mut() = incoming_request_headers;
+    let (parts, _) = synthetic_request.into_parts();
+
+    let Some(provider_request_body) =
+        crate::provider_transport::build_windsurf_cascade_request_body(
+            &request_body,
+            request_model,
+            &auth_value,
+            transport.endpoint.body_rules.as_ref(),
+            Some(&parts.headers),
+            upstream_is_stream,
+        )
+    else {
+        return Ok(provider_query_skipped_execution_outcome(
+            request_body,
+            "Provider request body could not be built for windsurf".to_string(),
+        ));
+    };
+    let Some(request_url) = crate::provider_transport::build_windsurf_cascade_upstream_url(
+        transport.endpoint.base_url.as_str(),
+        parts.uri.query(),
+    ) else {
+        return Ok(provider_query_skipped_execution_outcome(
+            provider_request_body,
+            "Provider request URL is unavailable for windsurf".to_string(),
+        ));
+    };
+    let Some(request_headers) = crate::provider_transport::build_windsurf_cascade_headers(
+        &parts.headers,
+        &provider_request_body,
+        &request_body,
+        transport.endpoint.header_rules.as_ref(),
+        &auth_header,
+        &auth_value,
+        upstream_is_stream,
+    ) else {
+        return Ok(ProviderQueryExecutionOutcome {
+            status: "failed",
+            skip_reason: None,
+            error_message: Some("provider request headers build failed".to_string()),
+            status_code: None,
+            latency_ms: None,
+            request_url,
+            request_headers: BTreeMap::new(),
+            request_body: provider_request_body,
+            response_headers: BTreeMap::new(),
+            response_body: None,
+        });
+    };
+
+    let plan = ExecutionPlan {
+        request_id: trace_id.to_string(),
+        candidate_id: Some(format!("provider-query-{}", candidate.key.id)),
+        provider_name: Some(provider.name.clone()),
+        provider_id: provider.id.clone(),
+        endpoint_id: candidate.endpoint.id.clone(),
+        key_id: candidate.key.id.clone(),
+        method: "POST".to_string(),
+        url: request_url.clone(),
+        headers: request_headers.clone(),
+        content_type: Some("application/connect+json".to_string()),
+        content_encoding: None,
+        body: RequestBody::from_json(provider_request_body.clone()),
+        stream: upstream_is_stream,
+        client_api_format: "openai:chat".to_string(),
+        provider_api_format: candidate.endpoint.api_format.clone(),
+        model_name: Some(request_model.to_string()),
+        proxy: state
+            .resolve_transport_proxy_snapshot_with_tunnel_affinity(&transport)
+            .await,
+        transport_profile: state.resolve_transport_profile(&transport),
+        timeouts: state.resolve_transport_execution_timeouts(&transport),
+    };
+
+    let result = state
+        .execute_execution_runtime_sync_plan(Some(trace_id), &plan)
+        .await?;
+    let response_body = if result.status_code < 400 {
+        provider_query_finalize_windsurf_result(
+            route_path,
+            trace_id,
+            request_model,
+            request_model,
+            &request_body,
+            &result,
+        )
+        .await?
+    } else {
+        result.body.as_ref().and_then(|body| body.json_body.clone())
+    };
+    let missing_success_body = result.status_code < 400 && response_body.is_none();
+    let did_fail = result.status_code >= 400 || missing_success_body;
+    let error_message = if did_fail {
+        provider_query_extract_error_message(&result).or_else(|| {
+            missing_success_body.then(|| {
+                format!(
+                    "Provider returned HTTP {} without a model-test response body",
+                    result.status_code
+                )
+            })
+        })
+    } else {
+        None
+    };
+
+    Ok(ProviderQueryExecutionOutcome {
+        status: if did_fail { "failed" } else { "success" },
+        skip_reason: None,
+        error_message,
+        status_code: Some(result.status_code),
+        latency_ms: result.telemetry.as_ref().and_then(|value| value.elapsed_ms),
+        request_url,
+        request_headers,
+        request_body: provider_request_body,
+        response_headers: result.headers,
+        response_body,
+    })
+}
+
 async fn build_admin_provider_query_kiro_failover_response(
     state: &AdminAppState<'_>,
     payload: &Value,

apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/adapter.rs

@@ -46,6 +46,22 @@ pub(super) fn provider_query_standard_test_unsupported_reason(
     api_format: &str,
 ) -> String {
     let normalized_api_format = crate::ai_serving::normalize_api_format_alias(api_format);
+    if crate::provider_transport::is_windsurf_provider_transport(transport)
+        && normalized_api_format == "openai:chat"
+    {
+        let reason =
+            crate::provider_transport::local_windsurf_request_transport_unsupported_reason_with_network(
+                transport,
+            );
+        return match reason {
+            Some(reason) => format!(
+                "{} ({reason})",
+                provider_query_unsupported_test_api_format_message(api_format)
+            ),
+            None => provider_query_unsupported_test_api_format_message(api_format),
+        };
+    }
+
     let reason = match normalized_api_format.as_str() {
         "openai:chat" => {
             crate::provider_transport::policy::local_openai_chat_transport_unsupported_reason(
@@ -294,6 +310,15 @@ pub(super) fn provider_query_transport_supports_model_test_execution(
     transport: &AdminGatewayProviderTransportSnapshot,
     api_format: &str,
 ) -> bool {
+    if crate::provider_transport::is_windsurf_provider_transport(transport)
+        && provider_query_normalize_api_format_alias(api_format) == "openai:chat"
+    {
+        return crate::provider_transport::local_windsurf_request_transport_unsupported_reason_with_network(
+            transport,
+        )
+        .is_none();
+    }
+
     match provider_query_test_adapter_for_provider_api_format(
         transport.provider.provider_type.as_str(),
         api_format,

apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/summary.rs

@@ -42,9 +42,9 @@ pub(super) fn provider_query_test_attempt_payload(
         "status_code": execution.status_code,
         "latency_ms": execution.latency_ms,
         "request_url": execution.request_url,
-        "request_headers": provider_query_redact_diagnostic_headers(&execution.request_headers),
-        "request_body": execution.request_body,
-        "response_headers": provider_query_redact_diagnostic_headers(&execution.response_headers),
+        "request_headers": redacted_provider_query_headers(&execution.request_headers),
+        "request_body": redacted_provider_query_value(&execution.request_body),
+        "response_headers": redacted_provider_query_headers(&execution.response_headers),
         "response_body": execution.response_body,
     })
 }
@@ -172,34 +172,84 @@ fn provider_query_endpoint_route_payload(
     })
 }
 
-fn provider_query_redact_diagnostic_headers(
-    headers: &BTreeMap<String, String>,
-) -> BTreeMap<String, String> {
+fn redacted_provider_query_headers(headers: &BTreeMap<String, String>) -> BTreeMap<String, String> {
     headers
         .iter()
-        .map(|(name, value)| {
-            if provider_query_header_is_sensitive(name) {
-                (name.clone(), "<redacted>".to_string())
+        .map(|(key, value)| {
+            if provider_query_field_is_sensitive(key) {
+                (key.clone(), "[REDACTED]".to_string())
             } else {
-                (name.clone(), value.clone())
+                (key.clone(), value.clone())
             }
         })
         .collect()
 }
 
-fn provider_query_header_is_sensitive(name: &str) -> bool {
+fn redacted_provider_query_value(value: &Value) -> Value {
+    match value {
+        Value::Object(object) => Value::Object(
+            object
+                .iter()
+                .map(|(key, value)| {
+                    if provider_query_field_is_sensitive(key) {
+                        (key.clone(), Value::String("[REDACTED]".to_string()))
+                    } else {
+                        (key.clone(), redacted_provider_query_value(value))
+                    }
+                })
+                .collect(),
+        ),
+        Value::Array(items) => Value::Array(
+            items
+                .iter()
+                .map(redacted_provider_query_value)
+                .collect::<Vec<_>>(),
+        ),
+        other => other.clone(),
+    }
+}
+
+fn provider_query_field_is_sensitive(key: &str) -> bool {
+    let key = key.trim().to_ascii_lowercase();
+    let normalized = key
+        .chars()
+        .filter(|ch| ch.is_ascii_alphanumeric())
+        .collect::<String>();
+    if matches!(
+        normalized.as_str(),
+        "maxtokens"
+            | "maxoutputtokens"
+            | "inputtokens"
+            | "outputtokens"
+            | "prompttokens"
+            | "completiontokens"
+            | "totaltokens"
+    ) {
+        return false;
+    }
     matches!(
-        name.trim().to_ascii_lowercase().as_str(),
+        key.as_str(),
         "authorization"
             | "proxy-authorization"
             | "cookie"
             | "set-cookie"
-            | "x-api-key"
+            | "api_key"
+            | "apikey"
             | "api-key"
+            | "x-api-key"
             | "x-goog-api-key"
             | "anthropic-api-key"
             | "openai-api-key"
-    )
+            | "x-codeium-csrf-token"
+            | "access_token"
+            | "refresh_token"
+            | "id_token"
+            | "password"
+            | "secret"
+    ) || normalized.ends_with("token")
+        || normalized.contains("secret")
+        || normalized.contains("apikey")
+        || normalized.contains("authorization")
 }
 
 pub(super) fn provider_query_candidate_summary_payload(
@@ -309,34 +359,93 @@ pub(super) fn provider_query_candidate_summary_payload(
 
 #[cfg(test)]
 mod tests {
-    use super::*;
+    use super::{redacted_provider_query_headers, redacted_provider_query_value};
+    use serde_json::json;
+    use std::collections::BTreeMap;
 
     #[test]
-    fn provider_query_diagnostic_headers_redact_credentials() {
+    fn redacts_sensitive_provider_query_headers() {
         let headers = BTreeMap::from([
             ("cookie".to_string(), "sso=secret".to_string()),
-            ("authorization".to_string(), "Bearer secret".to_string()),
+            (
+                "authorization".to_string(),
+                "Bearer secret-token".to_string(),
+            ),
             ("x-goog-api-key".to_string(), "secret".to_string()),
             ("content-type".to_string(), "application/json".to_string()),
+            (
+                "x-codeium-csrf-token".to_string(),
+                "csrf-secret".to_string(),
+            ),
         ]);
 
-        let redacted = provider_query_redact_diagnostic_headers(&headers);
+        let redacted = redacted_provider_query_headers(&headers);
 
         assert_eq!(
             redacted.get("cookie").map(String::as_str),
-            Some("<redacted>")
+            Some("[REDACTED]")
         );
         assert_eq!(
             redacted.get("authorization").map(String::as_str),
-            Some("<redacted>")
+            Some("[REDACTED]")
         );
         assert_eq!(
             redacted.get("x-goog-api-key").map(String::as_str),
-            Some("<redacted>")
+            Some("[REDACTED]")
+        );
+        assert_eq!(
+            redacted.get("x-codeium-csrf-token").map(String::as_str),
+            Some("[REDACTED]")
         );
         assert_eq!(
             redacted.get("content-type").map(String::as_str),
             Some("application/json")
         );
     }
+
+    #[test]
+    fn redacts_sensitive_provider_query_request_body_fields() {
+        let body = json!({
+            "metadata": {
+                "apiKey": "devin-session-token$secret",
+                "ideName": "windsurf"
+            },
+            "messages": [{"role": "user", "content": "hello"}],
+            "stream": true
+        });
+
+        let redacted = redacted_provider_query_value(&body);
+
+        assert_eq!(
+            redacted.pointer("/metadata/apiKey"),
+            Some(&json!("[REDACTED]"))
+        );
+        assert_eq!(
+            redacted.pointer("/metadata/ideName"),
+            Some(&json!("windsurf"))
+        );
+        assert_eq!(redacted.pointer("/stream"), Some(&json!(true)));
+    }
+
+    #[test]
+    fn keeps_non_secret_token_count_fields_visible() {
+        let body = json!({
+            "maxTokens": 64,
+            "usage": {
+                "inputTokens": 10,
+                "outputTokens": 2,
+                "accessToken": "secret"
+            }
+        });
+
+        let redacted = redacted_provider_query_value(&body);
+
+        assert_eq!(redacted.pointer("/maxTokens"), Some(&json!(64)));
+        assert_eq!(redacted.pointer("/usage/inputTokens"), Some(&json!(10)));
+        assert_eq!(redacted.pointer("/usage/outputTokens"), Some(&json!(2)));
+        assert_eq!(
+            redacted.pointer("/usage/accessToken"),
+            Some(&json!("[REDACTED]"))
+        );
+    }
 }

apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/tests.rs

@@ -90,6 +90,44 @@ fn provider_query_test_request_body_defaults_missing_model() {
     assert_eq!(body["model"], json!("fallback-model"));
 }
 
+#[test]
+fn provider_query_execution_json_body_decodes_stream_encoded_json_response() {
+    use base64::Engine as _;
+
+    let body = json!({
+        "created": 1,
+        "data": [{
+            "url": "https://example.test/image.png"
+        }]
+    });
+    let encoded_body = base64::engine::general_purpose::STANDARD
+        .encode(serde_json::to_vec(&body).expect("test body should serialize"));
+    let result = aether_contracts::ExecutionResult {
+        request_id: "request-1".to_string(),
+        candidate_id: None,
+        status_code: 200,
+        headers: std::collections::BTreeMap::from([(
+            "content-type".to_string(),
+            "application/json".to_string(),
+        )]),
+        body: Some(aether_contracts::ResponseBody {
+            json_body: None,
+            body_bytes_b64: Some(encoded_body),
+        }),
+        telemetry: None,
+        error: None,
+    };
+
+    assert_eq!(
+        provider_query_execution_json_body(&result),
+        Some(body.clone())
+    );
+    assert_eq!(
+        provider_query_standard_execution_response_body("openai:image", &result),
+        Some(body)
+    );
+}
+
 #[test]
 fn provider_query_test_request_body_fills_empty_conversation() {
     let payload = json!({
@@ -850,17 +888,39 @@ fn provider_query_grok_image_test_uses_grok_app_chat_upstream_url() {
     let transport = sample_openai_image_transport("grok");
 
     assert_eq!(
-        provider_query_openai_image_test_upstream_url(&transport, Some("trace=1")),
+        provider_query_openai_image_test_upstream_url(
+            &transport,
+            Some("/v1/images/generations"),
+            Some("trace=1"),
+        ),
         "https://grok.com/rest/app-chat/conversations/new"
     );
 }
 
+#[test]
+fn provider_query_custom_image_test_uses_images_upstream_url() {
+    let transport = sample_openai_image_transport("custom");
+
+    assert_eq!(
+        provider_query_openai_image_test_upstream_url(
+            &transport,
+            Some("/v1/images/generations"),
+            Some("trace=1"),
+        ),
+        "https://grok.com/v1/images/generations?trace=1"
+    );
+}
+
 #[test]
 fn provider_query_chatgpt_web_image_test_uses_internal_upstream_url() {
     let transport = sample_openai_image_transport("chatgpt_web");
 
     assert_eq!(
-        provider_query_openai_image_test_upstream_url(&transport, Some("trace=1")),
+        provider_query_openai_image_test_upstream_url(
+            &transport,
+            Some("/v1/images/generations"),
+            Some("trace=1"),
+        ),
         "https://grok.com/__aether/chatgpt-web-image"
     );
 }

apps/aether-gateway/src/handlers/admin/provider/shared/paths/endpoint_keys.rs

@@ -12,13 +12,6 @@ pub(crate) fn admin_provider_id_for_refresh_quota(request_path: &str) -> Option<
         .map(ToOwned::to_owned)
 }
 
-pub(crate) fn admin_provider_id_for_key_balance(request_path: &str) -> Option<String> {
-    request_path
-        .strip_prefix("/api/admin/endpoints/providers/")?
-        .strip_suffix("/key-balance")
-        .map(ToOwned::to_owned)
-}
-
 pub(crate) fn admin_reveal_key_id(request_path: &str) -> Option<String> {
     request_path
         .strip_prefix("/api/admin/endpoints/keys/")?

apps/aether-gateway/src/handlers/admin/provider/shared/paths/mod.rs

@@ -15,9 +15,9 @@ pub(crate) use self::crud::{
     is_admin_providers_root,
 };
 pub(crate) use self::endpoint_keys::{
-    admin_clear_oauth_invalid_key_id, admin_export_key_id, admin_provider_id_for_key_balance,
-    admin_provider_id_for_keys, admin_provider_id_for_refresh_quota,
-    admin_reset_cycle_stats_key_id, admin_reveal_key_id, admin_update_key_id,
+    admin_clear_oauth_invalid_key_id, admin_export_key_id, admin_provider_id_for_keys,
+    admin_provider_id_for_refresh_quota, admin_reset_cycle_stats_key_id, admin_reveal_key_id,
+    admin_update_key_id,
 };
 pub(crate) use self::oauth::{
     admin_provider_oauth_batch_import_provider_id, admin_provider_oauth_batch_import_task_path,

apps/aether-gateway/src/handlers/admin/provider/summary/aggregates.rs

@@ -31,7 +31,7 @@ pub(crate) async fn build_admin_provider_summary_payload(
         active_global_model_ids_result,
     ) = tokio::join!(
         state.list_provider_catalog_endpoints_by_provider_ids(&provider_ids),
-        state.list_provider_catalog_keys_by_provider_ids(&provider_ids),
+        state.list_provider_catalog_key_summaries_by_provider_ids(&provider_ids),
         state.read_provider_quota_snapshot(provider_id),
         state.list_provider_model_stats(&provider_ids),
         state.list_active_global_model_ids_by_provider_ids(&provider_ids),
@@ -197,7 +197,7 @@ pub(crate) async fn build_admin_providers_summary_payload(
     } else {
         let (endpoints_result, keys_result, model_stats_result, active_global_model_refs_result) = tokio::join!(
             state.list_provider_catalog_endpoints_by_provider_ids(&provider_ids),
-            state.list_provider_catalog_keys_by_provider_ids(&provider_ids),
+            state.list_provider_catalog_key_summaries_by_provider_ids(&provider_ids),
             state.list_provider_model_stats(&provider_ids),
             state.list_active_global_model_ids_by_provider_ids(&provider_ids),
         );

apps/aether-gateway/src/handlers/admin/provider/summary/value.rs

@@ -7,7 +7,7 @@ use aether_data_contracts::repository::candidates::{
 use aether_data_contracts::repository::provider_catalog::{
     StoredProviderCatalogEndpoint, StoredProviderCatalogKey, StoredProviderCatalogProvider,
 };
-use serde_json::{json, Map, Value};
+use serde_json::json;
 use std::collections::BTreeMap;
 
 fn json_truthy(value: &serde_json::Value) -> bool {
@@ -27,80 +27,6 @@ fn endpoint_timestamp_or_now(value: Option<u64>, now_unix_secs: u64) -> serde_js
         .unwrap_or(serde_json::Value::Null)
 }
 
-fn finite_json_number(value: Option<&Value>) -> Option<f64> {
-    match value {
-        Some(Value::Number(number)) => number.as_f64().filter(|value| value.is_finite()),
-        Some(Value::String(value)) => value
-            .trim()
-            .parse::<f64>()
-            .ok()
-            .filter(|value| value.is_finite()),
-        _ => None,
-    }
-}
-
-fn finite_json_u64(value: Option<&Value>) -> Option<u64> {
-    finite_json_number(value).and_then(|value| {
-        if value >= 0.0 {
-            Some(value as u64)
-        } else {
-            None
-        }
-    })
-}
-
-fn latest_key_balance_summary(keys: &[StoredProviderCatalogKey]) -> Value {
-    let mut selected: Option<(u64, &StoredProviderCatalogKey, &Map<String, Value>)> = None;
-
-    for key in keys {
-        let Some(balance) = key
-            .upstream_metadata
-            .as_ref()
-            .and_then(|metadata| metadata.get("balance_query"))
-            .and_then(Value::as_object)
-        else {
-            continue;
-        };
-        let Some(updated_at) = finite_json_u64(balance.get("updated_at")) else {
-            continue;
-        };
-        let has_balance_value = ["total_available", "total_used", "total_granted"]
-            .into_iter()
-            .any(|field| finite_json_number(balance.get(field)).is_some());
-        if !has_balance_value {
-            continue;
-        }
-
-        if selected
-            .as_ref()
-            .is_none_or(|(selected_updated_at, _, _)| updated_at > *selected_updated_at)
-        {
-            selected = Some((updated_at, key, balance));
-        }
-    }
-
-    let Some((updated_at, key, balance)) = selected else {
-        return Value::Null;
-    };
-
-    json!({
-        "key_id": key.id.clone(),
-        "key_name": key.name.clone(),
-        "updated_at": updated_at,
-        "architecture_id": balance.get("architecture_id").cloned().unwrap_or(Value::Null),
-        "status": balance.get("status").cloned().unwrap_or_else(|| json!("success")),
-        "executed_at": balance.get("executed_at").cloned().unwrap_or(Value::Null),
-        "response_time_ms": balance.get("response_time_ms").cloned().unwrap_or(Value::Null),
-        "total_available": balance.get("total_available").cloned().unwrap_or(Value::Null),
-        "total_used": balance.get("total_used").cloned().unwrap_or(Value::Null),
-        "total_granted": balance.get("total_granted").cloned().unwrap_or(Value::Null),
-        "currency": balance.get("currency").cloned().unwrap_or_else(|| json!("USD")),
-        "plan_name": balance.get("plan_name").cloned().unwrap_or(Value::Null),
-        "query_config": balance.get("query_config").cloned().unwrap_or(Value::Null),
-        "extra": balance.get("extra").cloned().unwrap_or(Value::Null),
-    })
-}
-
 pub(crate) fn build_admin_provider_summary_value(
     provider: &StoredProviderCatalogProvider,
     endpoints: &[StoredProviderCatalogEndpoint],
@@ -212,6 +138,13 @@ pub(crate) fn build_admin_provider_summary_value(
         .and_then(|cfg| cfg.get("simulated_cache_enabled"))
         .and_then(serde_json::Value::as_bool)
         .unwrap_or(false);
+    let ops_quota_alert_enabled = provider_ops_config
+        .and_then(serde_json::Value::as_object)
+        .and_then(|cfg| cfg.get("quota_alert"))
+        .and_then(serde_json::Value::as_object)
+        .and_then(|cfg| cfg.get("enabled"))
+        .and_then(serde_json::Value::as_bool)
+        .unwrap_or(false);
     let billing_type = quota_snapshot
         .map(|quota| quota.billing_type.clone())
         .or_else(|| provider.billing_type.clone());
@@ -232,7 +165,6 @@ pub(crate) fn build_admin_provider_summary_value(
         .and_then(|quota| quota.quota_expires_at_unix_secs)
         .or(provider.quota_expires_at_unix_secs)
         .and_then(unix_secs_to_rfc3339);
-    let key_balance_summary = latest_key_balance_summary(keys);
 
     json!({
         "id": provider.id.clone(),
@@ -271,8 +203,8 @@ pub(crate) fn build_admin_provider_summary_value(
         "endpoint_health_details": endpoint_health_details,
         "ops_configured": ops_configured,
         "ops_architecture_id": ops_architecture_id,
-        "key_balance_summary": key_balance_summary,
         "kiro_simulated_cache_enabled": kiro_simulated_cache_enabled,
+        "ops_quota_alert_enabled": ops_quota_alert_enabled,
         "created_at": endpoint_timestamp_or_now(provider.created_at_unix_ms, now_unix_secs),
         "updated_at": endpoint_timestamp_or_now(provider.updated_at_unix_secs, now_unix_secs),
     })

apps/aether-gateway/src/handlers/admin/provider/write/normalize.rs

@@ -4,9 +4,9 @@ pub(crate) fn normalize_provider_type_input(value: &str) -> Result<String, Strin
     let normalized = value.trim().to_ascii_lowercase();
     match normalized.as_str() {
         "custom" | "claude_code" | "kiro" | "codex" | "chatgpt_web" | "gemini_cli"
-        | "antigravity" | "vertex_ai" | "grok" => Ok(normalized),
+        | "antigravity" | "vertex_ai" | "grok" | "windsurf" => Ok(normalized),
         _ => Err(
-            "provider_type 仅支持 custom / claude_code / kiro / codex / chatgpt_web / gemini_cli / antigravity / vertex_ai / grok"
+            "provider_type 仅支持 custom / claude_code / kiro / codex / chatgpt_web / gemini_cli / antigravity / vertex_ai / grok / windsurf"
                 .to_string(),
         ),
     }

apps/aether-gateway/src/handlers/admin/request/provider/catalog.rs

@@ -177,21 +177,6 @@ impl<'a> AdminAppState<'a> {
         self.app.update_provider_catalog_key(key).await
     }
 
-    pub(crate) async fn update_provider_catalog_key_upstream_metadata(
-        &self,
-        key_id: &str,
-        upstream_metadata: Option<&serde_json::Value>,
-        updated_at_unix_secs: Option<u64>,
-    ) -> Result<bool, GatewayError> {
-        self.app
-            .update_provider_catalog_key_upstream_metadata(
-                key_id,
-                upstream_metadata,
-                updated_at_unix_secs,
-            )
-            .await
-    }
-
     pub(crate) async fn create_provider_catalog_key(
         &self,
         key: &aether_data_contracts::repository::provider_catalog::StoredProviderCatalogKey,

apps/aether-gateway/src/handlers/admin/request/system/import.rs

@@ -49,7 +49,7 @@ use std::collections::{BTreeMap, BTreeSet};
 use std::time::{SystemTime, UNIX_EPOCH};
 use uuid::Uuid;
 
-const ADMIN_SYSTEM_IMPORT_MAX_SIZE_BYTES: usize = 10 * 1024 * 1024;
+const ADMIN_SYSTEM_IMPORT_MAX_SIZE_BYTES: usize = 500 * 1024 * 1024;
 
 fn invalid_request(detail: impl Into<String>) -> (http::StatusCode, Value) {
     (
@@ -173,6 +173,9 @@ fn normalize_import_endpoint_format(value: &str) -> Result<String, String> {
     let normalized = match value.trim().to_ascii_lowercase().as_str() {
         "openai:cli" => "openai:responses",
         "openai:compact" => "openai:responses:compact",
+        "openai_image" | "images" | "image" | "/v1/images/generations" | "/v1/images/edits" => {
+            "openai:image"
+        }
         "claude:chat" | "claude:cli" => "claude:messages",
         "gemini:chat" | "gemini:cli" => "gemini:generate_content",
         _ => value.trim(),
@@ -956,7 +959,7 @@ impl<'a> AdminAppState<'a> {
         }
 
         if request_body.len() > ADMIN_SYSTEM_DATA_IMPORT_MAX_SIZE_BYTES {
-            return Ok(Err(invalid_request("请求体大小不能超过 20MB")));
+            return Ok(Err(invalid_request("请求体大小不能超过 500MB")));
         }
 
         let root = match serde_json::from_slice::<Value>(request_body) {
@@ -1053,7 +1056,7 @@ impl<'a> AdminAppState<'a> {
             )));
         }
         if request_body.len() > ADMIN_SYSTEM_IMPORT_MAX_SIZE_BYTES {
-            return Ok(Err(invalid_request("请求体大小不能超过 10MB")));
+            return Ok(Err(invalid_request("请求体大小不能超过 500MB")));
         }
 
         let parsed = routed!(parse_admin_system_config_import_request(request_body));
@@ -1993,7 +1996,7 @@ impl<'a> AdminAppState<'a> {
             )));
         }
         if request_body.len() > ADMIN_SYSTEM_IMPORT_MAX_SIZE_BYTES {
-            return Ok(Err(invalid_request("请求体大小不能超过 10MB")));
+            return Ok(Err(invalid_request("请求体大小不能超过 500MB")));
         }
 
         let root = match serde_json::from_slice::<Value>(request_body) {
@@ -3062,6 +3065,10 @@ mod tests {
         for (raw, expected) in [
             ("openai:cli", "openai:responses"),
             ("openai:compact", "openai:responses:compact"),
+            ("openai_image", "openai:image"),
+            ("images", "openai:image"),
+            ("/v1/images/generations", "openai:image"),
+            ("/v1/images/edits", "openai:image"),
             ("claude:chat", "claude:messages"),
             ("claude:cli", "claude:messages"),
             ("gemini:chat", "gemini:generate_content"),

apps/aether-gateway/src/handlers/admin/request/system/mod.rs

@@ -9,7 +9,7 @@ mod proxy_nodes;
 mod templates;
 
 const ADMIN_SYSTEM_DATA_EXPORT_VERSION: &str = "1.0";
-const ADMIN_SYSTEM_DATA_IMPORT_MAX_SIZE_BYTES: usize = 20 * 1024 * 1024;
+const ADMIN_SYSTEM_DATA_IMPORT_MAX_SIZE_BYTES: usize = 500 * 1024 * 1024;
 
 impl<'a> AdminAppState<'a> {
     pub(crate) async fn upsert_system_config_json_value(

apps/aether-gateway/src/handlers/admin/system/core/system_routes.rs

@@ -23,6 +23,7 @@ use crate::handlers::admin::system::shared::update::{
     read_update_history, read_update_task_status, start_admin_system_rollback_task,
     start_admin_system_update_task,
 };
+use crate::important_notification::build_important_notification_test_payload;
 use crate::maintenance::{ManualUsageCleanupMode, ManualUsageCleanupOptions};
 use crate::GatewayError;
 use aether_data_contracts::repository::usage::UsageCleanupTargets;
@@ -365,6 +366,16 @@ pub(super) async fn maybe_build_local_admin_core_system_response(
         ));
     }
 
+    if decision.route_kind.as_deref() == Some("important_notification_test")
+        && request_method == http::Method::POST
+        && request_path == "/api/admin/system/important-notification/test"
+    {
+        return Ok(Some(
+            Json(build_important_notification_test_payload(state, request_body).await?)
+                .into_response(),
+        ));
+    }
+
     if decision.route_kind.as_deref() == Some("cleanup") && request_method == http::Method::POST {
         return Ok(Some(attach_admin_audit_response(
             Json(build_admin_system_cleanup_payload(state).await?).into_response(),

apps/aether-gateway/src/handlers/admin/system/shared/configs.rs

@@ -33,6 +33,21 @@ fn admin_system_config_default_value(key: &str) -> Option<serde_json::Value> {
     admin_system_config_default_value_pure(key)
 }
 
+fn legacy_admin_system_config_fallback_key(normalized_key: &str) -> Option<&'static str> {
+    match normalized_key {
+        "module.server_chan_push.enabled" => {
+            Some("module.important_notification.server_chan_enabled")
+        }
+        "module.server_chan_push.send_key" => {
+            Some("module.important_notification.server_chan_send_key")
+        }
+        "module.server_chan_push.template" => {
+            Some("module.important_notification.server_chan_template")
+        }
+        _ => None,
+    }
+}
+
 pub(crate) fn build_admin_system_configs_payload(
     entries: &[aether_data::repository::system::StoredSystemConfigEntry],
 ) -> serde_json::Value {
@@ -44,12 +59,14 @@ pub(crate) async fn build_admin_system_config_detail_payload(
     requested_key: &str,
 ) -> Result<Result<serde_json::Value, (http::StatusCode, serde_json::Value)>, GatewayError> {
     let requested_key = requested_key.trim();
-    let value = state
-        .read_system_config_json_value(&normalize_admin_system_config_key(requested_key))
-        .await?
-        .or_else(|| {
-            admin_system_config_default_value(&normalize_admin_system_config_key(requested_key))
-        });
+    let normalized_key = normalize_admin_system_config_key(requested_key);
+    let mut value = state.read_system_config_json_value(&normalized_key).await?;
+    if value.is_none() {
+        if let Some(legacy_key) = legacy_admin_system_config_fallback_key(&normalized_key) {
+            value = state.read_system_config_json_value(legacy_key).await?;
+        }
+    }
+    let value = value.or_else(|| admin_system_config_default_value(&normalized_key));
     Ok(build_admin_system_config_detail_payload_pure(
         requested_key,
         value,

apps/aether-gateway/src/handlers/admin/system/shared/modules.rs

@@ -1,5 +1,11 @@
+use crate::bark_push::bark_push_configured;
 use crate::handlers::admin::request::AdminAppState;
 use crate::handlers::shared::{module_available_from_env, system_config_bool};
+use crate::important_notification::{
+    important_notification_configured, IMPORTANT_NOTIFICATION_ENABLED_KEY,
+    LEGACY_NOTIFICATION_EMAIL_ENABLED_KEY,
+};
+use crate::server_chan_push::server_chan_push_configured;
 use crate::system_features::ENABLE_MODEL_DIRECTIVES_CONFIG_KEY;
 use crate::GatewayError;
 use aether_admin::system as admin_system_kernel;
@@ -68,17 +74,41 @@ pub(crate) const ADMIN_MODULE_DEFINITIONS: &[AdminModuleDefinition] = &[
         admin_menu_order: 59,
     },
     AdminModuleDefinition {
-        name: "notification_email",
-        display_name: "异常通知",
-        description: "为 5xx 异常发送邮件通知，可在模块管理中启用或禁用",
+        name: "important_notification",
+        display_name: "通知服务",
+        description: "统一管理通知项、模板和推送服务选择，供后台任务和用户通知使用",
         category: "integration",
-        env_key: "NOTIFICATION_EMAIL_AVAILABLE",
+        env_key: "IMPORTANT_NOTIFICATION_AVAILABLE",
         default_available: true,
-        admin_route: None,
-        admin_menu_icon: Some("Mail"),
-        admin_menu_group: Some("system"),
+        admin_route: Some("/admin/notification-service"),
+        admin_menu_icon: Some("BellRing"),
+        admin_menu_group: None,
         admin_menu_order: 58,
     },
+    AdminModuleDefinition {
+        name: "server_chan_push",
+        display_name: "Server 酱推送",
+        description: "第三方推送服务，配置 Server 酱 Turbo SendKey 并测试微信推送",
+        category: "integration",
+        env_key: "SERVER_CHAN_PUSH_AVAILABLE",
+        default_available: true,
+        admin_route: Some("/admin/modules/server-chan"),
+        admin_menu_icon: Some("Send"),
+        admin_menu_group: Some("system"),
+        admin_menu_order: 59,
+    },
+    AdminModuleDefinition {
+        name: "bark_push",
+        display_name: "Bark 推送",
+        description: "第三方推送服务，配置 Bark Device Key 并测试 iOS 推送",
+        category: "integration",
+        env_key: "BARK_PUSH_AVAILABLE",
+        default_available: true,
+        admin_route: Some("/admin/modules/bark"),
+        admin_menu_icon: Some("Send"),
+        admin_menu_group: Some("system"),
+        admin_menu_order: 59,
+    },
     AdminModuleDefinition {
         name: "model_directives",
         display_name: "模型后缀参数",
@@ -150,10 +180,17 @@ pub(crate) struct AdminModuleRuntimeState {
     oauth_providers: Vec<aether_data::repository::auth_modules::StoredOAuthProviderModuleConfig>,
     ldap_config: Option<aether_data::repository::auth_modules::StoredLdapModuleConfig>,
     gemini_files_has_capable_key: bool,
-    smtp_configured: bool,
+    important_notification_configured: bool,
+    server_chan_push_configured: bool,
+    bark_push_configured: bool,
 }
 
 pub(crate) fn admin_module_by_name(name: &str) -> Option<&'static AdminModuleDefinition> {
+    let name = if name == "notification_email" {
+        "important_notification"
+    } else {
+        name
+    };
     ADMIN_MODULE_DEFINITIONS
         .iter()
         .find(|module| module.name == name)
@@ -170,11 +207,22 @@ pub(crate) fn admin_module_name_from_enabled_path(request_path: &str) -> Option<
 pub(crate) fn admin_module_enabled_config_key(module: &AdminModuleDefinition) -> String {
     if module.name == "model_directives" {
         ENABLE_MODEL_DIRECTIVES_CONFIG_KEY.to_string()
+    } else if module.name == "important_notification" {
+        IMPORTANT_NOTIFICATION_ENABLED_KEY.to_string()
     } else {
         format!("module.{}.enabled", module.name)
     }
 }
 
+fn admin_module_available(module: &AdminModuleDefinition) -> bool {
+    if module.name == "important_notification" {
+        let legacy_default =
+            module_available_from_env("NOTIFICATION_EMAIL_AVAILABLE", module.default_available);
+        return module_available_from_env(module.env_key, legacy_default);
+    }
+    module_available_from_env(module.env_key, module.default_available)
+}
+
 pub(crate) fn oauth_module_config_is_valid(
     providers: &[aether_data::repository::auth_modules::StoredOAuthProviderModuleConfig],
 ) -> bool {
@@ -221,28 +269,17 @@ pub(crate) async fn build_admin_module_runtime_state(
             })
     };
 
-    let smtp_host = state.read_system_config_json_value("smtp_host").await?;
-    let smtp_from_email = state
-        .read_system_config_json_value("smtp_from_email")
-        .await?;
-    let smtp_configured = smtp_host
-        .as_ref()
-        .and_then(serde_json::Value::as_str)
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-        .is_some()
-        && smtp_from_email
-            .as_ref()
-            .and_then(serde_json::Value::as_str)
-            .map(str::trim)
-            .filter(|value| !value.is_empty())
-            .is_some();
+    let notification_configured = important_notification_configured(state.app()).await?;
+    let server_chan_configured = server_chan_push_configured(state.app()).await?;
+    let bark_configured = bark_push_configured(state.app()).await?;
 
     Ok(AdminModuleRuntimeState {
         oauth_providers,
         ldap_config,
         gemini_files_has_capable_key,
-        smtp_configured,
+        important_notification_configured: notification_configured,
+        server_chan_push_configured: server_chan_configured,
+        bark_push_configured: bark_configured,
     })
 }
 
@@ -255,7 +292,9 @@ pub(crate) fn build_admin_module_validation_result(
         &runtime.oauth_providers,
         runtime.ldap_config.as_ref(),
         runtime.gemini_files_has_capable_key,
-        runtime.smtp_configured,
+        runtime.important_notification_configured,
+        runtime.server_chan_push_configured,
+        runtime.bark_push_configured,
     )
 }
 
@@ -274,12 +313,19 @@ pub(crate) async fn build_admin_module_status_payload(
     module: &AdminModuleDefinition,
     runtime: &AdminModuleRuntimeState,
 ) -> Result<serde_json::Value, GatewayError> {
-    let available = module_available_from_env(module.env_key, module.default_available);
+    let available = admin_module_available(module);
     let enabled = if available {
-        let enabled = state
+        let enabled_value = state
             .read_system_config_json_value(&admin_module_enabled_config_key(module))
             .await?;
-        system_config_bool(enabled.as_ref(), false)
+        let enabled_value = if module.name == "important_notification" && enabled_value.is_none() {
+            state
+                .read_system_config_json_value(LEGACY_NOTIFICATION_EMAIL_ENABLED_KEY)
+                .await?
+        } else {
+            enabled_value
+        };
+        system_config_bool(enabled_value.as_ref(), false)
     } else {
         false
     };

apps/aether-gateway/src/handlers/admin/system/shared/smtp.rs

@@ -1,14 +1,10 @@
+use crate::email_delivery::{probe_smtp_connection, system_config_u16, SmtpDeliveryConfig};
 use crate::handlers::admin::request::AdminAppState;
 use crate::handlers::shared::{system_config_bool, system_config_string};
 use crate::GatewayError;
 use axum::body::Bytes;
-use base64::Engine;
 use serde::Deserialize;
 use serde_json::json;
-use std::io::{BufRead, Write};
-use std::time::Duration;
-
-const SMTP_TIMEOUT_SECS: u64 = 30;
 
 #[derive(Debug, Default, Deserialize)]
 struct AdminSmtpTestRequest {
@@ -53,12 +49,12 @@ pub(crate) async fn build_admin_smtp_test_payload(
         }));
     }
 
-    let result = tokio::task::spawn_blocking(move || test_smtp_connection_blocking(config))
-        .await
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let result = probe_smtp_connection(config.into_delivery_config()).await;
     Ok(match result {
         Ok(()) => json!({ "success": true, "message": "SMTP 连接测试成功" }),
-        Err(error) => json!({ "success": false, "message": translate_smtp_error(&error) }),
+        Err(error) => {
+            json!({ "success": false, "message": translate_smtp_error(&smtp_gateway_error_message(&error)) })
+        }
     })
 }
 
@@ -94,8 +90,8 @@ async fn resolve_admin_smtp_config(
         port: request
             .smtp_port
             .as_ref()
-            .map(|value| system_config_u16(value, 587))
-            .unwrap_or_else(|| system_config_u16_opt(smtp_port.as_ref(), 587)),
+            .map(|value| system_config_u16(Some(value), 587))
+            .unwrap_or_else(|| system_config_u16(smtp_port.as_ref(), 587)),
         user: request
             .smtp_user
             .as_ref()
@@ -130,6 +126,21 @@ async fn resolve_admin_smtp_config(
     })
 }
 
+impl ResolvedSmtpConfig {
+    fn into_delivery_config(self) -> SmtpDeliveryConfig {
+        SmtpDeliveryConfig {
+            host: self.host.unwrap_or_default(),
+            port: self.port,
+            user: self.user,
+            password: self.password,
+            use_tls: self.use_tls,
+            use_ssl: self.use_ssl,
+            from_email: self.from_email.unwrap_or_default(),
+            from_name: self.from_name,
+        }
+    }
+}
+
 fn missing_smtp_fields(config: &ResolvedSmtpConfig) -> Vec<&'static str> {
     let mut fields = Vec::new();
     if config
@@ -171,178 +182,13 @@ fn missing_smtp_fields(config: &ResolvedSmtpConfig) -> Vec<&'static str> {
     fields
 }
 
-fn system_config_u16_opt(value: Option<&serde_json::Value>, default: u16) -> u16 {
-    value
-        .map(|value| system_config_u16(value, default))
-        .unwrap_or(default)
-}
-
-fn system_config_u16(value: &serde_json::Value, default: u16) -> u16 {
-    match value {
-        serde_json::Value::Number(value) => value
-            .as_u64()
-            .and_then(|value| u16::try_from(value).ok())
-            .unwrap_or(default),
-        serde_json::Value::String(value) => value.trim().parse::<u16>().unwrap_or(default),
-        _ => default,
+fn smtp_gateway_error_message(error: &GatewayError) -> String {
+    match error {
+        GatewayError::Internal(message) => message.clone(),
+        _ => format!("{error:?}"),
     }
 }
 
-fn build_tls_config() -> std::sync::Arc<rustls::ClientConfig> {
-    let _ = rustls::crypto::ring::default_provider().install_default();
-    let root_store =
-        rustls::RootCertStore::from_iter(webpki_roots::TLS_SERVER_ROOTS.iter().cloned());
-    std::sync::Arc::new(
-        rustls::ClientConfig::builder()
-            .with_root_certificates(root_store)
-            .with_no_client_auth(),
-    )
-}
-
-fn resolve_server_name(host: &str) -> Result<rustls::pki_types::ServerName<'static>, String> {
-    let host = host.trim().trim_start_matches('[').trim_end_matches(']');
-    if let Ok(ip) = host.parse::<std::net::IpAddr>() {
-        return Ok(rustls::pki_types::ServerName::from(ip));
-    }
-    rustls::pki_types::ServerName::try_from(host.to_string()).map_err(|err| err.to_string())
-}
-
-fn connect_tcp_stream(config: &ResolvedSmtpConfig) -> Result<std::net::TcpStream, String> {
-    let host = config.host.as_deref().unwrap_or_default();
-    let stream =
-        std::net::TcpStream::connect((host, config.port)).map_err(|err| err.to_string())?;
-    stream
-        .set_read_timeout(Some(Duration::from_secs(SMTP_TIMEOUT_SECS)))
-        .map_err(|err| err.to_string())?;
-    stream
-        .set_write_timeout(Some(Duration::from_secs(SMTP_TIMEOUT_SECS)))
-        .map_err(|err| err.to_string())?;
-    Ok(stream)
-}
-
-fn wrap_tls_stream(
-    stream: std::net::TcpStream,
-    host: &str,
-) -> Result<rustls::StreamOwned<rustls::ClientConnection, std::net::TcpStream>, String> {
-    let server_name = resolve_server_name(host)?;
-    let connection = rustls::ClientConnection::new(build_tls_config(), server_name)
-        .map_err(|err| err.to_string())?;
-    Ok(rustls::StreamOwned::new(connection, stream))
-}
-
-fn smtp_read_response<T: BufRead>(reader: &mut T) -> Result<(u16, String), String> {
-    let mut message = String::new();
-    let code = loop {
-        let mut line = String::new();
-        let bytes = reader.read_line(&mut line).map_err(|err| err.to_string())?;
-        if bytes == 0 {
-            return Err("smtp connection closed unexpectedly".to_string());
-        }
-        let trimmed = line.trim_end_matches(['\r', '\n']).to_string();
-        if trimmed.len() < 3 {
-            return Err("invalid smtp response".to_string());
-        }
-        let parsed_code = trimmed[..3].parse::<u16>().map_err(|err| err.to_string())?;
-        let continuation = trimmed.as_bytes().get(3).copied() == Some(b'-');
-        if !message.is_empty() {
-            message.push('\n');
-        }
-        message.push_str(&trimmed);
-        if !continuation {
-            break parsed_code;
-        }
-    };
-    Ok((code, message))
-}
-
-fn smtp_expect<T: BufRead>(reader: &mut T, allowed_codes: &[u16]) -> Result<String, String> {
-    let (code, message) = smtp_read_response(reader)?;
-    if allowed_codes.contains(&code) {
-        return Ok(message);
-    }
-    Err(format!("unexpected smtp response {code}: {message}"))
-}
-
-fn smtp_write_line<T: Write>(writer: &mut T, line: &str) -> Result<(), String> {
-    writer
-        .write_all(line.as_bytes())
-        .map_err(|err| err.to_string())?;
-    writer.write_all(b"\r\n").map_err(|err| err.to_string())?;
-    writer.flush().map_err(|err| err.to_string())
-}
-
-fn smtp_send_command<S: std::io::Read + Write>(
-    reader: &mut std::io::BufReader<S>,
-    command: &str,
-    allowed_codes: &[u16],
-) -> Result<String, String> {
-    smtp_write_line(reader.get_mut(), command)?;
-    smtp_expect(reader, allowed_codes)
-}
-
-fn smtp_authenticate<S: std::io::Read + Write>(
-    reader: &mut std::io::BufReader<S>,
-    config: &ResolvedSmtpConfig,
-) -> Result<(), String> {
-    let Some(username) = config
-        .user
-        .as_deref()
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-    else {
-        return Ok(());
-    };
-    let password = config.password.as_deref().unwrap_or_default();
-    smtp_send_command(reader, "AUTH LOGIN", &[334])?;
-    smtp_send_command(
-        reader,
-        &base64::engine::general_purpose::STANDARD.encode(username.as_bytes()),
-        &[334],
-    )?;
-    smtp_send_command(
-        reader,
-        &base64::engine::general_purpose::STANDARD.encode(password.as_bytes()),
-        &[235],
-    )?;
-    Ok(())
-}
-
-fn smtp_probe<S: std::io::Read + Write>(
-    reader: &mut std::io::BufReader<S>,
-    config: &ResolvedSmtpConfig,
-) -> Result<(), String> {
-    smtp_send_command(reader, "EHLO aether.local", &[250])?;
-    smtp_authenticate(reader, config)?;
-    let _ = smtp_send_command(reader, "QUIT", &[221]);
-    Ok(())
-}
-
-fn test_smtp_connection_blocking(config: ResolvedSmtpConfig) -> Result<(), String> {
-    if config.use_ssl {
-        let stream = connect_tcp_stream(&config)?;
-        let tls_stream = wrap_tls_stream(stream, config.host.as_deref().unwrap_or_default())?;
-        let mut reader = std::io::BufReader::new(tls_stream);
-        smtp_expect(&mut reader, &[220])?;
-        return smtp_probe(&mut reader, &config);
-    }
-
-    let stream = connect_tcp_stream(&config)?;
-    let mut reader = std::io::BufReader::new(stream);
-    smtp_expect(&mut reader, &[220])?;
-    smtp_send_command(&mut reader, "EHLO aether.local", &[250])?;
-    if config.use_tls {
-        smtp_send_command(&mut reader, "STARTTLS", &[220])?;
-        let stream = reader.into_inner();
-        let tls_stream = wrap_tls_stream(stream, config.host.as_deref().unwrap_or_default())?;
-        let mut reader = std::io::BufReader::new(tls_stream);
-        return smtp_probe(&mut reader, &config);
-    }
-
-    smtp_authenticate(&mut reader, &config)?;
-    let _ = smtp_send_command(&mut reader, "QUIT", &[221]);
-    Ok(())
-}
-
 fn translate_smtp_error(error: &str) -> String {
     let error_lower = error.to_ascii_lowercase();
 

apps/aether-gateway/src/handlers/public/support/auth_email.rs

@@ -1,11 +1,11 @@
 use super::{
-    decrypt_catalog_secret_with_fallbacks, escape_admin_email_template_html, json,
-    read_admin_email_template_payload, render_admin_email_template_html, system_config_bool,
-    system_config_string, system_config_u16, AppState, GatewayError,
+    escape_admin_email_template_html, json, read_admin_email_template_payload,
+    render_admin_email_template_html, system_config_string, AppState, GatewayError,
     AUTH_EMAIL_VERIFICATION_PREFIX, AUTH_EMAIL_VERIFIED_PREFIX, AUTH_EMAIL_VERIFIED_TTL_SECS,
-    AUTH_SMTP_TIMEOUT_SECS,
 };
-use base64::Engine;
+use crate::email_delivery::{
+    read_smtp_delivery_config, send_smtp_email, ComposedEmail, SmtpDeliveryConfig,
+};
 
 #[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
 pub(super) struct StoredAuthEmailVerificationCode {
@@ -13,25 +13,8 @@ pub(super) struct StoredAuthEmailVerificationCode {
     pub(super) created_at: String,
 }
 
-#[derive(Debug, Clone)]
-pub(super) struct AuthSmtpConfig {
-    pub(super) host: String,
-    pub(super) port: u16,
-    pub(super) user: Option<String>,
-    pub(super) password: Option<String>,
-    pub(super) use_tls: bool,
-    pub(super) use_ssl: bool,
-    pub(super) from_email: String,
-    pub(super) from_name: String,
-}
-
-#[derive(Debug, Clone)]
-pub(super) struct AuthComposedEmail {
-    pub(super) to_email: String,
-    pub(super) subject: String,
-    pub(super) html_body: String,
-    pub(super) text_body: String,
-}
+pub(super) type AuthSmtpConfig = SmtpDeliveryConfig;
+pub(super) type AuthComposedEmail = ComposedEmail;
 
 pub(super) fn auth_email_verification_key(email: &str) -> String {
     format!("{AUTH_EMAIL_VERIFICATION_PREFIX}{email}")
@@ -84,25 +67,6 @@ fn render_auth_template_string(
     Ok(rendered)
 }
 
-fn auth_encode_mime_header(value: &str) -> String {
-    if value.is_ascii() {
-        return value.to_string();
-    }
-    format!(
-        "=?UTF-8?B?{}?=",
-        base64::engine::general_purpose::STANDARD.encode(value.as_bytes())
-    )
-}
-
-fn auth_wrap_base64(value: &str) -> String {
-    let mut wrapped = String::new();
-    for chunk in value.as_bytes().chunks(76) {
-        wrapped.push_str(std::str::from_utf8(chunk).unwrap_or_default());
-        wrapped.push_str("\r\n");
-    }
-    wrapped
-}
-
 fn auth_build_verification_text_body(
     app_name: &str,
     email: &str,
@@ -114,244 +78,6 @@ fn auth_build_verification_text_body(
     )
 }
 
-fn auth_build_tls_config() -> std::sync::Arc<rustls::ClientConfig> {
-    let _ = rustls::crypto::ring::default_provider().install_default();
-    let root_store =
-        rustls::RootCertStore::from_iter(webpki_roots::TLS_SERVER_ROOTS.iter().cloned());
-    let config = rustls::ClientConfig::builder()
-        .with_root_certificates(root_store)
-        .with_no_client_auth();
-    std::sync::Arc::new(config)
-}
-
-fn auth_resolve_server_name(
-    host: &str,
-) -> Result<rustls::pki_types::ServerName<'static>, GatewayError> {
-    let host = host.trim().trim_start_matches('[').trim_end_matches(']');
-    if let Ok(ip) = host.parse::<std::net::IpAddr>() {
-        return Ok(rustls::pki_types::ServerName::from(ip));
-    }
-    rustls::pki_types::ServerName::try_from(host.to_string())
-        .map_err(|err| GatewayError::Internal(err.to_string()))
-}
-
-fn auth_connect_tcp_stream(config: &AuthSmtpConfig) -> Result<std::net::TcpStream, GatewayError> {
-    let stream = std::net::TcpStream::connect((config.host.as_str(), config.port))
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    stream
-        .set_read_timeout(Some(std::time::Duration::from_secs(AUTH_SMTP_TIMEOUT_SECS)))
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    stream
-        .set_write_timeout(Some(std::time::Duration::from_secs(AUTH_SMTP_TIMEOUT_SECS)))
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    Ok(stream)
-}
-
-fn auth_wrap_tls_stream(
-    stream: std::net::TcpStream,
-    host: &str,
-) -> Result<rustls::StreamOwned<rustls::ClientConnection, std::net::TcpStream>, GatewayError> {
-    let server_name = auth_resolve_server_name(host)?;
-    let connection = rustls::ClientConnection::new(auth_build_tls_config(), server_name)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    Ok(rustls::StreamOwned::new(connection, stream))
-}
-
-fn auth_smtp_read_response<T: std::io::BufRead>(
-    reader: &mut T,
-) -> Result<(u16, String), GatewayError> {
-    let mut message = String::new();
-    let code = loop {
-        let parsed_code;
-        let continuation;
-        let trimmed;
-        {
-            let mut line = String::new();
-            let bytes = reader
-                .read_line(&mut line)
-                .map_err(|err| GatewayError::Internal(err.to_string()))?;
-            if bytes == 0 {
-                return Err(GatewayError::Internal(
-                    "smtp connection closed unexpectedly".to_string(),
-                ));
-            }
-            trimmed = line.trim_end_matches(['\r', '\n']).to_string();
-            if trimmed.len() < 3 {
-                return Err(GatewayError::Internal("invalid smtp response".to_string()));
-            }
-            parsed_code = trimmed[..3]
-                .parse::<u16>()
-                .map_err(|err| GatewayError::Internal(err.to_string()))?;
-            continuation = trimmed.as_bytes().get(3).copied() == Some(b'-');
-        }
-        if !message.is_empty() {
-            message.push('\n');
-        }
-        message.push_str(&trimmed);
-        if !continuation {
-            break parsed_code;
-        }
-    };
-    Ok((code, message))
-}
-
-fn auth_smtp_expect<T: std::io::BufRead>(
-    reader: &mut T,
-    allowed_codes: &[u16],
-) -> Result<String, GatewayError> {
-    let (code, message) = auth_smtp_read_response(reader)?;
-    if allowed_codes.contains(&code) {
-        return Ok(message);
-    }
-    Err(GatewayError::Internal(format!(
-        "unexpected smtp response {code}: {message}"
-    )))
-}
-
-fn auth_smtp_write_line<T: std::io::Write>(writer: &mut T, line: &str) -> Result<(), GatewayError> {
-    writer
-        .write_all(line.as_bytes())
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    writer
-        .write_all(b"\r\n")
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    writer
-        .flush()
-        .map_err(|err| GatewayError::Internal(err.to_string()))
-}
-
-fn auth_smtp_send_command<S: std::io::Read + std::io::Write>(
-    reader: &mut std::io::BufReader<S>,
-    command: &str,
-    allowed_codes: &[u16],
-) -> Result<String, GatewayError> {
-    auth_smtp_write_line(reader.get_mut(), command)?;
-    auth_smtp_expect(reader, allowed_codes)
-}
-
-fn auth_build_email_message(config: &AuthSmtpConfig, email: &AuthComposedEmail) -> String {
-    let boundary = format!("aether-{}", uuid::Uuid::new_v4().simple());
-    let text_body = auth_wrap_base64(
-        &base64::engine::general_purpose::STANDARD.encode(email.text_body.as_bytes()),
-    );
-    let html_body = auth_wrap_base64(
-        &base64::engine::general_purpose::STANDARD.encode(email.html_body.as_bytes()),
-    );
-    let from_header = if config.from_name.trim().is_empty() {
-        format!("<{}>", config.from_email)
-    } else {
-        format!(
-            "{} <{}>",
-            auth_encode_mime_header(config.from_name.trim()),
-            config.from_email
-        )
-    };
-    format!(
-        "From: {from_header}\r\nTo: <{to_email}>\r\nSubject: {subject}\r\nMIME-Version: 1.0\r\nContent-Type: multipart/alternative; boundary=\"{boundary}\"\r\n\r\n--{boundary}\r\nContent-Type: text/plain; charset=\"utf-8\"\r\nContent-Transfer-Encoding: base64\r\n\r\n{text_body}--{boundary}\r\nContent-Type: text/html; charset=\"utf-8\"\r\nContent-Transfer-Encoding: base64\r\n\r\n{html_body}--{boundary}--\r\n",
-        to_email = email.to_email,
-        subject = auth_encode_mime_header(&email.subject),
-    )
-}
-
-fn auth_smtp_authenticate<S: std::io::Read + std::io::Write>(
-    reader: &mut std::io::BufReader<S>,
-    config: &AuthSmtpConfig,
-) -> Result<(), GatewayError> {
-    let Some(username) = config
-        .user
-        .as_deref()
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-    else {
-        return Ok(());
-    };
-    let password = config.password.as_deref().unwrap_or("");
-    auth_smtp_send_command(reader, "AUTH LOGIN", &[334])?;
-    auth_smtp_send_command(
-        reader,
-        &base64::engine::general_purpose::STANDARD.encode(username.as_bytes()),
-        &[334],
-    )?;
-    auth_smtp_send_command(
-        reader,
-        &base64::engine::general_purpose::STANDARD.encode(password.as_bytes()),
-        &[235],
-    )?;
-    Ok(())
-}
-
-fn auth_smtp_deliver_message<S: std::io::Read + std::io::Write>(
-    reader: &mut std::io::BufReader<S>,
-    config: &AuthSmtpConfig,
-    email: &AuthComposedEmail,
-) -> Result<(), GatewayError> {
-    auth_smtp_send_command(
-        reader,
-        &format!("MAIL FROM:<{}>", config.from_email),
-        &[250],
-    )?;
-    auth_smtp_send_command(
-        reader,
-        &format!("RCPT TO:<{}>", email.to_email),
-        &[250, 251],
-    )?;
-    auth_smtp_send_command(reader, "DATA", &[354])?;
-    let message = auth_build_email_message(config, email);
-    reader
-        .get_mut()
-        .write_all(message.as_bytes())
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    reader
-        .get_mut()
-        .write_all(b"\r\n.\r\n")
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    reader
-        .get_mut()
-        .flush()
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let _ = auth_smtp_expect(reader, &[250])?;
-    let _ = auth_smtp_send_command(reader, "QUIT", &[221]);
-    Ok(())
-}
-
-fn auth_smtp_send_message<S: std::io::Read + std::io::Write>(
-    reader: &mut std::io::BufReader<S>,
-    config: &AuthSmtpConfig,
-    email: &AuthComposedEmail,
-) -> Result<(), GatewayError> {
-    auth_smtp_send_command(reader, "EHLO aether.local", &[250])?;
-    auth_smtp_authenticate(reader, config)?;
-    auth_smtp_deliver_message(reader, config, email)
-}
-
-fn send_auth_email_blocking(
-    config: AuthSmtpConfig,
-    email: AuthComposedEmail,
-) -> Result<(), GatewayError> {
-    if config.use_ssl {
-        let stream = auth_connect_tcp_stream(&config)?;
-        let tls_stream = auth_wrap_tls_stream(stream, &config.host)?;
-        let mut reader = std::io::BufReader::new(tls_stream);
-        let _ = auth_smtp_expect(&mut reader, &[220])?;
-        return auth_smtp_send_message(&mut reader, &config, &email);
-    }
-
-    let stream = auth_connect_tcp_stream(&config)?;
-    let mut reader = std::io::BufReader::new(stream);
-    let _ = auth_smtp_expect(&mut reader, &[220])?;
-    let _ = auth_smtp_send_command(&mut reader, "EHLO aether.local", &[250])?;
-    if config.use_tls {
-        let _ = auth_smtp_send_command(&mut reader, "STARTTLS", &[220])?;
-        let stream = reader.into_inner();
-        let tls_stream = auth_wrap_tls_stream(stream, &config.host)?;
-        let mut reader = std::io::BufReader::new(tls_stream);
-        return auth_smtp_send_message(&mut reader, &config, &email);
-    }
-
-    auth_smtp_authenticate(&mut reader, &config)?;
-    auth_smtp_deliver_message(&mut reader, &config, &email)
-}
-
 pub(super) async fn read_auth_email_verification_code(
     state: &AppState,
     email: &str,
@@ -423,40 +149,7 @@ pub(super) async fn store_auth_email_verification_code(
 pub(super) async fn read_auth_smtp_config(
     state: &AppState,
 ) -> Result<Option<AuthSmtpConfig>, GatewayError> {
-    let smtp_host = state.read_system_config_json_value("smtp_host").await?;
-    let smtp_from_email = state
-        .read_system_config_json_value("smtp_from_email")
-        .await?;
-    let Some(host) = system_config_string(smtp_host.as_ref()) else {
-        return Ok(None);
-    };
-    let Some(from_email) = system_config_string(smtp_from_email.as_ref()) else {
-        return Ok(None);
-    };
-    let smtp_port = state.read_system_config_json_value("smtp_port").await?;
-    let smtp_user = state.read_system_config_json_value("smtp_user").await?;
-    let smtp_password = state.read_system_config_json_value("smtp_password").await?;
-    let smtp_use_tls = state.read_system_config_json_value("smtp_use_tls").await?;
-    let smtp_use_ssl = state.read_system_config_json_value("smtp_use_ssl").await?;
-    let smtp_from_name = state
-        .read_system_config_json_value("smtp_from_name")
-        .await?;
-
-    let password = system_config_string(smtp_password.as_ref()).map(|value| {
-        decrypt_catalog_secret_with_fallbacks(state.encryption_key(), &value).unwrap_or(value)
-    });
-
-    Ok(Some(AuthSmtpConfig {
-        host,
-        port: system_config_u16(smtp_port.as_ref(), 587),
-        user: system_config_string(smtp_user.as_ref()),
-        password,
-        use_tls: system_config_bool(smtp_use_tls.as_ref(), true),
-        use_ssl: system_config_bool(smtp_use_ssl.as_ref(), false),
-        from_email,
-        from_name: system_config_string(smtp_from_name.as_ref())
-            .unwrap_or_else(|| "Aether".to_string()),
-    }))
+    read_smtp_delivery_config(state).await
 }
 
 pub(super) async fn auth_email_app_name(state: &AppState) -> Result<String, GatewayError> {
@@ -516,27 +209,20 @@ pub(super) async fn send_auth_email(
     if record_auth_email_delivery_for_tests(
         state,
         json!({
-            "to_email": email.to_email,
-            "subject": email.subject,
-            "html_body": email.html_body,
-            "text_body": email.text_body,
+            "to_email": email.to_email.clone(),
+            "subject": email.subject.clone(),
+            "html_body": email.html_body.clone(),
+            "text_body": email.text_body.clone(),
         }),
     ) {
         return Ok(());
     }
 
-    tokio::task::spawn_blocking(move || send_auth_email_blocking(config, email))
-        .await
-        .map_err(|err| GatewayError::Internal(err.to_string()))?
+    send_smtp_email(config, email).await
 }
 
 pub(super) async fn auth_registration_email_configured(
     state: &AppState,
 ) -> Result<bool, GatewayError> {
-    let smtp_host = state.read_system_config_json_value("smtp_host").await?;
-    let smtp_from_email = state
-        .read_system_config_json_value("smtp_from_email")
-        .await?;
-    Ok(system_config_string(smtp_host.as_ref()).is_some()
-        && system_config_string(smtp_from_email.as_ref()).is_some())
+    Ok(read_smtp_delivery_config(state).await?.is_some())
 }

apps/aether-gateway/src/handlers/public/support/auth_helpers.rs

@@ -121,7 +121,6 @@ pub(super) const AUTH_REFRESH_TOKEN_EXPIRATION_DAYS: i64 = 7;
 pub(super) const AUTH_EMAIL_VERIFICATION_PREFIX: &str = "email:verification:";
 pub(super) const AUTH_EMAIL_VERIFIED_PREFIX: &str = "email:verified:";
 pub(super) const AUTH_EMAIL_VERIFIED_TTL_SECS: u64 = 3600;
-pub(super) const AUTH_SMTP_TIMEOUT_SECS: u64 = 30;
 
 pub(crate) fn build_auth_json_response(
     status: http::StatusCode,

apps/aether-gateway/src/handlers/public/support/user_me_profile.rs

@@ -7,7 +7,9 @@ use axum::{
 use serde::Deserialize;
 use serde_json::json;
 
-use crate::handlers::shared::{deserialize_optional_json_patch, normalize_feature_settings};
+use crate::handlers::shared::{
+    deserialize_optional_json_patch, normalize_user_self_feature_settings_update,
+};
 
 use super::{
     auth_password_policy_level, build_auth_error_response, resolve_authenticated_local_user,
@@ -65,12 +67,24 @@ pub(super) async fn handle_users_me_detail_put(
     let email = normalize_users_me_optional_non_empty_string(payload.email);
     let username = normalize_users_me_optional_non_empty_string(payload.username);
     let feature_settings = match payload.feature_settings {
-        Some(value) => match normalize_feature_settings(value) {
-            Ok(value) => Some(value),
-            Err(detail) => {
-                return build_auth_error_response(http::StatusCode::BAD_REQUEST, detail, false);
+        Some(value) => {
+            let current = match state.read_user_feature_settings(&auth.user.id).await {
+                Ok(value) => value,
+                Err(err) => {
+                    return build_auth_error_response(
+                        http::StatusCode::INTERNAL_SERVER_ERROR,
+                        format!("user feature settings lookup failed: {err:?}"),
+                        false,
+                    )
+                }
+            };
+            match normalize_user_self_feature_settings_update(value, current) {
+                Ok(value) => Some(value),
+                Err(detail) => {
+                    return build_auth_error_response(http::StatusCode::BAD_REQUEST, detail, false);
+                }
             }
-        },
+        }
         None => None,
     };
 

apps/aether-gateway/src/handlers/public/support/user_me_usage.rs

@@ -1,5 +1,6 @@
 use std::collections::{BTreeMap, BTreeSet};
 
+use aether_ai_serving::UPSTREAM_IS_STREAM_KEY;
 use aether_billing::{
     normalize_input_tokens_for_billing, normalize_total_input_context_for_cache_hit_rate,
 };
@@ -314,7 +315,7 @@ fn users_me_usage_upstream_is_stream(item: &StoredRequestUsageAudit) -> bool {
     item.request_metadata
         .as_ref()
         .and_then(serde_json::Value::as_object)
-        .and_then(|metadata| metadata.get("upstream_is_stream"))
+        .and_then(|metadata| metadata.get(UPSTREAM_IS_STREAM_KEY))
         .and_then(serde_json::Value::as_bool)
         .or_else(|| users_me_usage_headers_stream_flag(item.response_headers.as_ref()))
         .or_else(|| users_me_usage_infer_upstream_stream_from_captured_bodies(item))

apps/aether-gateway/src/handlers/shared/catalog.rs

@@ -21,7 +21,6 @@ const OAUTH_ACCOUNT_BLOCK_PREFIX: &str = "[ACCOUNT_BLOCK] ";
 const OAUTH_EXPIRED_PREFIX: &str = "[OAUTH_EXPIRED] ";
 const OAUTH_REFRESH_FAILED_PREFIX: &str = "[REFRESH_FAILED] ";
 const OAUTH_REQUEST_FAILED_PREFIX: &str = "[REQUEST_FAILED] ";
-const BALANCE_QUERY_SECRET_CIPHERTEXT_KEY: &str = "secret_ciphertext";
 
 pub(crate) fn provider_catalog_key_supports_format(
     key: &StoredProviderCatalogKey,
@@ -174,31 +173,6 @@ pub(crate) fn parse_catalog_auth_config_json(
         .cloned()
 }
 
-fn sanitized_admin_upstream_metadata(upstream_metadata: Option<&Value>) -> Value {
-    let Some(mut metadata) = upstream_metadata.cloned() else {
-        return Value::Null;
-    };
-    let Some(balance_query) = metadata
-        .as_object_mut()
-        .and_then(|metadata| metadata.get_mut("balance_query"))
-        .and_then(Value::as_object_mut)
-    else {
-        return metadata;
-    };
-    let has_saved_secret = balance_query
-        .remove(BALANCE_QUERY_SECRET_CIPHERTEXT_KEY)
-        .is_some();
-    if has_saved_secret {
-        let query_config = balance_query
-            .entry("query_config".to_string())
-            .or_insert_with(|| Value::Object(Map::new()));
-        if let Some(query_config) = query_config.as_object_mut() {
-            query_config.insert("has_saved_secret".to_string(), Value::Bool(true));
-        }
-    }
-    metadata
-}
-
 pub(crate) fn default_provider_key_status_snapshot() -> serde_json::Value {
     json!({
         "oauth": {
@@ -1112,6 +1086,278 @@ fn build_chatgpt_web_quota_status_snapshot(
     }))
 }
 
+fn windsurf_percent_quota_window_snapshot(
+    metadata: &Map<String, Value>,
+    code: &str,
+    label: &str,
+    remaining_percent_key: &str,
+    reset_at_key: &str,
+    observed_at_unix_secs: Option<u64>,
+) -> Option<Value> {
+    let remaining_percent = metadata
+        .get(remaining_percent_key)
+        .and_then(admin_provider_quota_pure::coerce_json_f64);
+    let reset_at = provider_quota_timestamp_unix_secs(metadata.get(reset_at_key));
+    if remaining_percent.is_none() && reset_at.is_none() {
+        return None;
+    }
+
+    let remaining_ratio = remaining_percent.map(|value| (value / 100.0).clamp(0.0, 1.0));
+    let used_ratio = remaining_ratio.map(|value| (1.0 - value).clamp(0.0, 1.0));
+    let reset_seconds = quota_window_reset_seconds(observed_at_unix_secs, reset_at);
+
+    Some(json!({
+        "code": code,
+        "label": label,
+        "scope": "account",
+        "unit": "percent",
+        "used_ratio": used_ratio,
+        "remaining_ratio": remaining_ratio,
+        "reset_at": reset_at,
+        "reset_seconds": reset_seconds,
+        "is_exhausted": remaining_ratio.map(|value| value <= 1e-6),
+    }))
+}
+
+fn windsurf_count_quota_window_snapshot(
+    metadata: &Map<String, Value>,
+    code: &str,
+    label: &str,
+    used_key: &str,
+    limit_key: &str,
+    remaining_key: &str,
+) -> Option<Value> {
+    let used = metadata
+        .get(used_key)
+        .and_then(admin_provider_quota_pure::coerce_json_f64);
+    let limit = metadata
+        .get(limit_key)
+        .and_then(admin_provider_quota_pure::coerce_json_f64);
+    let remaining = metadata
+        .get(remaining_key)
+        .and_then(admin_provider_quota_pure::coerce_json_f64)
+        .or_else(|| limit.zip(used).map(|(limit, used)| (limit - used).max(0.0)));
+    if used.is_none() && limit.is_none() && remaining.is_none() {
+        return None;
+    }
+
+    let used_ratio = used
+        .zip(limit)
+        .and_then(|(used, limit)| (limit > 0.0).then_some((used / limit).clamp(0.0, 1.0)));
+    let remaining_ratio = remaining.zip(limit).and_then(|(remaining, limit)| {
+        (limit > 0.0).then_some((remaining / limit).clamp(0.0, 1.0))
+    });
+
+    Some(json!({
+        "code": code,
+        "label": label,
+        "scope": "account",
+        "unit": "count",
+        "used_ratio": used_ratio,
+        "remaining_ratio": remaining_ratio,
+        "used_value": used,
+        "remaining_value": remaining,
+        "limit_value": limit,
+        "is_exhausted": remaining.is_some_and(|value| value <= 0.0),
+    }))
+}
+
+fn build_windsurf_quota_status_snapshot(
+    upstream_metadata: Option<&Value>,
+    source: &str,
+) -> Option<Value> {
+    let metadata = provider_quota_metadata_bucket(upstream_metadata, "windsurf")?;
+    let observed_at_unix_secs = provider_quota_timestamp_unix_secs(metadata.get("updated_at"));
+    let plan_type = metadata
+        .get("plan_name")
+        .or_else(|| metadata.get("plan_type"))
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned);
+    let rate_limit = metadata
+        .get("rate_limit")
+        .cloned()
+        .filter(|value| !value.is_null());
+    let last_error = metadata
+        .get("last_error")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned);
+    let banned = metadata
+        .get("banned")
+        .or_else(|| metadata.get("is_banned"))
+        .and_then(admin_provider_quota_pure::coerce_json_bool)
+        == Some(true);
+    let quarantined = metadata
+        .get("quarantined")
+        .or_else(|| metadata.get("is_quarantined"))
+        .and_then(admin_provider_quota_pure::coerce_json_bool)
+        == Some(true);
+
+    let mut windows = [
+        windsurf_percent_quota_window_snapshot(
+            metadata,
+            "daily",
+            "日",
+            "daily_remaining_percent",
+            "daily_reset_at",
+            observed_at_unix_secs,
+        ),
+        windsurf_percent_quota_window_snapshot(
+            metadata,
+            "weekly",
+            "周",
+            "weekly_remaining_percent",
+            "weekly_reset_at",
+            observed_at_unix_secs,
+        ),
+        windsurf_count_quota_window_snapshot(
+            metadata,
+            "prompt",
+            "Prompt",
+            "prompt_used",
+            "prompt_limit",
+            "prompt_remaining",
+        ),
+        windsurf_count_quota_window_snapshot(
+            metadata,
+            "flex",
+            "Flex",
+            "flex_used",
+            "flex_limit",
+            "flex_remaining",
+        ),
+    ]
+    .into_iter()
+    .flatten()
+    .collect::<Vec<_>>();
+
+    let mut rate_limit_cooling = false;
+    let mut rate_limit_reset_seconds = None;
+    let mut rate_limit_reason = None::<String>;
+    if let Some(rate_limit) = rate_limit.as_ref() {
+        if let Some(rate_limit_object) = rate_limit.as_object() {
+            let retry_after_ms = rate_limit_object
+                .get("retry_after_ms")
+                .or_else(|| rate_limit_object.get("retryAfterMs"))
+                .and_then(admin_provider_quota_pure::coerce_json_u64)
+                .filter(|value| *value > 0);
+            if let Some(retry_after_ms) = retry_after_ms {
+                rate_limit_cooling = true;
+                rate_limit_reset_seconds = Some(retry_after_ms.saturating_add(999) / 1000);
+                rate_limit_reason = rate_limit_object
+                    .get("message")
+                    .and_then(Value::as_str)
+                    .map(str::trim)
+                    .filter(|value| !value.is_empty())
+                    .map(ToOwned::to_owned);
+                windows.push(json!({
+                    "code": "rate_limit",
+                    "label": "速率",
+                    "scope": "account",
+                    "unit": "count",
+                    "is_exhausted": false,
+                    "reset_seconds": rate_limit_reset_seconds,
+                }));
+            }
+        }
+    }
+
+    let allowed_models_count = metadata
+        .get("allowed_models_count")
+        .or_else(|| metadata.get("models_count"))
+        .and_then(admin_provider_quota_pure::coerce_json_u64);
+
+    if windows.is_empty()
+        && plan_type.is_none()
+        && observed_at_unix_secs.is_none()
+        && rate_limit.is_none()
+        && allowed_models_count.is_none()
+        && !banned
+        && !quarantined
+    {
+        return None;
+    }
+
+    let usage_ratio = quota_windows_usage_ratio(&windows);
+    let reset_seconds = if rate_limit_cooling {
+        rate_limit_reset_seconds.or_else(|| quota_windows_min_reset_seconds(&windows))
+    } else {
+        quota_windows_min_reset_seconds(&windows)
+    };
+    let reset_at = if rate_limit_cooling {
+        None
+    } else {
+        quota_windows_min_reset_at(&windows)
+    };
+    let exhausted_by_window = windows.iter().filter_map(Value::as_object).any(|window| {
+        window
+            .get("code")
+            .and_then(Value::as_str)
+            .is_some_and(|code| {
+                code.eq_ignore_ascii_case("daily") || code.eq_ignore_ascii_case("weekly")
+            })
+            && window
+                .get("is_exhausted")
+                .and_then(admin_provider_quota_pure::coerce_json_bool)
+                .unwrap_or(false)
+    });
+    let exhausted = banned || quarantined || exhausted_by_window;
+    let (code, label, reason) = if banned {
+        (
+            "banned",
+            Some("账号已封禁"),
+            last_error
+                .clone()
+                .or_else(|| Some("账号被 Windsurf 标记为不可用".to_string())),
+        )
+    } else if quarantined {
+        (
+            "quarantined",
+            Some("账号隔离中"),
+            last_error
+                .clone()
+                .or_else(|| Some("账号处于隔离状态".to_string())),
+        )
+    } else if rate_limit_cooling {
+        (
+            "cooldown",
+            Some("冷却中"),
+            last_error.clone().or(rate_limit_reason),
+        )
+    } else if exhausted {
+        (
+            "exhausted",
+            Some("额度耗尽"),
+            Some("额度窗口已耗尽".to_string()),
+        )
+    } else {
+        ("ok", None, last_error)
+    };
+
+    Some(json!({
+        "version": 2,
+        "provider_type": "windsurf",
+        "code": code,
+        "label": label,
+        "reason": reason,
+        "freshness": "fresh",
+        "source": source,
+        "observed_at": observed_at_unix_secs,
+        "exhausted": exhausted,
+        "usage_ratio": usage_ratio,
+        "updated_at": observed_at_unix_secs,
+        "reset_at": reset_at,
+        "reset_seconds": reset_seconds,
+        "plan_type": plan_type,
+        "allowed_models_count": allowed_models_count,
+        "rate_limit": rate_limit.unwrap_or(Value::Null),
+        "windows": windows,
+    }))
+}
+
 fn build_antigravity_quota_status_snapshot(
     upstream_metadata: Option<&Value>,
     source: &str,
@@ -1359,6 +1605,7 @@ pub(crate) fn sync_provider_key_quota_status_snapshot(
         "codex" => build_codex_quota_status_snapshot(upstream_metadata, source),
         "kiro" => build_kiro_quota_status_snapshot(upstream_metadata, source),
         "chatgpt_web" => build_chatgpt_web_quota_status_snapshot(upstream_metadata, source),
+        "windsurf" => build_windsurf_quota_status_snapshot(upstream_metadata, source),
         "antigravity" => build_antigravity_quota_status_snapshot(upstream_metadata, source),
         "grok" => build_grok_quota_status_snapshot(upstream_metadata, source),
         "gemini_cli" => build_gemini_cli_quota_status_snapshot(upstream_metadata, source),
@@ -1395,6 +1642,12 @@ fn quota_snapshot_has_materialized_data(
         return false;
     }
 
+    if normalized_provider_type == "windsurf"
+        && windsurf_quota_snapshot_has_stale_cooldown(quota_snapshot)
+    {
+        return false;
+    }
+
     if quota_snapshot
         .get("windows")
         .and_then(Value::as_array)
@@ -1420,6 +1673,61 @@ fn quota_snapshot_has_materialized_data(
         })
 }
 
+fn windsurf_quota_snapshot_has_stale_cooldown(quota_snapshot: &Map<String, Value>) -> bool {
+    let code = quota_snapshot
+        .get("code")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .unwrap_or_default();
+    if !code.eq_ignore_ascii_case("cooldown") {
+        return false;
+    }
+
+    let rate_limit = quota_snapshot.get("rate_limit").and_then(Value::as_object);
+    let retry_after_ms = rate_limit
+        .and_then(|rate_limit| {
+            rate_limit
+                .get("retry_after_ms")
+                .or_else(|| rate_limit.get("retryAfterMs"))
+                .and_then(admin_provider_quota_pure::coerce_json_u64)
+        })
+        .unwrap_or(0);
+    if retry_after_ms > 0 {
+        return false;
+    }
+
+    let has_positive_rate_limit_reset = quota_snapshot
+        .get("windows")
+        .and_then(Value::as_array)
+        .is_some_and(|windows| {
+            windows.iter().filter_map(Value::as_object).any(|window| {
+                window
+                    .get("code")
+                    .and_then(Value::as_str)
+                    .is_some_and(|code| code.eq_ignore_ascii_case("rate_limit"))
+                    && window
+                        .get("reset_seconds")
+                        .or_else(|| window.get("reset_at"))
+                        .and_then(admin_provider_quota_pure::coerce_json_u64)
+                        .is_some_and(|value| value > 0)
+            })
+        });
+    if has_positive_rate_limit_reset {
+        return false;
+    }
+
+    let exhausted = quota_snapshot
+        .get("exhausted")
+        .and_then(admin_provider_quota_pure::coerce_json_bool)
+        .unwrap_or(false);
+    let has_capacity = rate_limit
+        .and_then(|rate_limit| rate_limit.get("has_capacity"))
+        .and_then(admin_provider_quota_pure::coerce_json_bool)
+        .unwrap_or(false);
+
+    has_capacity || !exhausted
+}
+
 pub(crate) fn provider_key_status_snapshot_payload(
     key: &StoredProviderCatalogKey,
     provider_type: &str,
@@ -1963,7 +2271,7 @@ pub(crate) fn build_admin_provider_key_response(
     );
     payload.insert(
         "upstream_metadata".to_string(),
-        sanitized_admin_upstream_metadata(key.upstream_metadata.as_ref()),
+        json!(key.upstream_metadata),
     );
     payload.insert("proxy".to_string(), json!(key.proxy));
     payload.insert("fingerprint".to_string(), json!(key.fingerprint));
@@ -2298,6 +2606,274 @@ mod tests {
         assert_eq!(auto.get("used_value"), Some(&json!(90.0)));
     }
 
+    #[test]
+    fn provider_key_status_snapshot_payload_backfills_windsurf_daily_and_weekly_quota() {
+        let mut key = sample_catalog_key();
+        key.upstream_metadata = Some(json!({
+            "windsurf": {
+                "updated_at": 1_778_067_246u64,
+                "plan_name": "Pro",
+                "daily_remaining_percent": 40.0,
+                "weekly_remaining_percent": 65.0,
+                "daily_reset_at": 1_778_100_000u64,
+                "weekly_reset_at": 1_778_600_000u64,
+                "prompt_used": 12.0,
+                "prompt_limit": 100.0,
+                "prompt_remaining": 88.0,
+                "flex_used": 3.0,
+                "flex_limit": 10.0,
+                "flex_remaining": 7.0,
+                "allowed_models_count": 82
+            }
+        }));
+
+        let payload = provider_key_status_snapshot_payload(&key, "windsurf");
+        let quota = payload
+            .get("quota")
+            .and_then(Value::as_object)
+            .expect("quota snapshot should be object");
+        let windows = quota
+            .get("windows")
+            .and_then(Value::as_array)
+            .expect("windsurf quota windows should exist");
+        let daily = windows
+            .iter()
+            .filter_map(Value::as_object)
+            .find(|window| window.get("code") == Some(&json!("daily")))
+            .expect("daily quota window should exist");
+        let weekly = windows
+            .iter()
+            .filter_map(Value::as_object)
+            .find(|window| window.get("code") == Some(&json!("weekly")))
+            .expect("weekly quota window should exist");
+
+        assert_eq!(quota.get("provider_type"), Some(&json!("windsurf")));
+        assert_eq!(quota.get("code"), Some(&json!("ok")));
+        assert_eq!(quota.get("plan_type"), Some(&json!("Pro")));
+        assert_eq!(quota.get("usage_ratio"), Some(&json!(0.6)));
+        assert_eq!(quota.get("reset_at"), Some(&json!(1_778_100_000u64)));
+        assert_eq!(daily.get("remaining_ratio"), Some(&json!(0.4)));
+        assert_eq!(daily.get("used_ratio"), Some(&json!(0.6)));
+        assert_eq!(daily.get("reset_seconds"), Some(&json!(32_754u64)));
+        assert_eq!(weekly.get("remaining_ratio"), Some(&json!(0.65)));
+        assert_eq!(weekly.get("used_ratio"), Some(&json!(0.35)));
+        assert_eq!(weekly.get("reset_seconds"), Some(&json!(532_754u64)));
+        assert_eq!(quota.get("allowed_models_count"), Some(&json!(82)));
+    }
+
+    #[test]
+    fn provider_key_status_snapshot_payload_treats_windsurf_rate_limit_as_cooldown() {
+        let mut key = sample_catalog_key();
+        key.upstream_metadata = Some(json!({
+            "windsurf": {
+                "updated_at": 1_778_067_246u64,
+                "daily_remaining_percent": 80.0,
+                "rate_limit": {
+                    "limited": true,
+                    "retry_after_ms": 60_001u64,
+                    "message": "slow down"
+                },
+                "last_error": "slow down"
+            }
+        }));
+
+        let payload = provider_key_status_snapshot_payload(&key, "windsurf");
+        let quota = payload
+            .get("quota")
+            .and_then(Value::as_object)
+            .expect("quota snapshot should be object");
+        let rate_window = quota
+            .get("windows")
+            .and_then(Value::as_array)
+            .and_then(|windows| {
+                windows
+                    .iter()
+                    .filter_map(Value::as_object)
+                    .find(|window| window.get("code") == Some(&json!("rate_limit")))
+            })
+            .expect("rate limit window should exist");
+
+        assert_eq!(quota.get("code"), Some(&json!("cooldown")));
+        assert_eq!(quota.get("exhausted"), Some(&json!(false)));
+        assert_eq!(quota.get("reset_seconds"), Some(&json!(61u64)));
+        assert_eq!(rate_window.get("is_exhausted"), Some(&json!(false)));
+        assert_eq!(rate_window.get("reset_seconds"), Some(&json!(61u64)));
+    }
+
+    #[test]
+    fn provider_key_status_snapshot_payload_keeps_windsurf_capacity_probe_without_retry_after_ok() {
+        let mut key = sample_catalog_key();
+        key.upstream_metadata = Some(json!({
+            "windsurf": {
+                "updated_at": 1_778_067_246u64,
+                "daily_remaining_percent": 100.0,
+                "weekly_remaining_percent": 100.0,
+                "rate_limit": {
+                    "limited": true,
+                    "has_capacity": false,
+                    "messages_remaining": 0.0,
+                    "max_messages": 100.0
+                }
+            }
+        }));
+
+        let payload = provider_key_status_snapshot_payload(&key, "windsurf");
+        let quota = payload
+            .get("quota")
+            .and_then(Value::as_object)
+            .expect("quota snapshot should be object");
+        let has_rate_limit_window =
+            quota
+                .get("windows")
+                .and_then(Value::as_array)
+                .is_some_and(|windows| {
+                    windows
+                        .iter()
+                        .filter_map(Value::as_object)
+                        .any(|window| window.get("code") == Some(&json!("rate_limit")))
+                });
+
+        assert_eq!(quota.get("code"), Some(&json!("ok")));
+        assert_eq!(quota.get("label"), Some(&Value::Null));
+        assert_eq!(quota.get("exhausted"), Some(&json!(false)));
+        assert_eq!(
+            payload.pointer("/quota/rate_limit/limited"),
+            Some(&json!(true))
+        );
+        assert!(!has_rate_limit_window);
+    }
+
+    #[test]
+    fn provider_key_status_snapshot_payload_refreshes_stale_windsurf_cooldown_when_probe_has_capacity(
+    ) {
+        let mut key = sample_catalog_key();
+        key.status_snapshot = Some(json!({
+            "quota": {
+                "version": 2,
+                "provider_type": "windsurf",
+                "code": "cooldown",
+                "label": "冷却中",
+                "exhausted": false,
+                "windows": [
+                    {
+                        "code": "daily",
+                        "unit": "percent",
+                        "label": "日",
+                        "scope": "account",
+                        "remaining_ratio": 0.99,
+                        "is_exhausted": false
+                    },
+                    {
+                        "code": "rate_limit",
+                        "unit": "count",
+                        "label": "速率",
+                        "scope": "account",
+                        "is_exhausted": false,
+                        "reset_seconds": null
+                    }
+                ],
+                "rate_limit": {
+                    "limited": true,
+                    "has_capacity": true,
+                    "messages_remaining": -1,
+                    "max_messages": -1
+                }
+            }
+        }));
+        key.upstream_metadata = Some(json!({
+            "windsurf": {
+                "updated_at": 1_778_067_246u64,
+                "daily_remaining_percent": 99.0,
+                "weekly_remaining_percent": 100.0,
+                "allowed_models_count": 118,
+                "rate_limit": {
+                    "limited": true,
+                    "has_capacity": true,
+                    "messages_remaining": -1,
+                    "max_messages": -1
+                }
+            }
+        }));
+
+        let payload = provider_key_status_snapshot_payload(&key, "windsurf");
+        let quota = payload
+            .get("quota")
+            .and_then(Value::as_object)
+            .expect("quota snapshot should be object");
+        let has_rate_limit_window =
+            quota
+                .get("windows")
+                .and_then(Value::as_array)
+                .is_some_and(|windows| {
+                    windows
+                        .iter()
+                        .filter_map(Value::as_object)
+                        .any(|window| window.get("code") == Some(&json!("rate_limit")))
+                });
+
+        assert_eq!(quota.get("code"), Some(&json!("ok")));
+        assert_eq!(quota.get("label"), Some(&Value::Null));
+        assert_eq!(quota.get("allowed_models_count"), Some(&json!(118u64)));
+        assert!(!has_rate_limit_window);
+    }
+
+    #[test]
+    fn provider_key_status_snapshot_payload_marks_windsurf_banned_and_quarantined_blocking() {
+        let mut banned_key = sample_catalog_key();
+        banned_key.upstream_metadata = Some(json!({
+            "windsurf": {
+                "updated_at": 1_778_067_246u64,
+                "banned": true,
+                "reason": "forbidden"
+            }
+        }));
+        let banned_payload = provider_key_status_snapshot_payload(&banned_key, "windsurf");
+
+        assert_eq!(
+            banned_payload.pointer("/quota/code"),
+            Some(&json!("banned"))
+        );
+        assert_eq!(
+            banned_payload.pointer("/quota/exhausted"),
+            Some(&json!(true))
+        );
+        assert_eq!(
+            banned_payload.pointer("/account/code"),
+            Some(&json!("account_banned"))
+        );
+        assert_eq!(
+            banned_payload.pointer("/account/blocked"),
+            Some(&json!(true))
+        );
+
+        let mut quarantined_key = sample_catalog_key();
+        quarantined_key.upstream_metadata = Some(json!({
+            "windsurf": {
+                "updated_at": 1_778_067_246u64,
+                "quarantined": true
+            }
+        }));
+        let quarantined_payload =
+            provider_key_status_snapshot_payload(&quarantined_key, "windsurf");
+
+        assert_eq!(
+            quarantined_payload.pointer("/quota/code"),
+            Some(&json!("quarantined"))
+        );
+        assert_eq!(
+            quarantined_payload.pointer("/quota/exhausted"),
+            Some(&json!(true))
+        );
+        assert_eq!(
+            quarantined_payload.pointer("/account/code"),
+            Some(&json!("account_quarantined"))
+        );
+        assert_eq!(
+            quarantined_payload.pointer("/account/blocked"),
+            Some(&json!(true))
+        );
+    }
+
     #[test]
     fn provider_key_status_snapshot_payload_preserves_existing_materialized_quota_snapshot() {
         let mut key = sample_catalog_key();

apps/aether-gateway/src/handlers/shared/mod.rs

@@ -38,7 +38,8 @@ pub(crate) use self::external_models::OFFICIAL_EXTERNAL_MODEL_PROVIDERS;
 pub(crate) use self::normalize::{
     deserialize_optional_json_patch, deserialize_optional_string_list_patch, ip_rules_allow,
     json_ip_rules_allow, normalize_feature_settings, normalize_ip_rules, normalize_json_array,
-    normalize_json_object, normalize_string_list, parse_json_ip_rules,
+    normalize_json_object, normalize_string_list, normalize_user_self_feature_settings_update,
+    parse_json_ip_rules,
 };
 pub(crate) use self::payloads::{
     InternalGatewayAuthContextRequest, InternalGatewayExecuteRequest,

apps/aether-gateway/src/handlers/shared/normalize.rs

@@ -54,6 +54,7 @@ pub(crate) fn normalize_feature_settings(value: Option<Value>) -> Result<Option<
         Value::Null => Ok(None),
         Value::Object(ref mut settings) => {
             normalize_chat_pii_redaction_feature_settings(settings)?;
+            normalize_notification_push_service_feature_settings(settings)?;
             if settings.is_empty() {
                 Ok(None)
             } else {
@@ -64,6 +65,41 @@ pub(crate) fn normalize_feature_settings(value: Option<Value>) -> Result<Option<
     }
 }
 
+pub(crate) fn normalize_user_self_feature_settings_update(
+    value: Option<Value>,
+    current: Option<Value>,
+) -> Result<Option<Value>, String> {
+    let mut normalized = normalize_feature_settings(value)?;
+    let current_notification_push_service = current
+        .and_then(|value| match value {
+            Value::Object(mut settings) => settings.remove("notification_push_service"),
+            _ => None,
+        })
+        .and_then(|value| {
+            let mut wrapper = Map::new();
+            wrapper.insert("notification_push_service".to_string(), value);
+            normalize_notification_push_service_feature_settings(&mut wrapper)
+                .ok()
+                .and_then(|_| wrapper.remove("notification_push_service"))
+        });
+
+    match (&mut normalized, current_notification_push_service) {
+        (Some(Value::Object(settings)), Some(value)) => {
+            settings.insert("notification_push_service".to_string(), value);
+        }
+        (Some(Value::Object(settings)), None) => {
+            settings.remove("notification_push_service");
+        }
+        (None, Some(value)) => {
+            let mut settings = Map::new();
+            settings.insert("notification_push_service".to_string(), value);
+            normalized = Some(Value::Object(settings));
+        }
+        _ => {}
+    }
+    Ok(normalized)
+}
+
 pub(crate) fn normalize_ip_rules(
     values: Option<Vec<String>>,
 ) -> Result<Option<Vec<String>>, String> {
@@ -326,9 +362,47 @@ fn normalize_chat_pii_redaction_feature_object(
     Ok(())
 }
 
+fn normalize_notification_push_service_feature_settings(
+    settings: &mut Map<String, Value>,
+) -> Result<(), String> {
+    let Some(value) = settings.get_mut("notification_push_service") else {
+        return Ok(());
+    };
+    match value {
+        Value::Null => {
+            settings.remove("notification_push_service");
+            Ok(())
+        }
+        Value::Object(feature) => {
+            normalize_notification_push_service_feature_object(feature)?;
+            if feature.is_empty() {
+                settings.remove("notification_push_service");
+            }
+            Ok(())
+        }
+        _ => Err("notification_push_service 必须是对象".to_string()),
+    }
+}
+
+fn normalize_notification_push_service_feature_object(
+    feature: &mut Map<String, Value>,
+) -> Result<(), String> {
+    for key in ["enabled"] {
+        if let Some(value) = feature.get(key) {
+            if !value.is_boolean() {
+                return Err(format!("notification_push_service.{key} 必须是布尔值"));
+            }
+        }
+    }
+    Ok(())
+}
+
 #[cfg(test)]
 mod tests {
-    use super::{ip_rules_allow, json_ip_rules_allow, normalize_ip_rules, parse_json_ip_rules};
+    use super::{
+        ip_rules_allow, json_ip_rules_allow, normalize_feature_settings, normalize_ip_rules,
+        normalize_user_self_feature_settings_update, parse_json_ip_rules,
+    };
     use serde_json::json;
     use std::net::{IpAddr, Ipv4Addr};
 
@@ -358,6 +432,41 @@ mod tests {
         );
     }
 
+    #[test]
+    fn normalize_feature_settings_accepts_notification_push_service_permission() {
+        let normalized = normalize_feature_settings(Some(json!({
+            "notification_push_service": {"enabled": true}
+        })))
+        .expect("feature settings should normalize")
+        .expect("feature settings should remain set");
+
+        assert_eq!(
+            normalized["notification_push_service"]["enabled"],
+            json!(true)
+        );
+    }
+
+    #[test]
+    fn user_self_feature_update_preserves_notification_push_permission() {
+        let normalized = normalize_user_self_feature_settings_update(
+            Some(json!({
+                "chat_pii_redaction": {"enabled": true, "inject_model_instruction": false},
+                "notification_push_service": {"enabled": false}
+            })),
+            Some(json!({
+                "notification_push_service": {"enabled": true}
+            })),
+        )
+        .expect("feature settings should normalize")
+        .expect("feature settings should remain set");
+
+        assert_eq!(
+            normalized["notification_push_service"]["enabled"],
+            json!(true)
+        );
+        assert_eq!(normalized["chat_pii_redaction"]["enabled"], json!(true));
+    }
+
     #[test]
     fn ip_rules_allow_applies_allow_rules_and_deny_overrides() {
         let rules = vec![

apps/aether-gateway/src/handlers/shared/request_utils.rs

@@ -221,7 +221,6 @@ pub(crate) fn admin_proxy_local_requires_buffered_body(
                 | (Some("endpoints_manage"), http::Method::POST, Some("create_endpoint"))
                 | (Some("endpoints_manage"), http::Method::POST, Some("batch_delete_keys"))
                 | (Some("endpoints_manage"), http::Method::POST, Some("refresh_quota"))
-                | (Some("endpoints_manage"), http::Method::POST, Some("query_key_balance"))
                 | (Some("endpoints_manage"), http::Method::PUT, Some("update_key"))
                 | (Some("endpoints_manage"), http::Method::PUT, Some("update_endpoint"))
                 | (Some("modules_manage"), http::Method::PUT, Some("set_enabled"))
@@ -255,6 +254,11 @@ pub(crate) fn admin_proxy_local_requires_buffered_body(
                 | (Some("system_manage"), http::Method::PUT, Some("config_set"))
                 | (Some("system_manage"), http::Method::PUT, Some("email_template_set"))
                 | (Some("system_manage"), http::Method::POST, Some("email_template_preview"))
+                | (
+                    Some("system_manage"),
+                    http::Method::POST,
+                    Some("important_notification_test"),
+                )
                 | (
                     Some("provider_models_manage"),
                     http::Method::POST,

apps/aether-gateway/src/important_notification.rs

@@ -0,0 +1,890 @@
+use crate::admin_api::AdminAppState;
+use crate::bark_push::{read_bark_push_config, send_bark_push, BarkPushConfig};
+use crate::email_delivery::{
+    read_smtp_delivery_config, send_smtp_email, ComposedEmail, SmtpDeliveryConfig,
+};
+use crate::handlers::shared::{system_config_bool, system_config_string};
+use crate::server_chan_push::{
+    read_server_chan_push_config, send_server_chan_push, ServerChanPushConfig,
+};
+use crate::{AppState, GatewayError};
+use axum::body::Bytes;
+use serde::{Deserialize, Serialize};
+use serde_json::{json, Value};
+use tracing::warn;
+
+pub(crate) const IMPORTANT_NOTIFICATION_ENABLED_KEY: &str = "module.important_notification.enabled";
+pub(crate) const LEGACY_NOTIFICATION_EMAIL_ENABLED_KEY: &str = "module.notification_email.enabled";
+pub(crate) const IMPORTANT_NOTIFICATION_EMAIL_ENABLED_KEY: &str =
+    "module.important_notification.email_enabled";
+pub(crate) const IMPORTANT_NOTIFICATION_EMAIL_RECIPIENTS_KEY: &str =
+    "module.important_notification.email_recipients";
+pub(crate) const IMPORTANT_NOTIFICATION_DEFAULT_CHANNEL_KEY: &str =
+    "module.important_notification.default_channel";
+pub(crate) const IMPORTANT_NOTIFICATION_ITEMS_KEY: &str = "module.important_notification.items";
+pub(crate) const PROVIDER_QUOTA_ALERT_ITEM_KEY: &str = "provider_quota_alert";
+
+#[derive(Debug, Clone)]
+pub(crate) struct ImportantNotification {
+    pub(crate) title: String,
+    pub(crate) markdown_body: String,
+    pub(crate) text_body: String,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum ImportantNotificationChannelFilter {
+    All,
+    Email,
+    ServerChan,
+    Bark,
+}
+
+#[derive(Debug, Clone)]
+struct ImportantNotificationConfig {
+    module_enabled: bool,
+    email_enabled: bool,
+    email_recipients: Vec<String>,
+    default_channel: ImportantNotificationChannelFilter,
+    items: Vec<ImportantNotificationItemConfig>,
+    server_chan: ServerChanPushConfig,
+    bark: BarkPushConfig,
+}
+
+#[derive(Debug, Clone)]
+struct ImportantNotificationItemConfig {
+    key: String,
+    name: String,
+    enabled: bool,
+    channel: Option<ImportantNotificationChannelFilter>,
+    title_template: Option<String>,
+    markdown_template: Option<String>,
+    text_template: Option<String>,
+    user_email_enabled: bool,
+}
+
+#[derive(Debug, Clone, Copy)]
+struct NotificationChannelReadiness {
+    email: bool,
+    server_chan: bool,
+    bark: bool,
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub(crate) struct ImportantNotificationChannelReport {
+    pub(crate) channel: &'static str,
+    pub(crate) success: bool,
+    pub(crate) message: String,
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub(crate) struct ImportantNotificationDeliveryReport {
+    pub(crate) success: bool,
+    pub(crate) channels: Vec<ImportantNotificationChannelReport>,
+}
+
+#[derive(Debug, Deserialize)]
+struct ImportantNotificationTestRequest {
+    #[serde(default)]
+    channel: Option<String>,
+    #[serde(default)]
+    item_key: Option<String>,
+}
+
+pub(crate) async fn important_notification_module_enabled(
+    state: &AppState,
+) -> Result<bool, GatewayError> {
+    let canonical = state
+        .read_system_config_json_value(IMPORTANT_NOTIFICATION_ENABLED_KEY)
+        .await?;
+    if canonical.is_some() {
+        return Ok(system_config_bool(canonical.as_ref(), false));
+    }
+    let legacy = state
+        .read_system_config_json_value(LEGACY_NOTIFICATION_EMAIL_ENABLED_KEY)
+        .await?;
+    Ok(system_config_bool(legacy.as_ref(), false))
+}
+
+pub(crate) async fn important_notification_configured(
+    state: &AppState,
+) -> Result<bool, GatewayError> {
+    let config = read_important_notification_config(state).await?;
+    important_notification_has_configured_channel(state, &config).await
+}
+
+pub(crate) async fn important_notification_dispatch_ready_for_item(
+    state: &AppState,
+    item_key: &str,
+) -> Result<bool, GatewayError> {
+    let config = read_important_notification_config(state).await?;
+    if !config.module_enabled {
+        return Ok(false);
+    }
+    if let Some(item) = find_notification_item(&config, item_key) {
+        if !item.enabled {
+            return Ok(false);
+        }
+    }
+    let readiness = read_notification_channel_readiness(state, &config).await?;
+    Ok(channel_filter_has_ready_channel(
+        notification_item_channel_filter(&config, item_key),
+        readiness,
+    ))
+}
+
+pub(crate) async fn send_important_notification(
+    state: &AppState,
+    notification: ImportantNotification,
+) -> Result<ImportantNotificationDeliveryReport, GatewayError> {
+    send_important_notification_with_filter(
+        state,
+        notification,
+        ImportantNotificationChannelFilter::All,
+    )
+    .await
+}
+
+pub(crate) async fn send_important_notification_for_item(
+    state: &AppState,
+    item_key: &str,
+    notification: ImportantNotification,
+    variables: &[(&str, String)],
+) -> Result<ImportantNotificationDeliveryReport, GatewayError> {
+    dispatch_important_notification(state, Some(item_key), notification, variables, None, false)
+        .await
+}
+
+pub(crate) async fn send_important_notification_with_filter(
+    state: &AppState,
+    notification: ImportantNotification,
+    channel_filter: ImportantNotificationChannelFilter,
+) -> Result<ImportantNotificationDeliveryReport, GatewayError> {
+    dispatch_important_notification(state, None, notification, &[], Some(channel_filter), false)
+        .await
+}
+
+pub(crate) async fn send_user_important_notification_email(
+    state: &AppState,
+    item_key: &str,
+    user_email: &str,
+    notification: ImportantNotification,
+    variables: &[(&str, String)],
+) -> Result<ImportantNotificationDeliveryReport, GatewayError> {
+    let config = read_important_notification_config(state).await?;
+    if !config.module_enabled {
+        return Ok(single_report("module", false, "通知服务未启用"));
+    }
+    let Some(item) = find_notification_item(&config, item_key) else {
+        return Ok(single_report("item", false, "通知项未定义"));
+    };
+    if !item.enabled {
+        return Ok(single_report("item", false, "通知项未启用"));
+    }
+    if !item.user_email_enabled {
+        return Ok(single_report("user_email", false, "通知项未启用用户邮件"));
+    }
+    let notification = apply_notification_item_template(Some(item), notification, variables);
+    let smtp_config = match read_smtp_delivery_config(state).await? {
+        Some(config) => config,
+        None => return Ok(single_report("user_email", false, "SMTP 配置不完整")),
+    };
+    let user_email = user_email.trim();
+    if user_email.is_empty() {
+        return Ok(single_report("user_email", false, "用户邮箱为空"));
+    }
+    match send_single_email_notification(smtp_config, user_email, &notification).await {
+        Ok(()) => Ok(single_report("user_email", true, "用户邮件通知已发送")),
+        Err(err) => {
+            warn!(error = ?err, user_email = %user_email, "failed to send user notification email");
+            Ok(single_report(
+                "user_email",
+                false,
+                format!("用户邮件通知发送失败: {err:?}"),
+            ))
+        }
+    }
+}
+
+async fn important_notification_has_configured_channel(
+    state: &AppState,
+    config: &ImportantNotificationConfig,
+) -> Result<bool, GatewayError> {
+    let readiness = read_notification_channel_readiness(state, config).await?;
+    if channel_filter_has_ready_channel(config.default_channel, readiness) {
+        return Ok(true);
+    }
+    Ok(config.items.iter().any(|item| {
+        item.enabled
+            && channel_filter_has_ready_channel(
+                item.channel.unwrap_or(config.default_channel),
+                readiness,
+            )
+    }))
+}
+
+async fn read_notification_channel_readiness(
+    state: &AppState,
+    config: &ImportantNotificationConfig,
+) -> Result<NotificationChannelReadiness, GatewayError> {
+    let smtp_config = read_smtp_delivery_config(state).await?;
+    Ok(NotificationChannelReadiness {
+        email: config.email_enabled && !config.email_recipients.is_empty() && smtp_config.is_some(),
+        server_chan: config.server_chan.enabled && config.server_chan.send_key.is_some(),
+        bark: config.bark.enabled && config.bark.device_key.is_some(),
+    })
+}
+
+fn channel_filter_has_ready_channel(
+    filter: ImportantNotificationChannelFilter,
+    readiness: NotificationChannelReadiness,
+) -> bool {
+    match filter {
+        ImportantNotificationChannelFilter::All => {
+            readiness.email || readiness.server_chan || readiness.bark
+        }
+        ImportantNotificationChannelFilter::Email => readiness.email,
+        ImportantNotificationChannelFilter::ServerChan => readiness.server_chan,
+        ImportantNotificationChannelFilter::Bark => readiness.bark,
+    }
+}
+
+async fn dispatch_important_notification(
+    state: &AppState,
+    item_key: Option<&str>,
+    notification: ImportantNotification,
+    variables: &[(&str, String)],
+    channel_override: Option<ImportantNotificationChannelFilter>,
+    bypass_enable_checks: bool,
+) -> Result<ImportantNotificationDeliveryReport, GatewayError> {
+    let config = read_important_notification_config(state).await?;
+    if !bypass_enable_checks && !config.module_enabled {
+        return Ok(single_report("module", false, "通知服务未启用"));
+    }
+
+    let item = item_key.and_then(|key| find_notification_item(&config, key));
+    if !bypass_enable_checks && item.is_some_and(|item| !item.enabled) {
+        return Ok(single_report("item", false, "通知项未启用"));
+    }
+    let notification = apply_notification_item_template(item, notification, variables);
+    let channel_filter = channel_override.unwrap_or_else(|| {
+        item.and_then(|item| item.channel)
+            .unwrap_or(config.default_channel)
+    });
+
+    let mut reports = Vec::new();
+    if matches!(
+        channel_filter,
+        ImportantNotificationChannelFilter::All | ImportantNotificationChannelFilter::Email
+    ) {
+        maybe_send_email_notification(
+            state,
+            &config,
+            &notification,
+            bypass_enable_checks,
+            &mut reports,
+        )
+        .await;
+    }
+
+    if matches!(
+        channel_filter,
+        ImportantNotificationChannelFilter::All | ImportantNotificationChannelFilter::ServerChan
+    ) {
+        maybe_send_server_chan_notification(
+            state,
+            &config,
+            &notification,
+            bypass_enable_checks,
+            &mut reports,
+        )
+        .await;
+    }
+
+    if matches!(
+        channel_filter,
+        ImportantNotificationChannelFilter::All | ImportantNotificationChannelFilter::Bark
+    ) {
+        maybe_send_bark_notification(
+            state,
+            &config,
+            &notification,
+            bypass_enable_checks,
+            &mut reports,
+        )
+        .await;
+    }
+
+    if reports.is_empty() {
+        reports.push(ImportantNotificationChannelReport {
+            channel: "none",
+            success: false,
+            message: "未启用可用的通知通道".to_string(),
+        });
+    }
+    let success = reports.iter().any(|report| report.success);
+    Ok(ImportantNotificationDeliveryReport {
+        success,
+        channels: reports,
+    })
+}
+
+pub(crate) async fn build_important_notification_test_payload(
+    state: &AdminAppState<'_>,
+    request_body: Option<&Bytes>,
+) -> Result<Value, GatewayError> {
+    let request = match request_body.filter(|body| !body.is_empty()) {
+        Some(body) => serde_json::from_slice::<ImportantNotificationTestRequest>(body).unwrap_or(
+            ImportantNotificationTestRequest {
+                channel: None,
+                item_key: None,
+            },
+        ),
+        None => ImportantNotificationTestRequest {
+            channel: None,
+            item_key: None,
+        },
+    };
+    let filter = request.channel.as_deref().and_then(parse_channel_filter);
+    let item_key = request
+        .item_key
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty());
+    let variables = vec![
+        ("provider_name", "示例 Provider".to_string()),
+        ("provider_id", "provider-demo".to_string()),
+        ("total_available", "8.0000".to_string()),
+        ("threshold_amount", "10.0000".to_string()),
+        ("user_email", "user@example.com".to_string()),
+        ("balance", "1.0000".to_string()),
+    ];
+    let report = dispatch_important_notification(
+        state.app(),
+        item_key,
+        ImportantNotification {
+            title: "Aether 通知服务测试".to_string(),
+            markdown_body: "这是一条来自 Aether 的通知服务测试。".to_string(),
+            text_body: "这是一条来自 Aether 的通知服务测试。".to_string(),
+        },
+        &variables,
+        filter,
+        true,
+    )
+    .await?;
+
+    Ok(json!({
+        "success": report.success,
+        "message": if report.success { "测试通知已发送" } else { "测试通知发送失败" },
+        "channels": report.channels,
+    }))
+}
+
+async fn read_important_notification_config(
+    state: &AppState,
+) -> Result<ImportantNotificationConfig, GatewayError> {
+    let module_enabled = important_notification_module_enabled(state).await?;
+    let email_enabled = state
+        .read_system_config_json_value(IMPORTANT_NOTIFICATION_EMAIL_ENABLED_KEY)
+        .await?;
+    let email_recipients = state
+        .read_system_config_json_value(IMPORTANT_NOTIFICATION_EMAIL_RECIPIENTS_KEY)
+        .await?;
+    let default_channel = state
+        .read_system_config_json_value(IMPORTANT_NOTIFICATION_DEFAULT_CHANNEL_KEY)
+        .await?;
+    let items = state
+        .read_system_config_json_value(IMPORTANT_NOTIFICATION_ITEMS_KEY)
+        .await?;
+
+    Ok(ImportantNotificationConfig {
+        module_enabled,
+        email_enabled: system_config_bool(email_enabled.as_ref(), false),
+        email_recipients: parse_recipient_list(email_recipients.as_ref()),
+        default_channel: default_channel
+            .as_ref()
+            .and_then(|value| value.as_str())
+            .and_then(parse_channel_filter)
+            .unwrap_or(ImportantNotificationChannelFilter::All),
+        items: parse_notification_items(items.as_ref()),
+        server_chan: read_server_chan_push_config(state).await?,
+        bark: read_bark_push_config(state).await?,
+    })
+}
+
+fn parse_channel_filter(raw: &str) -> Option<ImportantNotificationChannelFilter> {
+    match raw.trim().to_ascii_lowercase().as_str() {
+        "all" => Some(ImportantNotificationChannelFilter::All),
+        "email" => Some(ImportantNotificationChannelFilter::Email),
+        "server_chan" | "serverchan" | "serve_chan" => {
+            Some(ImportantNotificationChannelFilter::ServerChan)
+        }
+        "bark" => Some(ImportantNotificationChannelFilter::Bark),
+        "global" | "" => None,
+        _ => None,
+    }
+}
+
+fn parse_notification_items(value: Option<&Value>) -> Vec<ImportantNotificationItemConfig> {
+    let Some(Value::Array(items)) = value else {
+        return default_notification_items();
+    };
+    items
+        .iter()
+        .filter_map(parse_notification_item)
+        .collect::<Vec<_>>()
+}
+
+fn parse_notification_item(value: &Value) -> Option<ImportantNotificationItemConfig> {
+    let item = value.as_object()?;
+    let key = item.get("key")?.as_str()?.trim();
+    if key.is_empty() {
+        return None;
+    }
+    let name = item
+        .get("name")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .unwrap_or(key);
+    Some(ImportantNotificationItemConfig {
+        key: key.to_string(),
+        name: name.to_string(),
+        enabled: item.get("enabled").and_then(Value::as_bool).unwrap_or(true),
+        channel: item
+            .get("channel")
+            .and_then(Value::as_str)
+            .and_then(parse_channel_filter),
+        title_template: optional_non_empty_string(item.get("title_template")),
+        markdown_template: optional_non_empty_string(item.get("markdown_template")),
+        text_template: optional_non_empty_string(item.get("text_template")),
+        user_email_enabled: item
+            .get("user_email_enabled")
+            .and_then(Value::as_bool)
+            .unwrap_or(false),
+    })
+}
+
+fn default_notification_items() -> Vec<ImportantNotificationItemConfig> {
+    vec![
+        ImportantNotificationItemConfig {
+            key: PROVIDER_QUOTA_ALERT_ITEM_KEY.to_string(),
+            name: "号池额度不足".to_string(),
+            enabled: true,
+            channel: None,
+            title_template: None,
+            markdown_template: None,
+            text_template: None,
+            user_email_enabled: false,
+        },
+        ImportantNotificationItemConfig {
+            key: "provider_pool_abnormal".to_string(),
+            name: "号池异常".to_string(),
+            enabled: true,
+            channel: None,
+            title_template: Some("号池异常：{provider_name}".to_string()),
+            markdown_template: Some(
+                "号池 `{provider_name}` 出现异常，请检查服务状态。".to_string(),
+            ),
+            text_template: Some("号池 {provider_name} 出现异常，请检查服务状态。".to_string()),
+            user_email_enabled: false,
+        },
+        ImportantNotificationItemConfig {
+            key: "user_balance_low".to_string(),
+            name: "用户余额不足".to_string(),
+            enabled: true,
+            channel: Some(ImportantNotificationChannelFilter::Email),
+            title_template: Some("余额不足提醒".to_string()),
+            markdown_template: Some("你的账户余额已低于提醒阈值，请及时处理。".to_string()),
+            text_template: Some("你的账户余额已低于提醒阈值，请及时处理。".to_string()),
+            user_email_enabled: true,
+        },
+    ]
+}
+
+fn optional_non_empty_string(value: Option<&Value>) -> Option<String> {
+    value
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned)
+}
+
+fn find_notification_item<'a>(
+    config: &'a ImportantNotificationConfig,
+    item_key: &str,
+) -> Option<&'a ImportantNotificationItemConfig> {
+    let item_key = item_key.trim();
+    config.items.iter().find(|item| item.key == item_key)
+}
+
+fn notification_item_channel_filter(
+    config: &ImportantNotificationConfig,
+    item_key: &str,
+) -> ImportantNotificationChannelFilter {
+    find_notification_item(config, item_key)
+        .and_then(|item| item.channel)
+        .unwrap_or(config.default_channel)
+}
+
+fn apply_notification_item_template(
+    item: Option<&ImportantNotificationItemConfig>,
+    notification: ImportantNotification,
+    variables: &[(&str, String)],
+) -> ImportantNotification {
+    let Some(item) = item else {
+        return notification;
+    };
+    let title = render_template(
+        item.title_template.as_deref(),
+        &notification.title,
+        &notification,
+        variables,
+    );
+    let markdown_body = render_template(
+        item.markdown_template.as_deref(),
+        &notification.markdown_body,
+        &notification,
+        variables,
+    );
+    let text_body = render_template(
+        item.text_template.as_deref(),
+        &notification.text_body,
+        &notification,
+        variables,
+    );
+    ImportantNotification {
+        title,
+        markdown_body,
+        text_body,
+    }
+}
+
+fn render_template(
+    template: Option<&str>,
+    fallback: &str,
+    notification: &ImportantNotification,
+    variables: &[(&str, String)],
+) -> String {
+    let mut rendered = template
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .unwrap_or(fallback)
+        .to_string();
+    rendered = rendered
+        .replace("{title}", &notification.title)
+        .replace("{body}", &notification.markdown_body)
+        .replace("{text_body}", &notification.text_body);
+    for (key, value) in variables {
+        rendered = rendered.replace(&format!("{{{}}}", key.trim()), value);
+    }
+    rendered
+}
+
+async fn maybe_send_email_notification(
+    state: &AppState,
+    config: &ImportantNotificationConfig,
+    notification: &ImportantNotification,
+    bypass_channel_toggle: bool,
+    reports: &mut Vec<ImportantNotificationChannelReport>,
+) {
+    if !bypass_channel_toggle && !config.email_enabled {
+        return;
+    }
+    if config.email_recipients.is_empty() {
+        reports.push(ImportantNotificationChannelReport {
+            channel: "email",
+            success: false,
+            message: "未配置邮件收件人".to_string(),
+        });
+        return;
+    }
+    let smtp_config = match read_smtp_delivery_config(state).await {
+        Ok(Some(config)) => config,
+        Ok(None) => {
+            reports.push(ImportantNotificationChannelReport {
+                channel: "email",
+                success: false,
+                message: "SMTP 配置不完整".to_string(),
+            });
+            return;
+        }
+        Err(err) => {
+            reports.push(ImportantNotificationChannelReport {
+                channel: "email",
+                success: false,
+                message: format!("读取 SMTP 配置失败: {err:?}"),
+            });
+            return;
+        }
+    };
+
+    let mut sent = 0usize;
+    let mut failed = 0usize;
+    for recipient in &config.email_recipients {
+        match send_single_email_notification(smtp_config.clone(), recipient, notification).await {
+            Ok(()) => sent += 1,
+            Err(err) => {
+                failed += 1;
+                warn!(
+                    error = ?err,
+                    recipient = %recipient,
+                    "failed to send important notification email"
+                );
+            }
+        }
+    }
+
+    reports.push(ImportantNotificationChannelReport {
+        channel: "email",
+        success: sent > 0,
+        message: if failed == 0 {
+            format!("邮件通知已发送给 {sent} 个收件人")
+        } else {
+            format!("邮件通知成功 {sent} 个，失败 {failed} 个")
+        },
+    });
+}
+
+async fn send_single_email_notification(
+    smtp_config: SmtpDeliveryConfig,
+    recipient: &str,
+    notification: &ImportantNotification,
+) -> Result<(), GatewayError> {
+    send_smtp_email(
+        smtp_config,
+        ComposedEmail {
+            to_email: recipient.to_string(),
+            subject: notification.title.clone(),
+            html_body: build_notification_html(notification),
+            text_body: notification.text_body.clone(),
+        },
+    )
+    .await
+}
+
+async fn maybe_send_server_chan_notification(
+    state: &AppState,
+    config: &ImportantNotificationConfig,
+    notification: &ImportantNotification,
+    bypass_channel_toggle: bool,
+    reports: &mut Vec<ImportantNotificationChannelReport>,
+) {
+    if !bypass_channel_toggle && !config.server_chan.enabled {
+        return;
+    }
+    if config.server_chan.send_key.is_none() {
+        reports.push(ImportantNotificationChannelReport {
+            channel: "server_chan",
+            success: false,
+            message: "未配置 Server 酱 SendKey".to_string(),
+        });
+        return;
+    };
+    match send_server_chan_push(
+        state,
+        &config.server_chan,
+        &notification.title,
+        &notification.markdown_body,
+    )
+    .await
+    {
+        Ok(()) => reports.push(ImportantNotificationChannelReport {
+            channel: "server_chan",
+            success: true,
+            message: "Server 酱通知已发送".to_string(),
+        }),
+        Err(err) => {
+            warn!(error = ?err, "failed to send server chan important notification");
+            reports.push(ImportantNotificationChannelReport {
+                channel: "server_chan",
+                success: false,
+                message: format!("Server 酱通知发送失败: {err:?}"),
+            });
+        }
+    }
+}
+
+async fn maybe_send_bark_notification(
+    state: &AppState,
+    config: &ImportantNotificationConfig,
+    notification: &ImportantNotification,
+    bypass_channel_toggle: bool,
+    reports: &mut Vec<ImportantNotificationChannelReport>,
+) {
+    if !bypass_channel_toggle && !config.bark.enabled {
+        return;
+    }
+    if config.bark.device_key.is_none() {
+        reports.push(ImportantNotificationChannelReport {
+            channel: "bark",
+            success: false,
+            message: "未配置 Bark Device Key".to_string(),
+        });
+        return;
+    };
+    match send_bark_push(
+        state,
+        &config.bark,
+        &notification.title,
+        &notification.markdown_body,
+    )
+    .await
+    {
+        Ok(()) => reports.push(ImportantNotificationChannelReport {
+            channel: "bark",
+            success: true,
+            message: "Bark 通知已发送".to_string(),
+        }),
+        Err(err) => {
+            warn!(error = ?err, "failed to send bark important notification");
+            reports.push(ImportantNotificationChannelReport {
+                channel: "bark",
+                success: false,
+                message: format!("Bark 通知发送失败: {err:?}"),
+            });
+        }
+    }
+}
+
+fn single_report(
+    channel: &'static str,
+    success: bool,
+    message: impl Into<String>,
+) -> ImportantNotificationDeliveryReport {
+    ImportantNotificationDeliveryReport {
+        success,
+        channels: vec![ImportantNotificationChannelReport {
+            channel,
+            success,
+            message: message.into(),
+        }],
+    }
+}
+
+fn parse_recipient_list(value: Option<&Value>) -> Vec<String> {
+    let mut recipients = Vec::new();
+    match value {
+        Some(Value::Array(items)) => {
+            for item in items {
+                if let Some(raw) = item.as_str() {
+                    push_recipient_parts(&mut recipients, raw);
+                }
+            }
+        }
+        Some(Value::String(raw)) => push_recipient_parts(&mut recipients, raw),
+        _ => {}
+    }
+    recipients.sort();
+    recipients.dedup();
+    recipients
+}
+
+fn push_recipient_parts(recipients: &mut Vec<String>, raw: &str) {
+    for item in raw
+        .split([',', ';', '\n', '\r'])
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+    {
+        recipients.push(item.to_string());
+    }
+}
+
+fn build_notification_html(notification: &ImportantNotification) -> String {
+    format!(
+        "<!doctype html><html><body><h2>{}</h2><pre style=\"font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;white-space:pre-wrap;line-height:1.6\">{}</pre></body></html>",
+        escape_html(&notification.title),
+        escape_html(&notification.text_body),
+    )
+}
+
+fn escape_html(value: &str) -> String {
+    value
+        .replace('&', "&amp;")
+        .replace('<', "&lt;")
+        .replace('>', "&gt;")
+        .replace('"', "&quot;")
+        .replace('\'', "&#39;")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{
+        apply_notification_item_template, parse_channel_filter, parse_notification_items,
+        parse_recipient_list, ImportantNotification, ImportantNotificationChannelFilter,
+    };
+    use serde_json::json;
+
+    #[test]
+    fn parse_recipient_list_accepts_arrays_and_delimiters() {
+        assert_eq!(
+            parse_recipient_list(Some(&json!([
+                "ops@example.com, admin@example.com",
+                "ops@example.com"
+            ]))),
+            vec![
+                "admin@example.com".to_string(),
+                "ops@example.com".to_string()
+            ]
+        );
+    }
+
+    #[test]
+    fn parse_notification_items_reads_channel_and_user_email_flag() {
+        let items = parse_notification_items(Some(&json!([
+            {
+                "key": "user_balance_low",
+                "name": "用户余额不足",
+                "enabled": true,
+                "channel": "email",
+                "title_template": "余额提醒",
+                "markdown_template": "{user_email}: {balance}",
+                "user_email_enabled": true
+            }
+        ])));
+
+        assert_eq!(items.len(), 1);
+        assert_eq!(items[0].key, "user_balance_low");
+        assert_eq!(
+            items[0].channel,
+            Some(ImportantNotificationChannelFilter::Email)
+        );
+        assert!(items[0].user_email_enabled);
+    }
+
+    #[test]
+    fn parse_channel_filter_accepts_bark() {
+        assert_eq!(
+            parse_channel_filter("bark"),
+            Some(ImportantNotificationChannelFilter::Bark)
+        );
+    }
+
+    #[test]
+    fn item_template_renders_fallback_and_variables() {
+        let items = parse_notification_items(Some(&json!([
+            {
+                "key": "provider_quota_alert",
+                "name": "号池额度不足",
+                "title_template": "额度提醒：{provider_name}",
+                "markdown_template": "{body}\n剩余：{total_available}",
+                "text_template": "{text_body}\n剩余：{total_available}"
+            }
+        ])));
+        let rendered = apply_notification_item_template(
+            Some(&items[0]),
+            ImportantNotification {
+                title: "默认标题".to_string(),
+                markdown_body: "默认正文".to_string(),
+                text_body: "默认文本".to_string(),
+            },
+            &[
+                ("provider_name", "示例 Provider".to_string()),
+                ("total_available", "8.0000".to_string()),
+            ],
+        );
+
+        assert_eq!(rendered.title, "额度提醒：示例 Provider");
+        assert_eq!(rendered.markdown_body, "默认正文\n剩余：8.0000");
+        assert_eq!(rendered.text_body, "默认文本\n剩余：8.0000");
+    }
+}

apps/aether-gateway/src/lib.rs

@@ -30,6 +30,7 @@ mod api;
 mod async_task;
 mod audit;
 mod auth;
+mod bark_push;
 mod cache;
 mod client_session_affinity;
 mod clock;
@@ -37,6 +38,7 @@ mod constants;
 mod control;
 mod data;
 mod dispatch;
+mod email_delivery;
 mod error;
 mod execution_runtime;
 mod executor;
@@ -46,6 +48,7 @@ mod handlers;
 mod headers;
 mod hooks;
 mod image_capabilities;
+mod important_notification;
 mod log_ids;
 mod maintenance;
 pub(crate) mod middleware;
@@ -62,6 +65,7 @@ mod roles;
 mod router;
 mod routing;
 mod scheduler;
+mod server_chan_push;
 mod state;
 mod system_features;
 mod task_runtime;

apps/aether-gateway/src/main.rs

@@ -254,7 +254,7 @@ struct GatewayDataArgs {
     #[arg(
         long,
         env = "AETHER_GATEWAY_DATA_POSTGRES_MIN_CONNECTIONS",
-        default_value_t = 1
+        default_value_t = 4
     )]
     postgres_min_connections: u32,
 
@@ -497,6 +497,7 @@ impl GatewayUsageArgs {
     fn to_config(&self) -> UsageRuntimeConfig {
         UsageRuntimeConfig {
             enabled: true,
+            queue_terminal_events: true,
             stream_key: self.queue_stream_key.trim().to_string(),
             consumer_group: self.queue_group.trim().to_string(),
             dlq_stream_key: self.queue_dlq_stream_key.trim().to_string(),
@@ -1670,7 +1671,7 @@ mod tests {
                 encryption_key: None,
                 redis_url: None,
                 redis_key_prefix: None,
-                postgres_min_connections: 1,
+                postgres_min_connections: 4,
                 postgres_max_connections: 20,
                 postgres_acquire_timeout_ms: 10_000,
                 postgres_idle_timeout_ms: 30_000,

apps/aether-gateway/src/maintenance/mod.rs

@@ -7,8 +7,8 @@ pub(crate) use runtime::{
     ensure_provider_key_pool_scores_for_keys, inspect_proxy_upgrade_rollout,
     list_admin_cleanup_run_records, perform_account_self_check_once,
     perform_oauth_token_refresh_once, perform_pool_quota_probe_once, perform_provider_checkin_once,
-    pool_quota_probe_target_count, preview_manual_usage_cleanup, rebuild_admin_stats_once,
-    record_completed_cleanup_run, record_proxy_upgrade_traffic_success,
+    perform_provider_quota_alert_once, pool_quota_probe_target_count, preview_manual_usage_cleanup,
+    rebuild_admin_stats_once, record_completed_cleanup_run, record_proxy_upgrade_traffic_success,
     restore_proxy_upgrade_rollout_skipped_nodes, retry_proxy_upgrade_rollout_node,
     run_admin_system_cleanup_once, run_manual_usage_cleanup_once, skip_proxy_upgrade_rollout_node,
     spawn_account_self_check_worker, spawn_audit_cleanup_worker, spawn_db_maintenance_worker,
@@ -16,16 +16,17 @@ pub(crate) use runtime::{
     spawn_pending_cleanup_worker, spawn_pool_monitor_worker,
     spawn_pool_quota_probe_replenish_for_request, spawn_pool_quota_probe_worker,
     spawn_pool_score_rebuild_worker, spawn_provider_checkin_worker,
-    spawn_proxy_node_metrics_cleanup_worker, spawn_proxy_node_stale_cleanup_worker,
-    spawn_proxy_upgrade_rollout_worker, spawn_request_candidate_cleanup_worker,
-    spawn_stats_aggregation_worker, spawn_stats_hourly_aggregation_worker,
-    spawn_usage_cleanup_worker, spawn_usage_counter_flush_worker,
-    spawn_wallet_daily_usage_aggregation_worker, start_admin_request_body_cleanup_task,
-    start_admin_system_purge_task, start_manual_usage_cleanup_task, start_proxy_upgrade_rollout,
-    AccountSelfCheckRunSummary, AdminCleanupRunRecord, AdminCleanupTaskKind,
-    AdminStatsRebuildSummary, AdminSystemCleanupSummary, ManualUsageCleanupError,
-    ManualUsageCleanupMode, ManualUsageCleanupOptions, OAuthTokenRefreshRunSummary,
-    PoolQuotaProbeRunSummary, PoolQuotaProbeWorkerConfig, ProviderCheckinRunSummary,
+    spawn_provider_quota_alert_worker, spawn_proxy_node_metrics_cleanup_worker,
+    spawn_proxy_node_stale_cleanup_worker, spawn_proxy_upgrade_rollout_worker,
+    spawn_request_candidate_cleanup_worker, spawn_stats_aggregation_worker,
+    spawn_stats_hourly_aggregation_worker, spawn_usage_cleanup_worker,
+    spawn_usage_counter_flush_worker, spawn_wallet_daily_usage_aggregation_worker,
+    start_admin_request_body_cleanup_task, start_admin_system_purge_task,
+    start_manual_usage_cleanup_task, start_proxy_upgrade_rollout, AccountSelfCheckRunSummary,
+    AdminCleanupRunRecord, AdminCleanupTaskKind, AdminStatsRebuildSummary,
+    AdminSystemCleanupSummary, ManualUsageCleanupError, ManualUsageCleanupMode,
+    ManualUsageCleanupOptions, OAuthTokenRefreshRunSummary, PoolQuotaProbeRunSummary,
+    PoolQuotaProbeWorkerConfig, ProviderCheckinRunSummary, ProviderQuotaAlertRunSummary,
     ProxyUpgradeRolloutCancelSummary, ProxyUpgradeRolloutConflictClearSummary,
     ProxyUpgradeRolloutNodeActionSummary, ProxyUpgradeRolloutProbeConfig,
     ProxyUpgradeRolloutSkippedRestoreSummary, ProxyUpgradeRolloutStatus,

apps/aether-gateway/src/maintenance/runtime.rs

@@ -26,6 +26,8 @@ mod pool_quota_probe;
 mod pool_score_rebuild;
 #[path = "runtime/provider_checkin.rs"]
 mod provider_checkin;
+#[path = "runtime/provider_quota_alert.rs"]
+mod provider_quota_alert;
 #[path = "runtime/proxy_node_metrics_cleanup.rs"]
 mod proxy_node_metrics_cleanup;
 #[path = "runtime/proxy_node_staleness.rs"]
@@ -85,6 +87,9 @@ pub(crate) use pool_score_rebuild::{
     PoolScoreRebuildRunSummary, PoolScoreRebuildWorkerConfig,
 };
 pub(crate) use provider_checkin::{perform_provider_checkin_once, ProviderCheckinRunSummary};
+pub(crate) use provider_quota_alert::{
+    perform_provider_quota_alert_once, ProviderQuotaAlertRunSummary,
+};
 use proxy_node_metrics_cleanup::*;
 use proxy_node_staleness::*;
 use proxy_upgrade_rollout::*;
@@ -138,6 +143,8 @@ const USAGE_COUNTER_DELTA_CLEANUP_INTERVAL: Duration = Duration::from_secs(60);
 const USAGE_COUNTER_DELTA_CLEANUP_BATCH_SIZE: usize = 5_000;
 const USAGE_COUNTER_DELTA_RETENTION_SECS: u64 = 7 * 24 * 60 * 60;
 const PROVIDER_CHECKIN_CONCURRENCY: usize = 3;
+const PROVIDER_QUOTA_ALERT_CONCURRENCY: usize = 3;
+const PROVIDER_QUOTA_ALERT_INTERVAL: Duration = Duration::from_secs(5);
 const PROVIDER_CHECKIN_DEFAULT_TIME: &str = "01:05";
 const REQUEST_CANDIDATE_CLEANUP_INTERVAL: Duration = Duration::from_secs(24 * 60 * 60);
 const STATS_DAILY_AGGREGATION_HOUR: u32 = 0;

apps/aether-gateway/src/maintenance/runtime/account_self_check.rs

@@ -738,8 +738,21 @@ pub(crate) fn spawn_account_self_check_worker(
         let mut interval = tokio::time::interval(config.scan_interval);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if state
+                .data
+                .should_defer_maintenance_for_database_pool_pressure(&mut deferred_since)
+            {
+                debug!(
+                    event_name = "maintenance_worker_deferred",
+                    log_type = "ops",
+                    worker = "account_self_check",
+                    "gateway account self-check deferred because database pool has no idle reserve"
+                );
+                continue;
+            }
             if let Err(err) = perform_account_self_check_once_with_config(&state, config).await {
                 warn!(
                     error = ?err,

apps/aether-gateway/src/maintenance/runtime/pool_score_rebuild.rs

@@ -393,8 +393,21 @@ pub(crate) fn spawn_pool_score_rebuild_worker(
         }
         let mut interval = tokio::time::interval(config.interval);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if state
+                .data
+                .should_defer_maintenance_for_database_pool_pressure(&mut deferred_since)
+            {
+                debug!(
+                    event_name = "maintenance_worker_deferred",
+                    log_type = "ops",
+                    worker = "pool_score_rebuild",
+                    "gateway pool score rebuild deferred because database pool has no idle reserve"
+                );
+                continue;
+            }
             match perform_pool_score_rebuild_once_with_config(&state, config).await {
                 Ok(summary) if summary.scores_upserted > 0 => {
                     info!(

apps/aether-gateway/src/maintenance/runtime/provider_quota_alert.rs

@@ -0,0 +1,486 @@
+use std::collections::HashMap;
+
+use aether_data_contracts::repository::provider_catalog::{
+    StoredProviderCatalogEndpoint, StoredProviderCatalogProvider,
+};
+use futures_util::stream::{self, StreamExt};
+use serde::{Deserialize, Serialize};
+use serde_json::Value;
+use tracing::warn;
+
+use crate::admin_api::{
+    admin_provider_ops_local_action_response, store_admin_provider_ops_balance_cache, AdminAppState,
+};
+use crate::important_notification::{
+    important_notification_dispatch_ready_for_item, send_important_notification_for_item,
+    ImportantNotification, PROVIDER_QUOTA_ALERT_ITEM_KEY,
+};
+use crate::{AppState, GatewayError};
+
+use super::PROVIDER_QUOTA_ALERT_CONCURRENCY;
+
+const PROVIDER_QUOTA_ALERT_STATE_PREFIX: &str = "provider_ops:quota_alert:";
+const PROVIDER_QUOTA_ALERT_DEFAULT_FETCH_INTERVAL_SECS: u64 = 30;
+const PROVIDER_QUOTA_ALERT_MIN_FETCH_INTERVAL_SECS: u64 = 30;
+const PROVIDER_QUOTA_ALERT_REPEAT_COOLDOWN_SECS: u64 = 24 * 60 * 60;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) struct ProviderQuotaAlertRunSummary {
+    pub(crate) checked: usize,
+    pub(crate) alerted: usize,
+    pub(crate) skipped: usize,
+    pub(crate) failed: usize,
+}
+
+#[derive(Debug, Clone)]
+struct ProviderQuotaAlertTarget {
+    provider: StoredProviderCatalogProvider,
+    config: ProviderQuotaAlertConfig,
+}
+
+#[derive(Debug, Clone, Copy)]
+struct ProviderQuotaAlertConfig {
+    enabled: bool,
+    threshold_amount: f64,
+    fetch_interval_seconds: u64,
+}
+
+#[derive(Debug, Clone, Default, Serialize, Deserialize)]
+struct ProviderQuotaAlertRuntimeState {
+    #[serde(default)]
+    last_checked_at: u64,
+    #[serde(default)]
+    last_available: Option<f64>,
+    #[serde(default)]
+    below_threshold: bool,
+    #[serde(default)]
+    last_notified_at: Option<u64>,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum ProviderQuotaAlertStatus {
+    Checked,
+    Alerted,
+    Skipped,
+    Failed,
+}
+
+pub(crate) async fn perform_provider_quota_alert_once(
+    state: &AppState,
+) -> Result<ProviderQuotaAlertRunSummary, GatewayError> {
+    if !state.has_provider_catalog_data_reader() {
+        return Ok(ProviderQuotaAlertRunSummary {
+            checked: 0,
+            alerted: 0,
+            skipped: 0,
+            failed: 0,
+        });
+    }
+    if !important_notification_dispatch_ready_for_item(state, PROVIDER_QUOTA_ALERT_ITEM_KEY).await?
+    {
+        return Ok(ProviderQuotaAlertRunSummary {
+            checked: 0,
+            alerted: 0,
+            skipped: 0,
+            failed: 0,
+        });
+    }
+
+    let now_unix_secs = now_unix_secs();
+    let targets = select_provider_quota_alert_targets(state, now_unix_secs).await?;
+    if targets.is_empty() {
+        return Ok(ProviderQuotaAlertRunSummary {
+            checked: 0,
+            alerted: 0,
+            skipped: 0,
+            failed: 0,
+        });
+    }
+
+    let provider_ids = targets
+        .iter()
+        .map(|target| target.provider.id.clone())
+        .collect::<Vec<_>>();
+    let mut endpoints_by_provider = HashMap::<String, Vec<StoredProviderCatalogEndpoint>>::new();
+    for endpoint in state
+        .list_provider_catalog_endpoints_by_provider_ids(&provider_ids)
+        .await?
+    {
+        endpoints_by_provider
+            .entry(endpoint.provider_id.clone())
+            .or_default()
+            .push(endpoint);
+    }
+
+    let mut results = stream::iter(targets.into_iter().map(|target| {
+        let state = state.clone();
+        let provider_id = target.provider.id.clone();
+        let endpoints = endpoints_by_provider
+            .get(&provider_id)
+            .cloned()
+            .unwrap_or_default();
+        async move { run_provider_quota_alert_for_provider(&state, target, endpoints).await }
+    }))
+    .buffer_unordered(PROVIDER_QUOTA_ALERT_CONCURRENCY);
+
+    let mut summary = ProviderQuotaAlertRunSummary {
+        checked: 0,
+        alerted: 0,
+        skipped: 0,
+        failed: 0,
+    };
+    while let Some(status) = results.next().await {
+        match status {
+            ProviderQuotaAlertStatus::Checked => summary.checked += 1,
+            ProviderQuotaAlertStatus::Alerted => {
+                summary.checked += 1;
+                summary.alerted += 1;
+            }
+            ProviderQuotaAlertStatus::Skipped => summary.skipped += 1,
+            ProviderQuotaAlertStatus::Failed => summary.failed += 1,
+        }
+    }
+
+    Ok(summary)
+}
+
+async fn select_provider_quota_alert_targets(
+    state: &AppState,
+    now_unix_secs: u64,
+) -> Result<Vec<ProviderQuotaAlertTarget>, GatewayError> {
+    let providers = state
+        .list_provider_catalog_providers(true)
+        .await?
+        .into_iter()
+        .filter_map(|provider| {
+            let config = provider_quota_alert_config(&provider)?;
+            (config.enabled).then_some(ProviderQuotaAlertTarget { provider, config })
+        })
+        .collect::<Vec<_>>();
+    let mut due = Vec::new();
+    for target in providers {
+        let runtime = read_quota_alert_runtime_state(state, &target.provider.id).await;
+        let last_checked_at = runtime
+            .as_ref()
+            .map(|state| state.last_checked_at)
+            .unwrap_or(0);
+        if now_unix_secs.saturating_sub(last_checked_at)
+            >= target
+                .config
+                .fetch_interval_seconds
+                .max(PROVIDER_QUOTA_ALERT_MIN_FETCH_INTERVAL_SECS)
+        {
+            due.push(target);
+        }
+    }
+    Ok(due)
+}
+
+async fn run_provider_quota_alert_for_provider(
+    state: &AppState,
+    target: ProviderQuotaAlertTarget,
+    endpoints: Vec<StoredProviderCatalogEndpoint>,
+) -> ProviderQuotaAlertStatus {
+    let provider_id = target.provider.id.clone();
+    let admin_state = AdminAppState::new(state);
+    let payload = admin_provider_ops_local_action_response(
+        &admin_state,
+        &provider_id,
+        Some(&target.provider),
+        &endpoints,
+        "query_balance",
+        None,
+    )
+    .await;
+    store_admin_provider_ops_balance_cache(&admin_state, &provider_id, &payload).await;
+
+    let now_unix_secs = now_unix_secs();
+    let payload_status_success = payload.get("status").and_then(Value::as_str) == Some("success");
+    let Some(total_available) = extract_total_available(&payload) else {
+        warn!(
+            provider_id = %provider_id,
+            payload = %payload,
+            "provider quota alert skipped because balance payload has no total_available"
+        );
+        write_checked_runtime_state_without_balance(state, &provider_id, now_unix_secs).await;
+        return if payload_status_success {
+            ProviderQuotaAlertStatus::Skipped
+        } else {
+            ProviderQuotaAlertStatus::Failed
+        };
+    };
+
+    let previous = read_quota_alert_runtime_state(state, &provider_id).await;
+    let should_notify = provider_quota_alert_should_notify(
+        now_unix_secs,
+        total_available,
+        target.config.threshold_amount,
+        previous.as_ref(),
+    );
+    let mut next = ProviderQuotaAlertRuntimeState {
+        last_checked_at: now_unix_secs,
+        last_available: Some(total_available),
+        below_threshold: total_available <= target.config.threshold_amount,
+        last_notified_at: previous.and_then(|state| state.last_notified_at),
+    };
+
+    if should_notify {
+        let notification = build_provider_quota_alert_notification(
+            &target.provider,
+            total_available,
+            target.config.threshold_amount,
+        );
+        let variables = provider_quota_alert_notification_variables(
+            &target.provider,
+            total_available,
+            target.config.threshold_amount,
+        );
+        let report = send_important_notification_for_item(
+            state,
+            PROVIDER_QUOTA_ALERT_ITEM_KEY,
+            notification,
+            &variables,
+        )
+        .await;
+        let delivered = match &report {
+            Ok(report) if report.success => true,
+            Ok(report) => {
+                warn!(
+                    provider_id = %provider_id,
+                    report = ?report,
+                    "provider quota alert notification did not reach any channel"
+                );
+                false
+            }
+            Err(err) => {
+                warn!(
+                    provider_id = %provider_id,
+                    error = ?err,
+                    "provider quota alert notification failed"
+                );
+                false
+            }
+        };
+        if delivered {
+            next.last_notified_at = Some(now_unix_secs);
+            write_quota_alert_runtime_state(state, &provider_id, &next).await;
+            return ProviderQuotaAlertStatus::Alerted;
+        }
+        write_quota_alert_runtime_state(state, &provider_id, &next).await;
+        return ProviderQuotaAlertStatus::Failed;
+    }
+
+    write_quota_alert_runtime_state(state, &provider_id, &next).await;
+    ProviderQuotaAlertStatus::Checked
+}
+
+fn provider_quota_alert_config(
+    provider: &StoredProviderCatalogProvider,
+) -> Option<ProviderQuotaAlertConfig> {
+    let quota_alert = provider
+        .config
+        .as_ref()
+        .and_then(Value::as_object)
+        .and_then(|config| config.get("provider_ops"))
+        .and_then(Value::as_object)
+        .and_then(|provider_ops| provider_ops.get("quota_alert"))
+        .and_then(Value::as_object)?;
+    let enabled = quota_alert
+        .get("enabled")
+        .and_then(Value::as_bool)
+        .unwrap_or(false);
+    let threshold_amount = quota_alert
+        .get("threshold_amount")
+        .and_then(value_as_f64)
+        .filter(|value| value.is_finite() && *value >= 0.0)
+        .unwrap_or(0.0);
+    let fetch_interval_seconds = quota_alert
+        .get("fetch_interval_seconds")
+        .and_then(Value::as_u64)
+        .unwrap_or(PROVIDER_QUOTA_ALERT_DEFAULT_FETCH_INTERVAL_SECS)
+        .max(PROVIDER_QUOTA_ALERT_MIN_FETCH_INTERVAL_SECS);
+    Some(ProviderQuotaAlertConfig {
+        enabled,
+        threshold_amount,
+        fetch_interval_seconds,
+    })
+}
+
+fn provider_quota_alert_should_notify(
+    now_unix_secs: u64,
+    total_available: f64,
+    threshold_amount: f64,
+    previous: Option<&ProviderQuotaAlertRuntimeState>,
+) -> bool {
+    if total_available > threshold_amount {
+        return false;
+    }
+    let Some(previous) = previous else {
+        return true;
+    };
+    if !previous.below_threshold {
+        return true;
+    }
+    previous
+        .last_notified_at
+        .map(|last| now_unix_secs.saturating_sub(last) >= PROVIDER_QUOTA_ALERT_REPEAT_COOLDOWN_SECS)
+        .unwrap_or(true)
+}
+
+fn extract_total_available(payload: &Value) -> Option<f64> {
+    if payload.get("status").and_then(Value::as_str) != Some("success") {
+        return None;
+    }
+    payload
+        .get("data")
+        .and_then(|data| data.get("total_available"))
+        .and_then(value_as_f64)
+        .filter(|value| value.is_finite())
+}
+
+fn value_as_f64(value: &Value) -> Option<f64> {
+    value.as_f64().or_else(|| {
+        value
+            .as_str()
+            .and_then(|raw| raw.trim().parse::<f64>().ok())
+    })
+}
+
+fn build_provider_quota_alert_notification(
+    provider: &StoredProviderCatalogProvider,
+    total_available: f64,
+    threshold_amount: f64,
+) -> ImportantNotification {
+    let title = format!("提供商额度提醒：{}", provider.name);
+    let body = format!(
+        "提供商 `{}` 当前剩余额度为 `{:.4}`，已低于或等于提醒阈值 `{:.4}`。\n\nProvider ID: `{}`",
+        provider.name, total_available, threshold_amount, provider.id
+    );
+    let text_body = format!(
+        "提供商 {} 当前剩余额度为 {:.4}，已低于或等于提醒阈值 {:.4}。\n\nProvider ID: {}",
+        provider.name, total_available, threshold_amount, provider.id
+    );
+    ImportantNotification {
+        title,
+        markdown_body: body,
+        text_body,
+    }
+}
+
+fn provider_quota_alert_notification_variables(
+    provider: &StoredProviderCatalogProvider,
+    total_available: f64,
+    threshold_amount: f64,
+) -> Vec<(&'static str, String)> {
+    vec![
+        ("provider_name", provider.name.clone()),
+        ("provider_id", provider.id.clone()),
+        ("total_available", format!("{total_available:.4}")),
+        ("threshold_amount", format!("{threshold_amount:.4}")),
+    ]
+}
+
+async fn read_quota_alert_runtime_state(
+    state: &AppState,
+    provider_id: &str,
+) -> Option<ProviderQuotaAlertRuntimeState> {
+    let key = provider_quota_alert_state_key(provider_id);
+    state
+        .runtime_kv_get(&key)
+        .await
+        .ok()
+        .flatten()
+        .and_then(|raw| serde_json::from_str::<ProviderQuotaAlertRuntimeState>(&raw).ok())
+}
+
+async fn write_checked_runtime_state_without_balance(
+    state: &AppState,
+    provider_id: &str,
+    now_unix_secs: u64,
+) {
+    let mut next = read_quota_alert_runtime_state(state, provider_id)
+        .await
+        .unwrap_or_default();
+    next.last_checked_at = now_unix_secs;
+    write_quota_alert_runtime_state(state, provider_id, &next).await;
+}
+
+async fn write_quota_alert_runtime_state(
+    state: &AppState,
+    provider_id: &str,
+    runtime_state: &ProviderQuotaAlertRuntimeState,
+) {
+    let Ok(serialized) = serde_json::to_string(runtime_state) else {
+        return;
+    };
+    if let Err(err) = state
+        .runtime_state()
+        .kv_set(
+            &provider_quota_alert_state_key(provider_id),
+            serialized,
+            None,
+        )
+        .await
+    {
+        warn!(
+            error = %err,
+            provider_id,
+            "failed to write provider quota alert runtime state"
+        );
+    }
+}
+
+fn provider_quota_alert_state_key(provider_id: &str) -> String {
+    format!("{PROVIDER_QUOTA_ALERT_STATE_PREFIX}{provider_id}")
+}
+
+fn now_unix_secs() -> u64 {
+    chrono::Utc::now().timestamp().max(0) as u64
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{
+        provider_quota_alert_should_notify, ProviderQuotaAlertRuntimeState,
+        PROVIDER_QUOTA_ALERT_REPEAT_COOLDOWN_SECS,
+    };
+
+    #[test]
+    fn quota_alert_notifies_on_first_drop_and_after_cooldown() {
+        assert!(provider_quota_alert_should_notify(100, 3.0, 5.0, None));
+        assert!(provider_quota_alert_should_notify(
+            100,
+            3.0,
+            5.0,
+            Some(&ProviderQuotaAlertRuntimeState {
+                below_threshold: false,
+                ..ProviderQuotaAlertRuntimeState::default()
+            })
+        ));
+        assert!(!provider_quota_alert_should_notify(
+            100,
+            3.0,
+            5.0,
+            Some(&ProviderQuotaAlertRuntimeState {
+                below_threshold: true,
+                last_notified_at: Some(90),
+                ..ProviderQuotaAlertRuntimeState::default()
+            })
+        ));
+        assert!(provider_quota_alert_should_notify(
+            100 + PROVIDER_QUOTA_ALERT_REPEAT_COOLDOWN_SECS,
+            3.0,
+            5.0,
+            Some(&ProviderQuotaAlertRuntimeState {
+                below_threshold: true,
+                last_notified_at: Some(100),
+                ..ProviderQuotaAlertRuntimeState::default()
+            })
+        ));
+    }
+
+    #[test]
+    fn quota_alert_does_not_notify_above_threshold() {
+        assert!(!provider_quota_alert_should_notify(100, 6.0, 5.0, None));
+    }
+}

apps/aether-gateway/src/maintenance/runtime/workers.rs

@@ -1,7 +1,8 @@
 use std::sync::Arc;
+use std::time::Instant;
 
 use chrono::Utc;
-use tracing::warn;
+use tracing::{debug, warn};
 
 use crate::data::GatewayDataState;
 use crate::AppState;
@@ -10,22 +11,22 @@ use super::{
     cleanup_processed_usage_counter_deltas_once, duration_until_next_daily_run,
     duration_until_next_db_maintenance_run, duration_until_next_stats_aggregation_run,
     duration_until_next_stats_hourly_aggregation_run, maintenance_timezone, parse_hhmm_time,
-    perform_oauth_token_refresh_once, provider_checkin_schedule, run_audit_cleanup_once,
-    run_db_maintenance_once, run_gemini_file_mapping_cleanup_once, run_pending_cleanup_once,
-    run_pool_monitor_once, run_provider_checkin_once, run_proxy_node_metrics_cleanup_once,
-    run_proxy_node_stale_cleanup_once, run_proxy_upgrade_rollout_once,
-    run_request_candidate_cleanup_once, run_stats_aggregation_once,
+    perform_oauth_token_refresh_once, perform_provider_quota_alert_once, provider_checkin_schedule,
+    run_audit_cleanup_once, run_db_maintenance_once, run_gemini_file_mapping_cleanup_once,
+    run_pending_cleanup_once, run_pool_monitor_once, run_provider_checkin_once,
+    run_proxy_node_metrics_cleanup_once, run_proxy_node_stale_cleanup_once,
+    run_proxy_upgrade_rollout_once, run_request_candidate_cleanup_once, run_stats_aggregation_once,
     run_stats_hourly_aggregation_once, run_usage_cleanup_once, run_usage_counter_flush_once,
     run_wallet_daily_usage_aggregation_once, AUDIT_LOG_CLEANUP_INTERVAL,
     GEMINI_FILE_MAPPING_CLEANUP_INTERVAL, OAUTH_TOKEN_REFRESH_INTERVAL, PENDING_CLEANUP_INTERVAL,
-    POOL_MONITOR_INTERVAL, PROVIDER_CHECKIN_DEFAULT_TIME, PROXY_NODE_METRICS_CLEANUP_HOUR,
-    PROXY_NODE_METRICS_CLEANUP_MINUTE, PROXY_NODE_STALE_SWEEP_INTERVAL,
-    PROXY_UPGRADE_ROLLOUT_INTERVAL, REQUEST_CANDIDATE_CLEANUP_INTERVAL, USAGE_CLEANUP_HOUR,
-    USAGE_CLEANUP_MINUTE, USAGE_COUNTER_DELTA_CLEANUP_BATCH_SIZE,
-    USAGE_COUNTER_DELTA_CLEANUP_INTERVAL, USAGE_COUNTER_DELTA_RETENTION_SECS,
-    USAGE_COUNTER_FLUSH_BATCH_SIZE, USAGE_COUNTER_FLUSH_CATCH_UP_BURST_LIMIT,
-    USAGE_COUNTER_FLUSH_INTERVAL, WALLET_DAILY_USAGE_AGGREGATION_HOUR,
-    WALLET_DAILY_USAGE_AGGREGATION_MINUTE,
+    POOL_MONITOR_INTERVAL, PROVIDER_CHECKIN_DEFAULT_TIME, PROVIDER_QUOTA_ALERT_INTERVAL,
+    PROXY_NODE_METRICS_CLEANUP_HOUR, PROXY_NODE_METRICS_CLEANUP_MINUTE,
+    PROXY_NODE_STALE_SWEEP_INTERVAL, PROXY_UPGRADE_ROLLOUT_INTERVAL,
+    REQUEST_CANDIDATE_CLEANUP_INTERVAL, USAGE_CLEANUP_HOUR, USAGE_CLEANUP_MINUTE,
+    USAGE_COUNTER_DELTA_CLEANUP_BATCH_SIZE, USAGE_COUNTER_DELTA_CLEANUP_INTERVAL,
+    USAGE_COUNTER_DELTA_RETENTION_SECS, USAGE_COUNTER_FLUSH_BATCH_SIZE,
+    USAGE_COUNTER_FLUSH_CATCH_UP_BURST_LIMIT, USAGE_COUNTER_FLUSH_INTERVAL,
+    WALLET_DAILY_USAGE_AGGREGATION_HOUR, WALLET_DAILY_USAGE_AGGREGATION_MINUTE,
 };
 
 const STATS_DAILY_CATCH_UP_BURST_LIMIT: usize = 14;
@@ -46,6 +47,37 @@ fn log_maintenance_worker_failure(
     );
 }
 
+fn should_defer_for_database_pressure(
+    data: &GatewayDataState,
+    worker: &'static str,
+    deferred_since: &mut Option<Instant>,
+) -> bool {
+    let Some(summary) = data.database_pool_summary() else {
+        *deferred_since = None;
+        return false;
+    };
+    if !GatewayDataState::should_defer_maintenance_for_pool_pressure_state(
+        GatewayDataState::database_pool_summary_under_maintenance_pressure(&summary),
+        deferred_since,
+    ) {
+        return false;
+    }
+
+    debug!(
+        event_name = "maintenance_worker_deferred",
+        log_type = "ops",
+        worker,
+        driver = %summary.driver,
+        checked_out = summary.checked_out,
+        pool_size = summary.pool_size,
+        idle = summary.idle,
+        max_connections = summary.max_connections,
+        usage_rate = summary.usage_rate,
+        "gateway maintenance worker deferred because database pool has no idle reserve"
+    );
+    true
+}
+
 pub(crate) fn spawn_audit_cleanup_worker(
     data: Arc<GatewayDataState>,
 ) -> Option<tokio::task::JoinHandle<()>> {
@@ -177,8 +209,19 @@ pub(crate) fn spawn_usage_counter_flush_worker(
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
         let mut last_delta_cleanup = tokio::time::Instant::now();
+        let mut usage_counter_flush_deferred_since = None;
+        let mut usage_counter_delta_cleanup_deferred_since = None;
 
         loop {
+            if should_defer_for_database_pressure(
+                &data,
+                "usage_counter_flush",
+                &mut usage_counter_flush_deferred_since,
+            ) {
+                interval.tick().await;
+                continue;
+            }
+
             let mut batches = 0_usize;
             while batches < USAGE_COUNTER_FLUSH_CATCH_UP_BURST_LIMIT {
                 match run_usage_counter_flush_once(&data, USAGE_COUNTER_FLUSH_BATCH_SIZE).await {
@@ -197,7 +240,18 @@ pub(crate) fn spawn_usage_counter_flush_worker(
             }
 
             if last_delta_cleanup.elapsed() >= USAGE_COUNTER_DELTA_CLEANUP_INTERVAL {
-                if let Err(err) = cleanup_processed_usage_counter_deltas_once(
+                if should_defer_for_database_pressure(
+                    &data,
+                    "usage_counter_delta_cleanup",
+                    &mut usage_counter_delta_cleanup_deferred_since,
+                ) {
+                    debug!(
+                        event_name = "maintenance_worker_deferred",
+                        log_type = "ops",
+                        worker = "usage_counter_delta_cleanup",
+                        "gateway maintenance worker deferred cleanup under database pressure"
+                    );
+                } else if let Err(err) = cleanup_processed_usage_counter_deltas_once(
                     &data,
                     USAGE_COUNTER_DELTA_RETENTION_SECS,
                     USAGE_COUNTER_DELTA_CLEANUP_BATCH_SIZE,
@@ -254,6 +308,34 @@ pub(crate) fn spawn_provider_checkin_worker(
     }))
 }
 
+pub(crate) fn spawn_provider_quota_alert_worker(
+    state: AppState,
+) -> Option<tokio::task::JoinHandle<()>> {
+    if !state.has_provider_catalog_data_reader() {
+        return None;
+    }
+
+    Some(tokio::spawn(async move {
+        let mut interval = tokio::time::interval(PROVIDER_QUOTA_ALERT_INTERVAL);
+        interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
+        interval.tick().await;
+        let mut deferred_since = None;
+        loop {
+            interval.tick().await;
+            if should_defer_for_database_pressure(
+                &state.data,
+                "provider_quota_alert",
+                &mut deferred_since,
+            ) {
+                continue;
+            }
+            if let Err(err) = perform_provider_quota_alert_once(&state).await {
+                log_maintenance_worker_failure("provider_quota_alert", "tick", &err);
+            }
+        }
+    }))
+}
+
 pub(crate) fn spawn_oauth_token_refresh_worker(
     state: AppState,
 ) -> Option<tokio::task::JoinHandle<()>> {
@@ -268,8 +350,16 @@ pub(crate) fn spawn_oauth_token_refresh_worker(
         let mut interval = tokio::time::interval(OAUTH_TOKEN_REFRESH_INTERVAL);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if should_defer_for_database_pressure(
+                &state.data,
+                "oauth_token_refresh",
+                &mut deferred_since,
+            ) {
+                continue;
+            }
             if let Err(err) = perform_oauth_token_refresh_once(&state).await {
                 log_maintenance_worker_failure("oauth_token_refresh", "tick", &err);
             }
@@ -314,8 +404,12 @@ pub(crate) fn spawn_pending_cleanup_worker(
         let mut interval = tokio::time::interval(PENDING_CLEANUP_INTERVAL);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if should_defer_for_database_pressure(&data, "pending_cleanup", &mut deferred_since) {
+                continue;
+            }
             if let Err(err) = run_pending_cleanup_once(&data).await {
                 log_maintenance_worker_failure("pending_cleanup", "tick", &err);
             }
@@ -337,8 +431,16 @@ pub(crate) fn spawn_proxy_node_stale_cleanup_worker(
         let mut interval = tokio::time::interval(PROXY_NODE_STALE_SWEEP_INTERVAL);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if should_defer_for_database_pressure(
+                &data,
+                "proxy_node_stale_cleanup",
+                &mut deferred_since,
+            ) {
+                continue;
+            }
             if let Err(err) = run_proxy_node_stale_cleanup_once(&data).await {
                 log_maintenance_worker_failure("proxy_node_stale_cleanup", "tick", &err);
             }
@@ -387,8 +489,16 @@ pub(crate) fn spawn_proxy_upgrade_rollout_worker(
         let mut interval = tokio::time::interval(PROXY_UPGRADE_ROLLOUT_INTERVAL);
         interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
         interval.tick().await;
+        let mut deferred_since = None;
         loop {
             interval.tick().await;
+            if should_defer_for_database_pressure(
+                &state.data,
+                "proxy_upgrade_rollout",
+                &mut deferred_since,
+            ) {
+                continue;
+            }
             if let Err(err) = run_proxy_upgrade_rollout_once(&state).await {
                 log_maintenance_worker_failure("proxy_upgrade_rollout", "tick", &err);
             }

apps/aether-gateway/src/model_fetch/runtime.rs

@@ -426,6 +426,7 @@ mod tests {
         keys: Arc<Mutex<Vec<StoredProviderCatalogKey>>>,
         transports: Arc<HashMap<(String, String, String), GatewayProviderTransportSnapshot>>,
         execution_results: Arc<Mutex<VecDeque<ExecutionResult>>>,
+        executed_plans: Arc<Mutex<Vec<ExecutionPlan>>>,
         cached_models: Arc<Mutex<HashMap<(String, String), Vec<Value>>>>,
     }
 
@@ -443,6 +444,7 @@ mod tests {
                 keys: Arc::new(Mutex::new(keys)),
                 transports: Arc::new(transports),
                 execution_results: Arc::new(Mutex::new(VecDeque::from(execution_results))),
+                executed_plans: Arc::new(Mutex::new(Vec::new())),
                 cached_models: Arc::new(Mutex::new(HashMap::new())),
             }
         }
@@ -482,8 +484,12 @@ mod tests {
 
         async fn execute_model_fetch_execution_plan(
             &self,
-            _plan: &ExecutionPlan,
+            plan: &ExecutionPlan,
         ) -> Result<ExecutionResult, String> {
+            self.executed_plans
+                .lock()
+                .expect("executed plans mutex")
+                .push(plan.clone());
             self.execution_results
                 .lock()
                 .expect("execution result mutex")
@@ -910,6 +916,134 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn model_fetch_fetches_windsurf_model_configs_and_persists_allowed_models() {
+        let provider = sample_provider("provider-windsurf", "windsurf");
+        let endpoint = StoredProviderCatalogEndpoint::new(
+            "endpoint-windsurf-chat".to_string(),
+            "provider-windsurf".to_string(),
+            "openai:chat".to_string(),
+            None,
+            None,
+            true,
+        )
+        .expect("endpoint should build")
+        .with_transport_fields(
+            "https://server.codeium.com".to_string(),
+            None,
+            None,
+            None,
+            None,
+            None,
+            None,
+            None,
+        )
+        .expect("endpoint transport should build");
+        let key = sample_key(
+            "key-windsurf",
+            "provider-windsurf",
+            "api_key",
+            &["openai:chat"],
+        );
+        let mut transport = sample_transport(
+            "windsurf",
+            "provider-windsurf",
+            "endpoint-windsurf-chat",
+            "key-windsurf",
+            "openai:chat",
+            "api_key",
+            Some(r#"{"provider_type":"windsurf"}"#),
+        );
+        transport.endpoint.base_url = "https://server.codeium.com".to_string();
+        transport.key.decrypted_api_key = "devin-session-token$abc".to_string();
+        let state = TestState::new(
+            vec![provider],
+            vec![endpoint],
+            vec![key],
+            HashMap::from([(
+                (
+                    "provider-windsurf".to_string(),
+                    "endpoint-windsurf-chat".to_string(),
+                    "key-windsurf".to_string(),
+                ),
+                transport,
+            )]),
+            vec![execution_result(json!({
+                "clientModelConfigs": [
+                    {
+                        "modelUid": "claude-sonnet-4-6",
+                        "label": "Claude Sonnet 4.6",
+                        "provider": "anthropic",
+                        "supportsImages": true,
+                        "creditMultiplier": 4
+                    },
+                    {
+                        "modelUid": "gpt-5.4",
+                        "label": "GPT-5.4",
+                        "provider": "openai"
+                    }
+                ],
+                "defaultOverrideModelConfig": {
+                    "modelUid": "claude-sonnet-4-6"
+                }
+            }))],
+        );
+
+        let summary = perform_model_fetch_once_with_state(&state)
+            .await
+            .expect("fetch should succeed");
+
+        assert_eq!(summary.succeeded, 1);
+        let plans = state.executed_plans.lock().expect("executed plans mutex");
+        assert_eq!(plans.len(), 1);
+        assert_eq!(
+            plans[0].url,
+            "https://server.codeium.com/exa.api_server_pb.ApiServerService/GetCascadeModelConfigs"
+        );
+        assert_eq!(plans[0].method, "POST");
+        assert_eq!(plans[0].provider_api_format, "windsurf:model_configs");
+        assert_eq!(
+            plans[0]
+                .body
+                .json_body
+                .as_ref()
+                .and_then(|body| body.get("metadata"))
+                .and_then(|metadata| metadata.get("apiKey")),
+            Some(&json!("devin-session-token$abc"))
+        );
+        drop(plans);
+
+        let updated = state.key("key-windsurf");
+        assert_eq!(
+            updated.allowed_models,
+            Some(json!(["claude-sonnet-4-6", "gpt-5.4"]))
+        );
+        assert_eq!(
+            updated
+                .upstream_metadata
+                .as_ref()
+                .and_then(|value| value.get("windsurf"))
+                .and_then(|value| value.get("allowed_models_count")),
+            Some(&json!(2))
+        );
+        assert_eq!(
+            updated
+                .upstream_metadata
+                .as_ref()
+                .and_then(|value| value.get("windsurf"))
+                .and_then(|value| value.get("default_model_uid")),
+            Some(&json!("claude-sonnet-4-6"))
+        );
+        let cached = state.cached_models.lock().expect("cache mutex");
+        let cached_models = cached
+            .get(&("provider-windsurf".to_string(), "key-windsurf".to_string()))
+            .expect("cached models should be written");
+        assert_eq!(
+            cached_models[0]["api_formats"],
+            json!(["openai:chat", "openai:responses", "claude:messages"])
+        );
+    }
+
     #[tokio::test]
     async fn model_fetch_failure_keeps_existing_allowed_models() {
         let provider = sample_provider("provider-openai", "openai");