refactor: 大规模模块拆分与代码精简，新增 ai-pipeline/data-contracts 独立 crate

- 新增 aether-ai-pipeline 和 aether-data-contracts crate，将 pipeline 逻辑与数据契约从 gateway 中解耦
- 重构 admin handlers：拆分单体模块为 auth/billing/endpoint/features/model/observability/provider/system 等独立子模块
- 合并 chat/cli 重复代码路径：精简 conversion、finalize、planner 中的 sync/chat/cli 分支
- 重构 scheduler/executor/data 层，引入 facade 模式降低模块间耦合
- 移除冗余的 intent 模块，将 plan_fallback/policy/stream_path/sync_path 迁移至 executor
- 前端适配：调整 admin API 调用和 provider 模型测试对话框

Cargo.lock

@@ -19,11 +19,22 @@ dependencies = [
  "cpufeatures",
 ]
 
+[[package]]
+name = "aether-ai-pipeline"
+version = "0.1.0"
+dependencies = [
+ "aether-provider-transport",
+ "base64",
+ "http",
+ "serde_json",
+ "uuid",
+]
+
 [[package]]
 name = "aether-billing"
 version = "0.1.0"
 dependencies = [
- "aether-data",
+ "aether-data-contracts",
  "aether-usage-runtime",
  "async-trait",
  "serde",
@@ -66,6 +77,7 @@ name = "aether-data"
 version = "0.1.0"
 dependencies = [
  "aether-cache",
+ "aether-data-contracts",
  "aether-wallet",
  "async-trait",
  "chrono",
@@ -81,15 +93,27 @@ dependencies = [
  "uuid",
 ]
 
+[[package]]
+name = "aether-data-contracts"
+version = "0.1.0"
+dependencies = [
+ "async-trait",
+ "serde",
+ "serde_json",
+ "thiserror 2.0.18",
+]
+
 [[package]]
 name = "aether-gateway"
 version = "0.1.0"
 dependencies = [
+ "aether-ai-pipeline",
  "aether-billing",
  "aether-cache",
  "aether-contracts",
  "aether-crypto",
  "aether-data",
+ "aether-data-contracts",
  "aether-http",
  "aether-model-fetch",
  "aether-provider-transport",
@@ -120,6 +144,7 @@ dependencies = [
  "rustls",
  "serde",
  "serde_json",
+ "sha1",
  "sha2",
  "sqlx",
  "thiserror 2.0.18",
@@ -147,7 +172,7 @@ name = "aether-model-fetch"
 version = "0.1.0"
 dependencies = [
  "aether-contracts",
- "aether-data",
+ "aether-data-contracts",
  "aether-provider-transport",
  "aether-scheduler-core",
  "async-trait",
@@ -164,6 +189,7 @@ dependencies = [
  "aether-contracts",
  "aether-crypto",
  "aether-data",
+ "aether-data-contracts",
  "aether-video-tasks-core",
  "async-trait",
  "axum",
@@ -246,7 +272,7 @@ name = "aether-scheduler-core"
 version = "0.1.0"
 dependencies = [
  "aether-contracts",
- "aether-data",
+ "aether-data-contracts",
  "aether-wallet",
  "regex",
  "serde",
@@ -282,6 +308,7 @@ version = "0.1.0"
 dependencies = [
  "aether-contracts",
  "aether-data",
+ "aether-data-contracts",
  "async-trait",
  "base64",
  "serde",
@@ -295,7 +322,7 @@ name = "aether-video-tasks-core"
 version = "0.1.0"
 dependencies = [
  "aether-contracts",
- "aether-data",
+ "aether-data-contracts",
  "async-trait",
  "serde",
  "serde_json",

Cargo.toml

@@ -1,6 +1,8 @@
 [workspace]
 members = [
     "apps/aether-proxy",
+    "crates/aether-ai-pipeline",
+    "crates/aether-data-contracts",
     "crates/aether-cache",
     "crates/aether-billing",
     "crates/aether-wallet",
@@ -25,6 +27,8 @@ license = "LicenseRef-Aether-NonCommercial"
 repository = "https://github.com/fawney19/Aether.git"
 
 [workspace.dependencies]
+aether-ai-pipeline = { path = "crates/aether-ai-pipeline" }
+aether-data-contracts = { path = "crates/aether-data-contracts" }
 aether-cache = { path = "crates/aether-cache" }
 aether-billing = { path = "crates/aether-billing" }
 aether-wallet = { path = "crates/aether-wallet" }

apps/aether-gateway/Cargo.toml

@@ -7,11 +7,13 @@ repository.workspace = true
 description = "Rust ingress gateway for Aether phase 3a transparent proxy"
 
 [dependencies]
+aether-ai-pipeline.workspace = true
 aether-billing.workspace = true
 aether-cache.workspace = true
 aether-contracts.workspace = true
 aether-crypto.workspace = true
 aether-data.workspace = true
+aether-data-contracts.workspace = true
 aether-http.workspace = true
 aether-model-fetch.workspace = true
 aether-provider-transport.workspace = true
@@ -42,6 +44,7 @@ reqwest.workspace = true
 rustls.workspace = true
 serde.workspace = true
 serde_json.workspace = true
+sha1 = "0.10"
 sha2.workspace = true
 sqlx.workspace = true
 thiserror.workspace = true

apps/aether-gateway/src/ai_pipeline/adaptation/private_envelope.rs

@@ -8,9 +8,7 @@ use crate::ai_pipeline::adaptation::surfaces::{
     provider_adaptation_should_unwrap_stream_envelope, ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
     GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME,
 };
-use crate::ai_pipeline::runtime::adapters::kiro::{
+use crate::ai_pipeline::runtime::adapters::kiro::{KiroToClaudeCliStreamState, KIRO_ENVELOPE_NAME};
-    KiroToClaudeCliStreamState, KIRO_ENVELOPE_NAME,
-};
 use crate::{usage::GatewaySyncReportRequest, GatewayError};
 
 enum ProviderPrivateStreamNormalizeMode {

apps/aether-gateway/src/ai_pipeline/adaptation/surfaces.rs

@@ -1,190 +1,8 @@
-use crate::ai_pipeline::runtime::adapters::{
+pub(crate) use aether_ai_pipeline::adaptation::surfaces::{
-    antigravity::ANTIGRAVITY_PROVIDER_TYPE,
-    kiro::{KIRO_ENVELOPE_NAME, PROVIDER_TYPE as KIRO_PROVIDER_TYPE},
-};
-
-pub(crate) const ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME: &str = "antigravity:v1internal";
-pub(crate) const GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME: &str = "gemini_cli:v1internal";
-
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
-pub(crate) enum ProviderAdaptationSurface {
-    AntigravityGeminiChat,
-    AntigravityGeminiCli,
-    GeminiCliV1Internal,
-    KiroClaudeCli,
-}
-
-#[derive(Debug, Clone, Copy)]
-pub(crate) struct ProviderAdaptationDescriptor {
-    pub(crate) surface: ProviderAdaptationSurface,
-    pub(crate) provider_type: Option<&'static str>,
-    pub(crate) envelope_name: &'static str,
-    pub(crate) anchor_api_format: &'static str,
-    pub(crate) supports_request_bridge: bool,
-    pub(crate) supports_sync_finalize_bridge: bool,
-    pub(crate) supports_stream_bridge: bool,
-    pub(crate) requires_eventstream_accept: bool,
-    pub(crate) unwraps_response_envelope: bool,
-}
-
-const PROVIDER_ADAPTATION_SURFACES: &[ProviderAdaptationDescriptor] = &[
-    ProviderAdaptationDescriptor {
-        surface: ProviderAdaptationSurface::AntigravityGeminiChat,
-        provider_type: Some(ANTIGRAVITY_PROVIDER_TYPE),
-        envelope_name: ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
-        anchor_api_format: "gemini:chat",
-        supports_request_bridge: true,
-        supports_sync_finalize_bridge: true,
-        supports_stream_bridge: true,
-        requires_eventstream_accept: false,
-        unwraps_response_envelope: true,
-    },
-    ProviderAdaptationDescriptor {
-        surface: ProviderAdaptationSurface::AntigravityGeminiCli,
-        provider_type: Some(ANTIGRAVITY_PROVIDER_TYPE),
-        envelope_name: ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
-        anchor_api_format: "gemini:cli",
-        supports_request_bridge: true,
-        supports_sync_finalize_bridge: true,
-        supports_stream_bridge: true,
-        requires_eventstream_accept: false,
-        unwraps_response_envelope: true,
-    },
-    ProviderAdaptationDescriptor {
-        surface: ProviderAdaptationSurface::GeminiCliV1Internal,
-        provider_type: None,
-        envelope_name: GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME,
-        anchor_api_format: "gemini:cli",
-        supports_request_bridge: false,
-        supports_sync_finalize_bridge: true,
-        supports_stream_bridge: true,
-        requires_eventstream_accept: false,
-        unwraps_response_envelope: true,
-    },
-    ProviderAdaptationDescriptor {
-        surface: ProviderAdaptationSurface::KiroClaudeCli,
-        provider_type: Some(KIRO_PROVIDER_TYPE),
-        envelope_name: KIRO_ENVELOPE_NAME,
-        anchor_api_format: "claude:cli",
-        supports_request_bridge: true,
-        supports_sync_finalize_bridge: true,
-        supports_stream_bridge: true,
-        requires_eventstream_accept: true,
-        unwraps_response_envelope: false,
-    },
-];
-
-pub(crate) fn provider_adaptation_descriptor_for_envelope(
-    envelope_name: &str,
-    provider_api_format: &str,
-) -> Option<&'static ProviderAdaptationDescriptor> {
-    let envelope_name = envelope_name.trim();
-    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
-    PROVIDER_ADAPTATION_SURFACES.iter().find(|descriptor| {
-        descriptor.envelope_name.eq_ignore_ascii_case(envelope_name)
-            && descriptor
-                .anchor_api_format
-                .eq_ignore_ascii_case(provider_api_format.as_str())
-    })
-}
-
-pub(crate) fn provider_adaptation_descriptor_for_provider_type(
-    provider_type: &str,
-    provider_api_format: &str,
-) -> Option<&'static ProviderAdaptationDescriptor> {
-    let provider_type = provider_type.trim();
-    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
-    PROVIDER_ADAPTATION_SURFACES.iter().find(|descriptor| {
-        descriptor
-            .provider_type
-            .is_some_and(|value| value.eq_ignore_ascii_case(provider_type))
-            && descriptor
-                .anchor_api_format
-                .eq_ignore_ascii_case(provider_api_format.as_str())
-    })
-}
-
-pub(crate) fn provider_adaptation_anchor_api_format(
-    envelope_name: &str,
-    provider_api_format: &str,
-) -> Option<&'static str> {
-    provider_adaptation_descriptor_for_envelope(envelope_name, provider_api_format)
-        .map(|descriptor| descriptor.anchor_api_format)
-}
-
-pub(crate) fn provider_adaptation_allows_sync_finalize_envelope(
-    envelope_name: &str,
-    provider_api_format: &str,
-) -> bool {
-    provider_adaptation_descriptor_for_envelope(envelope_name, provider_api_format)
-        .is_some_and(|descriptor| descriptor.supports_sync_finalize_bridge)
-}
-
-pub(crate) fn provider_adaptation_requires_eventstream_accept(
-    envelope_name: Option<&str>,
-    provider_api_format: &str,
-) -> bool {
-    envelope_name
-        .and_then(|value| provider_adaptation_descriptor_for_envelope(value, provider_api_format))
-        .is_some_and(|descriptor| descriptor.requires_eventstream_accept)
-}
-
-pub(crate) fn provider_adaptation_should_unwrap_stream_envelope(
-    envelope_name: &str,
-    provider_api_format: &str,
-) -> bool {
-    provider_adaptation_descriptor_for_envelope(envelope_name, provider_api_format)
-        .is_some_and(|descriptor| descriptor.unwraps_response_envelope)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::{
     provider_adaptation_allows_sync_finalize_envelope, provider_adaptation_anchor_api_format,
+    provider_adaptation_descriptor_for_envelope, provider_adaptation_descriptor_for_provider_type,
     provider_adaptation_requires_eventstream_accept,
-        provider_adaptation_should_unwrap_stream_envelope, ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
+    provider_adaptation_should_unwrap_stream_envelope, ProviderAdaptationDescriptor,
-        GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME, KIRO_ENVELOPE_NAME,
+    ProviderAdaptationSurface, ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
+    GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME,
 };
-
-    #[test]
-    fn resolves_private_surface_anchor_contracts() {
-        assert_eq!(
-            provider_adaptation_anchor_api_format(
-                ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
-                "gemini:cli"
-            ),
-            Some("gemini:cli")
-        );
-        assert_eq!(
-            provider_adaptation_anchor_api_format(
-                GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME,
-                "gemini:cli"
-            ),
-            Some("gemini:cli")
-        );
-        assert_eq!(
-            provider_adaptation_anchor_api_format(KIRO_ENVELOPE_NAME, "claude:cli"),
-            Some("claude:cli")
-        );
-    }
-
-    #[test]
-    fn exposes_private_surface_capabilities() {
-        assert!(provider_adaptation_allows_sync_finalize_envelope(
-            ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME,
-            "gemini:chat"
-        ));
-        assert!(provider_adaptation_should_unwrap_stream_envelope(
-            GEMINI_CLI_V1INTERNAL_ENVELOPE_NAME,
-            "gemini:cli"
-        ));
-        assert!(provider_adaptation_requires_eventstream_accept(
-            Some(KIRO_ENVELOPE_NAME),
-            "claude:cli"
-        ));
-        assert!(!provider_adaptation_requires_eventstream_accept(
-            Some(ANTIGRAVITY_V1INTERNAL_ENVELOPE_NAME),
-            "gemini:cli"
-        ));
-    }
-}

apps/aether-gateway/src/ai_pipeline/contracts/actions.rs

@@ -1,5 +0,0 @@
-pub(crate) const EXECUTION_RUNTIME_SYNC_ACTION: &str = "execution_runtime_sync";
-pub(crate) const EXECUTION_RUNTIME_SYNC_DECISION_ACTION: &str = "execution_runtime_sync_decision";
-pub(crate) const EXECUTION_RUNTIME_STREAM_ACTION: &str = "execution_runtime_stream";
-pub(crate) const EXECUTION_RUNTIME_STREAM_DECISION_ACTION: &str =
-    "execution_runtime_stream_decision";

apps/aether-gateway/src/ai_pipeline/contracts/control_payloads.rs

@@ -4,9 +4,10 @@ use aether_contracts::{ExecutionPlan, ExecutionTimeouts, ProxySnapshot};
 use serde::{Deserialize, Serialize};
 use tracing::warn;
 
-use crate::control::GatewayControlAuthContext;
+use crate::ai_pipeline::control_facade::{
-use crate::control::GatewayControlDecision;
+    collect_control_headers, resolve_execution_runtime_auth_context, GatewayControlAuthContext,
-use crate::headers::collect_control_headers;
+    GatewayControlDecision,
+};
 use crate::{AppState, GatewayError};
 
 #[derive(Debug, Serialize)]
@@ -142,7 +143,7 @@ pub(crate) async fn build_gateway_plan_request(
     body_json: serde_json::Value,
     body_base64: Option<String>,
 ) -> Result<GatewayControlPlanRequest, GatewayError> {
-    let auth_context = crate::control::resolve_execution_runtime_auth_context(
+    let auth_context = resolve_execution_runtime_auth_context(
         state,
         decision,
         &parts.headers,

apps/aether-gateway/src/ai_pipeline/contracts/mod.rs

@@ -1,43 +1,38 @@
-pub(crate) mod actions;
 pub(crate) mod control_payloads;
-pub(crate) mod plan_kinds;
-pub(crate) mod report_kinds;
 
-pub(crate) use actions::{
+pub(crate) use aether_ai_pipeline::contracts::{
-    EXECUTION_RUNTIME_STREAM_ACTION, EXECUTION_RUNTIME_STREAM_DECISION_ACTION,
+    core_error_background_report_kind, core_error_default_client_api_format,
-    EXECUTION_RUNTIME_SYNC_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+    core_success_background_report_kind, implicit_sync_finalize_report_kind,
+    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_STREAM_SUCCESS_REPORT_KIND,
+    CLAUDE_CHAT_SYNC_ERROR_REPORT_KIND, CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND,
+    CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CHAT_SYNC_SUCCESS_REPORT_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
+    CLAUDE_CLI_STREAM_SUCCESS_REPORT_KIND, CLAUDE_CLI_SYNC_ERROR_REPORT_KIND,
+    CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND, CLAUDE_CLI_SYNC_PLAN_KIND,
+    CLAUDE_CLI_SYNC_SUCCESS_REPORT_KIND, EXECUTION_RUNTIME_STREAM_ACTION,
+    EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_CHAT_STREAM_PLAN_KIND,
+    GEMINI_CHAT_STREAM_SUCCESS_REPORT_KIND, GEMINI_CHAT_SYNC_ERROR_REPORT_KIND,
+    GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
+    GEMINI_CHAT_SYNC_SUCCESS_REPORT_KIND, GEMINI_CLI_STREAM_PLAN_KIND,
+    GEMINI_CLI_STREAM_SUCCESS_REPORT_KIND, GEMINI_CLI_SYNC_ERROR_REPORT_KIND,
+    GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
+    GEMINI_CLI_SYNC_SUCCESS_REPORT_KIND, GEMINI_FILES_DELETE_PLAN_KIND,
+    GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND,
+    GEMINI_FILES_UPLOAD_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    GEMINI_VIDEO_CREATE_SYNC_FINALIZE_REPORT_KIND, GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND,
+    OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_STREAM_SUCCESS_REPORT_KIND,
+    OPENAI_CHAT_SYNC_ERROR_REPORT_KIND, OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND,
+    OPENAI_CHAT_SYNC_PLAN_KIND, OPENAI_CHAT_SYNC_SUCCESS_REPORT_KIND, OPENAI_CLI_STREAM_PLAN_KIND,
+    OPENAI_CLI_STREAM_SUCCESS_REPORT_KIND, OPENAI_CLI_SYNC_ERROR_REPORT_KIND,
+    OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND, OPENAI_CLI_SYNC_PLAN_KIND,
+    OPENAI_CLI_SYNC_SUCCESS_REPORT_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
+    OPENAI_COMPACT_SYNC_ERROR_REPORT_KIND, OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND,
+    OPENAI_COMPACT_SYNC_PLAN_KIND, OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_FINALIZE_REPORT_KIND,
+    OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
 };
 pub(crate) use control_payloads::{
     build_gateway_plan_request, generic_decision_missing_exact_provider_request,
     GatewayControlPlanRequest, GatewayControlPlanResponse, GatewayControlSyncDecisionResponse,
 };
-pub(crate) use plan_kinds::{
-    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
-    CLAUDE_CLI_SYNC_PLAN_KIND, GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
-    GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND, GEMINI_FILES_DELETE_PLAN_KIND,
-    GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND,
-    GEMINI_FILES_UPLOAD_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
-    GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
-    OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
-    OPENAI_COMPACT_SYNC_PLAN_KIND, OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND,
-    OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
-    OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND, OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
-};
-pub(crate) use report_kinds::{
-    core_error_background_report_kind, core_error_default_client_api_format,
-    core_success_background_report_kind, implicit_sync_finalize_report_kind,
-    CLAUDE_CHAT_STREAM_SUCCESS_REPORT_KIND, CLAUDE_CHAT_SYNC_ERROR_REPORT_KIND,
-    CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND, CLAUDE_CHAT_SYNC_SUCCESS_REPORT_KIND,
-    CLAUDE_CLI_STREAM_SUCCESS_REPORT_KIND, CLAUDE_CLI_SYNC_ERROR_REPORT_KIND,
-    CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND, CLAUDE_CLI_SYNC_SUCCESS_REPORT_KIND,
-    GEMINI_CHAT_STREAM_SUCCESS_REPORT_KIND, GEMINI_CHAT_SYNC_ERROR_REPORT_KIND,
-    GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND, GEMINI_CHAT_SYNC_SUCCESS_REPORT_KIND,
-    GEMINI_CLI_STREAM_SUCCESS_REPORT_KIND, GEMINI_CLI_SYNC_ERROR_REPORT_KIND,
-    GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND, GEMINI_CLI_SYNC_SUCCESS_REPORT_KIND,
-    GEMINI_VIDEO_CREATE_SYNC_FINALIZE_REPORT_KIND, OPENAI_CHAT_STREAM_SUCCESS_REPORT_KIND,
-    OPENAI_CHAT_SYNC_ERROR_REPORT_KIND, OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND,
-    OPENAI_CHAT_SYNC_SUCCESS_REPORT_KIND, OPENAI_CLI_STREAM_SUCCESS_REPORT_KIND,
-    OPENAI_CLI_SYNC_ERROR_REPORT_KIND, OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND,
-    OPENAI_CLI_SYNC_SUCCESS_REPORT_KIND, OPENAI_COMPACT_SYNC_ERROR_REPORT_KIND,
-    OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND, OPENAI_VIDEO_CREATE_SYNC_FINALIZE_REPORT_KIND,
-};

apps/aether-gateway/src/ai_pipeline/contracts/plan_kinds.rs

@@ -1,26 +0,0 @@
-pub(crate) const GEMINI_FILES_GET_PLAN_KIND: &str = "gemini_files_get";
-pub(crate) const GEMINI_FILES_UPLOAD_PLAN_KIND: &str = "gemini_files_upload";
-pub(crate) const GEMINI_FILES_LIST_PLAN_KIND: &str = "gemini_files_list";
-pub(crate) const GEMINI_FILES_DELETE_PLAN_KIND: &str = "gemini_files_delete";
-pub(crate) const GEMINI_FILES_DOWNLOAD_PLAN_KIND: &str = "gemini_files_download";
-pub(crate) const OPENAI_VIDEO_CONTENT_PLAN_KIND: &str = "openai_video_content";
-pub(crate) const OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND: &str = "openai_video_cancel_sync";
-pub(crate) const OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND: &str = "openai_video_remix_sync";
-pub(crate) const OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND: &str = "openai_video_delete_sync";
-pub(crate) const GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND: &str = "gemini_video_create_sync";
-pub(crate) const GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND: &str = "gemini_video_cancel_sync";
-pub(crate) const OPENAI_CHAT_STREAM_PLAN_KIND: &str = "openai_chat_stream";
-pub(crate) const CLAUDE_CHAT_STREAM_PLAN_KIND: &str = "claude_chat_stream";
-pub(crate) const GEMINI_CHAT_STREAM_PLAN_KIND: &str = "gemini_chat_stream";
-pub(crate) const OPENAI_CLI_STREAM_PLAN_KIND: &str = "openai_cli_stream";
-pub(crate) const OPENAI_COMPACT_STREAM_PLAN_KIND: &str = "openai_compact_stream";
-pub(crate) const CLAUDE_CLI_STREAM_PLAN_KIND: &str = "claude_cli_stream";
-pub(crate) const GEMINI_CLI_STREAM_PLAN_KIND: &str = "gemini_cli_stream";
-pub(crate) const OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND: &str = "openai_video_create_sync";
-pub(crate) const OPENAI_CHAT_SYNC_PLAN_KIND: &str = "openai_chat_sync";
-pub(crate) const OPENAI_CLI_SYNC_PLAN_KIND: &str = "openai_cli_sync";
-pub(crate) const OPENAI_COMPACT_SYNC_PLAN_KIND: &str = "openai_compact_sync";
-pub(crate) const CLAUDE_CHAT_SYNC_PLAN_KIND: &str = "claude_chat_sync";
-pub(crate) const GEMINI_CHAT_SYNC_PLAN_KIND: &str = "gemini_chat_sync";
-pub(crate) const CLAUDE_CLI_SYNC_PLAN_KIND: &str = "claude_cli_sync";
-pub(crate) const GEMINI_CLI_SYNC_PLAN_KIND: &str = "gemini_cli_sync";

apps/aether-gateway/src/ai_pipeline/contracts/report_kinds.rs

@@ -1,92 +0,0 @@
-use super::plan_kinds::{
-    CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_SYNC_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
-    GEMINI_CLI_SYNC_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND,
-    OPENAI_COMPACT_SYNC_PLAN_KIND,
-};
-
-pub(crate) const OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND: &str = "openai_chat_sync_finalize";
-pub(crate) const CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND: &str = "claude_chat_sync_finalize";
-pub(crate) const GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND: &str = "gemini_chat_sync_finalize";
-pub(crate) const OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND: &str = "openai_cli_sync_finalize";
-pub(crate) const OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND: &str = "openai_compact_sync_finalize";
-pub(crate) const CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND: &str = "claude_cli_sync_finalize";
-pub(crate) const GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND: &str = "gemini_cli_sync_finalize";
-pub(crate) const OPENAI_VIDEO_CREATE_SYNC_FINALIZE_REPORT_KIND: &str =
-    "openai_video_create_sync_finalize";
-pub(crate) const GEMINI_VIDEO_CREATE_SYNC_FINALIZE_REPORT_KIND: &str =
-    "gemini_video_create_sync_finalize";
-
-pub(crate) const OPENAI_CHAT_SYNC_SUCCESS_REPORT_KIND: &str = "openai_chat_sync_success";
-pub(crate) const CLAUDE_CHAT_SYNC_SUCCESS_REPORT_KIND: &str = "claude_chat_sync_success";
-pub(crate) const GEMINI_CHAT_SYNC_SUCCESS_REPORT_KIND: &str = "gemini_chat_sync_success";
-pub(crate) const OPENAI_CLI_SYNC_SUCCESS_REPORT_KIND: &str = "openai_cli_sync_success";
-pub(crate) const CLAUDE_CLI_SYNC_SUCCESS_REPORT_KIND: &str = "claude_cli_sync_success";
-pub(crate) const GEMINI_CLI_SYNC_SUCCESS_REPORT_KIND: &str = "gemini_cli_sync_success";
-
-pub(crate) const OPENAI_CHAT_STREAM_SUCCESS_REPORT_KIND: &str = "openai_chat_stream_success";
-pub(crate) const CLAUDE_CHAT_STREAM_SUCCESS_REPORT_KIND: &str = "claude_chat_stream_success";
-pub(crate) const GEMINI_CHAT_STREAM_SUCCESS_REPORT_KIND: &str = "gemini_chat_stream_success";
-pub(crate) const OPENAI_CLI_STREAM_SUCCESS_REPORT_KIND: &str = "openai_cli_stream_success";
-pub(crate) const CLAUDE_CLI_STREAM_SUCCESS_REPORT_KIND: &str = "claude_cli_stream_success";
-pub(crate) const GEMINI_CLI_STREAM_SUCCESS_REPORT_KIND: &str = "gemini_cli_stream_success";
-
-pub(crate) const OPENAI_CHAT_SYNC_ERROR_REPORT_KIND: &str = "openai_chat_sync_error";
-pub(crate) const CLAUDE_CHAT_SYNC_ERROR_REPORT_KIND: &str = "claude_chat_sync_error";
-pub(crate) const GEMINI_CHAT_SYNC_ERROR_REPORT_KIND: &str = "gemini_chat_sync_error";
-pub(crate) const OPENAI_CLI_SYNC_ERROR_REPORT_KIND: &str = "openai_cli_sync_error";
-pub(crate) const OPENAI_COMPACT_SYNC_ERROR_REPORT_KIND: &str = "openai_compact_sync_error";
-pub(crate) const CLAUDE_CLI_SYNC_ERROR_REPORT_KIND: &str = "claude_cli_sync_error";
-pub(crate) const GEMINI_CLI_SYNC_ERROR_REPORT_KIND: &str = "gemini_cli_sync_error";
-
-pub(crate) fn implicit_sync_finalize_report_kind(plan_kind: &str) -> Option<&'static str> {
-    match plan_kind {
-        OPENAI_CHAT_SYNC_PLAN_KIND => Some(OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND),
-        CLAUDE_CHAT_SYNC_PLAN_KIND => Some(CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND),
-        GEMINI_CHAT_SYNC_PLAN_KIND => Some(GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND),
-        OPENAI_CLI_SYNC_PLAN_KIND => Some(OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND),
-        OPENAI_COMPACT_SYNC_PLAN_KIND => Some(OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND),
-        CLAUDE_CLI_SYNC_PLAN_KIND => Some(CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND),
-        GEMINI_CLI_SYNC_PLAN_KIND => Some(GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND),
-        _ => None,
-    }
-}
-
-pub(crate) fn core_error_default_client_api_format(report_kind: &str) -> Option<&'static str> {
-    match report_kind {
-        OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND => Some("openai:chat"),
-        CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND => Some("claude:chat"),
-        GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND => Some("gemini:chat"),
-        OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND => Some("openai:cli"),
-        OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND => Some("openai:compact"),
-        CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND => Some("claude:cli"),
-        GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND => Some("gemini:cli"),
-        _ => None,
-    }
-}
-
-pub(crate) fn core_error_background_report_kind(report_kind: &str) -> Option<&'static str> {
-    match report_kind {
-        OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND => Some(OPENAI_CHAT_SYNC_ERROR_REPORT_KIND),
-        CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND => Some(CLAUDE_CHAT_SYNC_ERROR_REPORT_KIND),
-        GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND => Some(GEMINI_CHAT_SYNC_ERROR_REPORT_KIND),
-        OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND => Some(OPENAI_CLI_SYNC_ERROR_REPORT_KIND),
-        OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND => Some(OPENAI_COMPACT_SYNC_ERROR_REPORT_KIND),
-        CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND => Some(CLAUDE_CLI_SYNC_ERROR_REPORT_KIND),
-        GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND => Some(GEMINI_CLI_SYNC_ERROR_REPORT_KIND),
-        _ => None,
-    }
-}
-
-pub(crate) fn core_success_background_report_kind(report_kind: &str) -> Option<&'static str> {
-    match report_kind {
-        OPENAI_CHAT_SYNC_FINALIZE_REPORT_KIND => Some(OPENAI_CHAT_SYNC_SUCCESS_REPORT_KIND),
-        CLAUDE_CHAT_SYNC_FINALIZE_REPORT_KIND => Some(CLAUDE_CHAT_SYNC_SUCCESS_REPORT_KIND),
-        GEMINI_CHAT_SYNC_FINALIZE_REPORT_KIND => Some(GEMINI_CHAT_SYNC_SUCCESS_REPORT_KIND),
-        OPENAI_CLI_SYNC_FINALIZE_REPORT_KIND | OPENAI_COMPACT_SYNC_FINALIZE_REPORT_KIND => {
-            Some(OPENAI_CLI_SYNC_SUCCESS_REPORT_KIND)
-        }
-        CLAUDE_CLI_SYNC_FINALIZE_REPORT_KIND => Some(CLAUDE_CLI_SYNC_SUCCESS_REPORT_KIND),
-        GEMINI_CLI_SYNC_FINALIZE_REPORT_KIND => Some(GEMINI_CLI_SYNC_SUCCESS_REPORT_KIND),
-        _ => None,
-    }
-}

apps/aether-gateway/src/ai_pipeline/control_facade.rs

@@ -0,0 +1,26 @@
+use axum::http::Uri;
+
+use crate::{AppState, GatewayError};
+
+pub(crate) use crate::control::{GatewayControlAuthContext, GatewayControlDecision};
+
+pub(crate) async fn resolve_execution_runtime_auth_context(
+    state: &AppState,
+    decision: &GatewayControlDecision,
+    headers: &http::HeaderMap,
+    uri: &Uri,
+    trace_id: &str,
+) -> Result<Option<GatewayControlAuthContext>, GatewayError> {
+    crate::control::resolve_execution_runtime_auth_context(state, decision, headers, uri, trace_id)
+        .await
+}
+
+pub(crate) fn collect_control_headers(
+    headers: &http::HeaderMap,
+) -> std::collections::BTreeMap<String, String> {
+    crate::headers::collect_control_headers(headers)
+}
+
+pub(crate) fn is_json_request(headers: &http::HeaderMap) -> bool {
+    crate::headers::is_json_request(headers)
+}

apps/aether-gateway/src/ai_pipeline/conversion/mod.rs

@@ -1,11 +1,10 @@
-pub(crate) mod error;
 pub(crate) mod registry;
 pub(crate) mod request;
 pub(crate) mod response;
 
 #[cfg(test)]
-pub(crate) use error::core_success_background_report_kind;
+pub(crate) use aether_ai_pipeline::conversion::core_success_background_report_kind;
-pub(crate) use error::{
+pub(crate) use aether_ai_pipeline::conversion::{
     build_core_error_body_for_client_format, core_error_background_report_kind,
     core_error_default_client_api_format, is_core_error_finalize_kind, LocalCoreSyncErrorKind,
 };

apps/aether-gateway/src/ai_pipeline/conversion/registry.rs

@@ -1,60 +1,17 @@
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::provider_transport_facade::auth::{
     resolve_local_gemini_auth, resolve_local_openai_chat_auth, resolve_local_standard_auth,
 };
-use crate::provider_transport::policy::{
+use crate::ai_pipeline::provider_transport_facade::policy::{
     supports_local_openai_chat_transport, supports_local_standard_transport_with_network,
 };
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     supports_local_gemini_transport_with_network, GatewayProviderTransportSnapshot,
 };
 
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) use aether_ai_pipeline::conversion::{
-pub(crate) enum RequestConversionKind {
+    request_conversion_kind, sync_chat_response_conversion_kind, sync_cli_response_conversion_kind,
-    ToOpenAIChat,
+    RequestConversionKind, SyncChatResponseConversionKind, SyncCliResponseConversionKind,
-    ToOpenAIFamilyCli,
+};
-    ToOpenAICompact,
-    ToClaudeStandard,
-    ToGeminiStandard,
-}
-
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
-pub(crate) enum SyncChatResponseConversionKind {
-    ToOpenAIChat,
-    ToClaudeChat,
-    ToGeminiChat,
-}
-
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
-pub(crate) enum SyncCliResponseConversionKind {
-    ToOpenAIFamilyCli,
-    ToClaudeCli,
-    ToGeminiCli,
-}
-
-pub(crate) fn request_conversion_kind(
-    client_api_format: &str,
-    provider_api_format: &str,
-) -> Option<RequestConversionKind> {
-    let client_api_format = client_api_format.trim().to_ascii_lowercase();
-    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
-    if client_api_format == provider_api_format {
-        return None;
-    }
-    if !is_standard_api_format(client_api_format.as_str())
-        || !is_standard_api_format(provider_api_format.as_str())
-    {
-        return None;
-    }
-
-    match provider_api_format.as_str() {
-        "openai:chat" => Some(RequestConversionKind::ToOpenAIChat),
-        "openai:cli" => Some(RequestConversionKind::ToOpenAIFamilyCli),
-        "openai:compact" => Some(RequestConversionKind::ToOpenAICompact),
-        "claude:chat" | "claude:cli" => Some(RequestConversionKind::ToClaudeStandard),
-        "gemini:chat" | "gemini:cli" => Some(RequestConversionKind::ToGeminiStandard),
-        _ => None,
-    }
-}
 
 pub(crate) fn request_conversion_transport_supported(
     transport: &GatewayProviderTransportSnapshot,
@@ -100,59 +57,6 @@ pub(crate) fn request_conversion_direct_auth(
     }
 }
 
-pub(crate) fn sync_chat_response_conversion_kind(
-    provider_api_format: &str,
-    client_api_format: &str,
-) -> Option<SyncChatResponseConversionKind> {
-    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
-    let client_api_format = client_api_format.trim().to_ascii_lowercase();
-    if provider_api_format == client_api_format {
-        return None;
-    }
-    if !is_standard_api_format(provider_api_format.as_str()) {
-        return None;
-    }
-    match client_api_format.as_str() {
-        "openai:chat" => Some(SyncChatResponseConversionKind::ToOpenAIChat),
-        "claude:chat" => Some(SyncChatResponseConversionKind::ToClaudeChat),
-        "gemini:chat" => Some(SyncChatResponseConversionKind::ToGeminiChat),
-        _ => None,
-    }
-}
-
-pub(crate) fn sync_cli_response_conversion_kind(
-    provider_api_format: &str,
-    client_api_format: &str,
-) -> Option<SyncCliResponseConversionKind> {
-    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
-    let client_api_format = client_api_format.trim().to_ascii_lowercase();
-    if provider_api_format == client_api_format {
-        return None;
-    }
-    if !is_standard_api_format(provider_api_format.as_str()) {
-        return None;
-    }
-    match client_api_format.as_str() {
-        "openai:cli" | "openai:compact" => Some(SyncCliResponseConversionKind::ToOpenAIFamilyCli),
-        "claude:cli" => Some(SyncCliResponseConversionKind::ToClaudeCli),
-        "gemini:cli" => Some(SyncCliResponseConversionKind::ToGeminiCli),
-        _ => None,
-    }
-}
-
-fn is_standard_api_format(api_format: &str) -> bool {
-    matches!(
-        api_format,
-        "openai:chat"
-            | "openai:cli"
-            | "openai:compact"
-            | "claude:chat"
-            | "claude:cli"
-            | "gemini:chat"
-            | "gemini:cli"
-    )
-}
-
 #[cfg(test)]
 mod tests {
     use super::{

apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/mod.rs

@@ -1,8 +0,0 @@
-mod claude;
-mod gemini;
-mod openai_cli;
-mod shared;
-
-pub(crate) use claude::convert_openai_chat_request_to_claude_request;
-pub(crate) use gemini::convert_openai_chat_request_to_gemini_request;
-pub(crate) use openai_cli::convert_openai_chat_request_to_openai_cli_request;

apps/aether-gateway/src/ai_pipeline/conversion/request/mod.rs

@@ -1,12 +1,7 @@
-mod from_openai_chat;
+pub(crate) use aether_ai_pipeline::conversion::request::{
-mod to_openai_chat;
-
-pub(crate) use from_openai_chat::{
     convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
-    convert_openai_chat_request_to_openai_cli_request,
+    convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
-};
+    normalize_claude_request_to_openai_chat_request,
-pub(crate) use to_openai_chat::{
-    extract_openai_text_content, normalize_claude_request_to_openai_chat_request,
     normalize_gemini_request_to_openai_chat_request,
     normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
 };

apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/mod.rs

@@ -1,9 +0,0 @@
-mod claude;
-mod gemini;
-mod openai_cli;
-mod shared;
-
-pub(crate) use claude::normalize_claude_request_to_openai_chat_request;
-pub(crate) use gemini::normalize_gemini_request_to_openai_chat_request;
-pub(crate) use openai_cli::normalize_openai_cli_request_to_openai_chat_request;
-pub(crate) use shared::{extract_openai_text_content, parse_openai_tool_result_content};

apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/mod.rs

@@ -1,9 +0,0 @@
-mod claude_chat;
-mod gemini_chat;
-mod openai_cli;
-mod shared;
-
-pub(crate) use claude_chat::convert_openai_chat_response_to_claude_chat;
-pub(crate) use gemini_chat::convert_openai_chat_response_to_gemini_chat;
-pub(crate) use openai_cli::convert_openai_chat_response_to_openai_cli;
-pub(crate) use shared::build_openai_cli_response;

apps/aether-gateway/src/ai_pipeline/conversion/response/mod.rs

@@ -1,12 +1,7 @@
-mod from_openai_chat;
+pub(crate) use aether_ai_pipeline::conversion::response::{
-mod to_openai_chat;
+    build_openai_cli_response, convert_claude_chat_response_to_openai_chat,
-
+    convert_claude_cli_response_to_openai_cli, convert_gemini_chat_response_to_openai_chat,
-pub(crate) use from_openai_chat::{
+    convert_gemini_cli_response_to_openai_cli, convert_openai_chat_response_to_claude_chat,
-    build_openai_cli_response, convert_openai_chat_response_to_claude_chat,
     convert_openai_chat_response_to_gemini_chat, convert_openai_chat_response_to_openai_cli,
-};
-pub(crate) use to_openai_chat::{
-    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
-    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
     convert_openai_cli_response_to_openai_chat,
 };

apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/mod.rs

@@ -1,12 +0,0 @@
-mod claude_chat;
-mod claude_cli;
-mod gemini_chat;
-mod gemini_cli;
-mod openai_cli;
-mod shared;
-
-pub(crate) use claude_chat::convert_claude_chat_response_to_openai_chat;
-pub(crate) use claude_cli::convert_claude_cli_response_to_openai_cli;
-pub(crate) use gemini_chat::convert_gemini_chat_response_to_openai_chat;
-pub(crate) use gemini_cli::convert_gemini_cli_response_to_openai_cli;
-pub(crate) use openai_cli::convert_openai_cli_response_to_openai_chat;

apps/aether-gateway/src/ai_pipeline/execution_facade.rs

@@ -0,0 +1,14 @@
+use axum::body::Body;
+use axum::http::Response;
+
+use crate::{usage::GatewaySyncReportRequest, GatewayError};
+
+pub(crate) use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+
+pub(crate) fn maybe_build_local_sync_finalize_response(
+    trace_id: &str,
+    decision: &crate::ai_pipeline::control_facade::GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    crate::execution_runtime::maybe_build_local_sync_finalize_response(trace_id, decision, payload)
+}

apps/aether-gateway/src/ai_pipeline/finalize/common.rs

@@ -9,8 +9,8 @@ pub(crate) use crate::ai_pipeline::adaptation::private_envelope::{
     provider_private_response_allows_sync_finalize as local_finalize_allows_envelope,
 };
 use crate::ai_pipeline::contracts::core_success_background_report_kind;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::api::response::build_client_response_from_parts;
-use crate::control::GatewayControlDecision;
 use crate::{usage::GatewaySyncReportRequest, GatewayError};
 
 pub(crate) struct LocalCoreSyncFinalizeOutcome {
@@ -141,58 +141,3 @@ pub(crate) fn canonicalize_tool_arguments(value: Option<Value>) -> String {
 pub(crate) fn build_generated_tool_call_id(index: usize) -> String {
     format!("call_auto_{index}")
 }
-
-pub(crate) fn parse_stream_json_events(body: &[u8]) -> Option<Vec<Value>> {
-    let text = std::str::from_utf8(body).ok()?;
-    let trimmed = text.trim();
-    if trimmed.is_empty() {
-        return Some(Vec::new());
-    }
-
-    if trimmed.starts_with('[') {
-        let array_value: Value = serde_json::from_str(trimmed).ok()?;
-        let array = array_value.as_array()?;
-        return Some(
-            array
-                .iter()
-                .filter(|value| value.is_object())
-                .cloned()
-                .collect(),
-        );
-    }
-
-    let mut events = Vec::new();
-    let mut current_event_type: Option<String> = None;
-
-    for raw_line in text.lines() {
-        let line = raw_line.trim_matches('\r').trim();
-        if line.is_empty() || line.starts_with(':') {
-            continue;
-        }
-        if let Some(event_name) = line.strip_prefix("event:") {
-            current_event_type = Some(event_name.trim().to_string());
-            continue;
-        }
-        let data_line = if let Some(rest) = line.strip_prefix("data:") {
-            rest.trim()
-        } else {
-            line
-        };
-        if data_line.is_empty() || data_line == "[DONE]" {
-            continue;
-        }
-
-        let mut event: Value = serde_json::from_str(data_line).ok()?;
-        if let Some(event_object) = event.as_object_mut() {
-            if !event_object.contains_key("type") {
-                if let Some(event_name) = current_event_type.take() {
-                    event_object.insert("type".to_string(), Value::String(event_name));
-                }
-            }
-        }
-        events.push(event);
-        current_event_type = None;
-    }
-
-    Some(events)
-}

apps/aether-gateway/src/ai_pipeline/finalize/internal/mod.rs

@@ -2,7 +2,7 @@ use axum::body::Body;
 use axum::http::Response;
 use serde_json::Value;
 
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::{usage::GatewaySyncReportRequest, GatewayError};
 
 #[path = "stream_rewrite.rs"]

apps/aether-gateway/src/ai_pipeline/finalize/internal/stream_rewrite.rs

@@ -4,6 +4,9 @@ use crate::ai_pipeline::adaptation::private_envelope::transform_provider_private
 use crate::ai_pipeline::finalize::standard::StreamingStandardConversionState;
 use crate::ai_pipeline::runtime::adapters::kiro::KiroToClaudeCliStreamState;
 use crate::GatewayError;
+use aether_ai_pipeline::finalize::{
+    resolve_finalize_stream_rewrite_mode, FinalizeStreamRewriteMode,
+};
 
 enum RewriteMode {
     EnvelopeUnwrap,
@@ -21,103 +24,13 @@ pub(crate) fn maybe_build_local_stream_rewriter(
     report_context: Option<&Value>,
 ) -> Option<LocalStreamRewriter> {
     let report_context = report_context?;
-    let needs_conversion = report_context
+    let mode = match resolve_finalize_stream_rewrite_mode(report_context)? {
-        .get("needs_conversion")
+        FinalizeStreamRewriteMode::EnvelopeUnwrap => RewriteMode::EnvelopeUnwrap,
-        .and_then(Value::as_bool)
+        FinalizeStreamRewriteMode::Standard => {
-        .unwrap_or(false);
-    let envelope_name = report_context
-        .get("envelope_name")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-
-    let mode = if needs_conversion {
-        match (
-            envelope_name.as_str(),
-            provider_api_format.as_str(),
-            client_api_format.as_str(),
-        ) {
-            ("", "claude:chat", "openai:chat") => {
             RewriteMode::Standard(StreamingStandardConversionState::default())
         }
-            ("", "gemini:chat", "openai:chat") => {
+        FinalizeStreamRewriteMode::KiroToClaudeCli => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("", "openai:cli", "openai:chat") | ("", "openai:compact", "openai:chat") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("", "claude:cli", "openai:cli") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("", "claude:cli", "openai:compact") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("", "gemini:cli", "openai:cli") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("", "gemini:cli", "openai:compact") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("antigravity:v1internal", "gemini:chat", "openai:chat") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("antigravity:v1internal", "gemini:cli", "openai:cli") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            ("antigravity:v1internal", "gemini:cli", "openai:compact") => {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            _ if is_standard_chat_client_api_format(client_api_format.as_str())
-                && is_standard_provider_api_format(provider_api_format.as_str()) =>
-            {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            _ if is_standard_cli_client_api_format(client_api_format.as_str())
-                && is_standard_provider_api_format(provider_api_format.as_str()) =>
-            {
-                RewriteMode::Standard(StreamingStandardConversionState::default())
-            }
-            _ => return None,
-        }
-    } else {
-        match envelope_name.as_str() {
-            "antigravity:v1internal" => {
-                if provider_api_format == client_api_format
-                    && matches!(provider_api_format.as_str(), "gemini:chat" | "gemini:cli")
-                {
-                    RewriteMode::EnvelopeUnwrap
-                } else {
-                    return None;
-                }
-            }
-            "gemini_cli:v1internal" => {
-                if provider_api_format == "gemini:cli" && client_api_format == "gemini:cli" {
-                    RewriteMode::EnvelopeUnwrap
-                } else {
-                    return None;
-                }
-            }
-            "kiro:generateassistantresponse" => {
-                if provider_api_format == "claude:cli" && client_api_format == "claude:cli" {
             RewriteMode::KiroToClaudeCli(KiroToClaudeCliStreamState::new(report_context))
-                } else {
-                    return None;
-                }
-            }
-            _ => return None,
         }
     };
 
@@ -175,30 +88,6 @@ impl LocalStreamRewriter {
     }
 }
 
-fn is_standard_provider_api_format(api_format: &str) -> bool {
-    matches!(
-        api_format,
-        "openai:chat"
-            | "openai:cli"
-            | "openai:compact"
-            | "claude:chat"
-            | "claude:cli"
-            | "gemini:chat"
-            | "gemini:cli"
-    )
-}
-
-fn is_standard_chat_client_api_format(api_format: &str) -> bool {
-    matches!(api_format, "openai:chat" | "claude:chat" | "gemini:chat")
-}
-
-fn is_standard_cli_client_api_format(api_format: &str) -> bool {
-    matches!(
-        api_format,
-        "openai:cli" | "openai:compact" | "claude:cli" | "gemini:cli"
-    )
-}
-
 #[cfg(test)]
 #[path = "../tests_stream.rs"]
 mod tests;

apps/aether-gateway/src/ai_pipeline/finalize/internal/sync_finalize.rs

@@ -1,43 +1,23 @@
-use std::collections::BTreeMap;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
-
-use axum::body::Body;
-use axum::http::Response;
-use base64::Engine as _;
-use serde_json::{json, Map, Value};
-
-use crate::ai_pipeline::conversion::{
-    sync_chat_response_conversion_kind, sync_cli_response_conversion_kind,
-};
-use crate::api::response::build_client_response_from_parts;
-use crate::control::GatewayControlDecision;
 use crate::{usage::GatewaySyncReportRequest, GatewayError};
 
+pub(crate) use crate::ai_pipeline::conversion::response::{
+    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
+    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
+};
 pub(crate) use crate::ai_pipeline::finalize::common::{
     build_generated_tool_call_id, build_local_success_outcome,
     build_local_success_outcome_with_conversion_report, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
+    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
     LocalCoreSyncFinalizeOutcome,
 };
 pub(crate) use crate::ai_pipeline::finalize::standard::{
+    maybe_build_standard_sync_finalize_product_from_normalized_payload,
+    StandardSyncFinalizeNormalizedProduct,
+};
+pub(crate) use aether_ai_pipeline::finalize::sync_products::{
     aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
     aggregate_openai_chat_stream_sync_response, aggregate_openai_cli_stream_sync_response,
-    aggregate_standard_chat_stream_sync_response, aggregate_standard_cli_stream_sync_response,
-    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
-    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
-    convert_openai_chat_response_to_claude_chat, convert_openai_chat_response_to_gemini_chat,
-    convert_openai_chat_response_to_openai_cli, convert_openai_cli_response_to_openai_chat,
-    convert_standard_chat_response, convert_standard_cli_response,
-    maybe_build_local_claude_cli_stream_sync_response,
-    maybe_build_local_claude_stream_sync_response, maybe_build_local_claude_sync_response,
-    maybe_build_local_gemini_cli_stream_sync_response,
-    maybe_build_local_gemini_stream_sync_response, maybe_build_local_gemini_sync_response,
-    maybe_build_local_openai_chat_cross_format_stream_sync_response,
-    maybe_build_local_openai_chat_cross_format_sync_response,
-    maybe_build_local_openai_chat_stream_sync_response,
-    maybe_build_local_openai_chat_sync_response,
-    maybe_build_local_openai_cli_cross_format_stream_sync_response,
-    maybe_build_local_openai_cli_cross_format_sync_response,
-    maybe_build_local_openai_cli_stream_sync_response,
 };
 
 pub(crate) fn maybe_build_local_core_sync_finalize_response(
@@ -51,325 +31,50 @@ pub(crate) fn maybe_build_local_core_sync_finalize_response(
         return Ok(None);
     };
     let payload = &normalized_payload;
-    if let Some(response) =
-        maybe_build_local_openai_chat_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_openai_chat_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) = maybe_build_local_openai_chat_cross_format_stream_sync_response(
-        trace_id, decision, payload,
-    )? {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_openai_cli_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_openai_cli_cross_format_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_claude_cli_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_gemini_cli_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_claude_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) = maybe_build_local_claude_sync_response(trace_id, decision, payload)? {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_gemini_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) = maybe_build_local_gemini_sync_response(trace_id, decision, payload)? {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_openai_chat_cross_format_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_openai_cli_cross_format_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) = maybe_build_local_standard_chat_cross_format_stream_sync_response(
-        trace_id, decision, payload,
-    )? {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_standard_chat_cross_format_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) = maybe_build_local_standard_cli_cross_format_stream_sync_response(
-        trace_id, decision, payload,
-    )? {
-        return Ok(Some(response));
-    }
-    if let Some(response) =
-        maybe_build_local_standard_cli_cross_format_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    Ok(None)
-}
-
-fn maybe_build_local_standard_chat_cross_format_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_chat_sync_finalize" | "claude_chat_sync_finalize" | "gemini_chat_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
     let Some(report_context) = payload.report_context.as_ref() else {
         return Ok(None);
     };
-    let provider_api_format = report_context
+    if !local_finalize_allows_envelope(report_context) {
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if !local_finalize_allows_envelope(report_context)
-        || sync_chat_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
-    {
         return Ok(None);
     }
-    let Some(body_base64) = payload.body_base64.as_deref() else {
+    let Some(product) = maybe_build_standard_sync_finalize_product_from_normalized_payload(
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let Some(aggregated) =
-        aggregate_standard_chat_stream_sync_response(&body_bytes, &provider_api_format)
-    else {
-        return Ok(None);
-    };
-    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
-        return Ok(None);
-    };
-    let Some(converted) = convert_standard_chat_response(
-        &aggregated,
-        &provider_api_format,
-        &client_api_format,
-        report_context,
-    ) else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, aggregated,
-    )?))
-}
-
-fn maybe_build_local_standard_chat_cross_format_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
         payload.report_kind.as_str(),
-        "openai_chat_sync_finalize" | "claude_chat_sync_finalize" | "gemini_chat_sync_finalize"
+        payload.status_code,
-    ) || payload.status_code >= 400
+        Some(report_context),
-    {
+        payload.body_json.as_ref(),
-        return Ok(None);
+        payload.body_base64.as_deref(),
-    }
+    )
-
+    .map_err(GatewayError::from)?
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if !local_finalize_allows_envelope(report_context)
-        || sync_chat_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
-    {
-        return Ok(None);
-    }
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
     else {
         return Ok(None);
     };
-    let Some(converted) = convert_standard_chat_response(
-        &body_json,
-        &provider_api_format,
-        &client_api_format,
-        report_context,
-    ) else {
-        return Ok(None);
-    };
 
-    Ok(Some(build_local_success_outcome_with_conversion_report(
+    match product {
-        trace_id, decision, payload, converted, body_json,
+        StandardSyncFinalizeNormalizedProduct::SuccessBody(body_json) => {
-    )?))
+            let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)?
-}
-
-fn maybe_build_local_standard_cli_cross_format_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize"
-            | "openai_compact_sync_finalize"
-            | "claude_cli_sync_finalize"
-            | "gemini_cli_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if !local_finalize_allows_envelope(report_context)
-        || sync_cli_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
-    {
-        return Ok(None);
-    }
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let Some(aggregated) =
-        aggregate_standard_cli_stream_sync_response(&body_bytes, &provider_api_format)
             else {
                 return Ok(None);
             };
-    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
+            Ok(Some(build_local_success_outcome(
-        return Ok(None);
+                trace_id, decision, payload, body_json,
-    };
-    let Some(converted) = convert_standard_cli_response(
-        &aggregated,
-        &provider_api_format,
-        &client_api_format,
-        report_context,
-    ) else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, aggregated,
             )?))
         }
-
+        StandardSyncFinalizeNormalizedProduct::CrossFormat(product) => {
-fn maybe_build_local_standard_cli_cross_format_sync_response(
+            let Some(provider_body_json) =
-    trace_id: &str,
+                unwrap_local_finalize_response_value(product.provider_body_json, report_context)?
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize"
-            | "openai_compact_sync_finalize"
-            | "claude_cli_sync_finalize"
-            | "gemini_cli_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if !local_finalize_allows_envelope(report_context)
-        || sync_cli_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
-    {
-        return Ok(None);
-    }
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
             else {
                 return Ok(None);
             };
-    let Some(converted) = convert_standard_cli_response(
-        &body_json,
-        &provider_api_format,
-        &client_api_format,
-        report_context,
-    ) else {
-        return Ok(None);
-    };
-
             Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, body_json,
+                trace_id,
+                decision,
+                payload,
+                product.client_body_json,
+                provider_body_json,
             )?))
         }
+    }
+}
 
 #[cfg(test)]
 #[path = "../tests_sync.rs"]

apps/aether-gateway/src/ai_pipeline/finalize/mod.rs

@@ -1,10 +1,8 @@
 pub(crate) mod common;
 pub(crate) mod sse;
 pub(crate) mod standard;
-pub(crate) use crate::api::response::{
+pub(crate) use crate::ai_pipeline::execution_facade::maybe_build_local_sync_finalize_response;
-    build_client_response, build_client_response_from_parts,
+pub(crate) use crate::api::response::{build_client_response, build_client_response_from_parts};
-};
-pub(crate) use crate::execution_runtime::maybe_build_local_sync_finalize_response;
 pub(crate) use common::build_local_success_outcome;
 pub(crate) use internal::{
     maybe_build_stream_response_rewriter, maybe_build_sync_finalize_outcome,

apps/aether-gateway/src/ai_pipeline/finalize/sse.rs

@@ -1,38 +1,14 @@
 use serde_json::Value;
 
 use crate::GatewayError;
+use aether_ai_pipeline::finalize::{self, PipelineFinalizeError};
 
-pub(crate) fn map_claude_stop_reason(
+pub(crate) use finalize::sse::{encode_done_sse, map_claude_stop_reason};
-    stop_reason: Option<&str>,
-    has_tool_calls: bool,
-) -> Option<&'static str> {
-    let mapped = match stop_reason {
-        Some("end_turn") | Some("stop_sequence") => Some("stop"),
-        Some("max_tokens") => Some("length"),
-        Some("tool_use") => Some("tool_calls"),
-        Some("pause_turn") => Some("stop"),
-        _ => None,
-    };
-    if has_tool_calls && mapped.is_none_or(|value| value == "stop") {
-        Some("tool_calls")
-    } else {
-        mapped
-    }
-}
 
-pub(crate) fn encode_done_sse() -> Vec<u8> {
+fn map_error(err: PipelineFinalizeError) -> GatewayError {
-    b"data: [DONE]\n\n".to_vec()
+    err.into()
 }
 
 pub(crate) fn encode_json_sse(event: Option<&str>, value: &Value) -> Result<Vec<u8>, GatewayError> {
-    let mut out = Vec::new();
+    finalize::sse::encode_json_sse(event, value).map_err(map_error)
-    if let Some(event) = event.filter(|value| !value.trim().is_empty()) {
-        out.extend_from_slice(b"event: ");
-        out.extend_from_slice(event.as_bytes());
-        out.push(b'\n');
-    }
-    out.extend_from_slice(b"data: ");
-    out.extend(serde_json::to_vec(value).map_err(|err| GatewayError::Internal(err.to_string()))?);
-    out.extend_from_slice(b"\n\n");
-    Ok(out)
 }

apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/mod.rs

@@ -1,9 +1,6 @@
 pub(super) mod stream;
-pub(super) mod sync;
 
-pub(crate) use sync::{
+pub(crate) use crate::ai_pipeline::conversion::response::{
-    aggregate_claude_stream_sync_response, convert_claude_chat_response_to_openai_chat,
+    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
-    convert_claude_cli_response_to_openai_cli, convert_openai_chat_response_to_claude_chat,
+    convert_openai_chat_response_to_claude_chat,
-    maybe_build_local_claude_cli_stream_sync_response,
-    maybe_build_local_claude_stream_sync_response, maybe_build_local_claude_sync_response,
 };

apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/stream.rs

@@ -1,531 +1,3 @@
-use std::collections::BTreeMap;
+pub(crate) use aether_ai_pipeline::finalize::standard::claude::stream::{
-
+    ClaudeClientEmitter, ClaudeProviderState,
-use serde_json::{json, Map, Value};
-
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, canonicalize_tool_arguments,
 };
-use crate::ai_pipeline::finalize::sse::{
-    encode_done_sse, encode_json_sse, map_claude_stop_reason,
-};
-use crate::GatewayError;
-
-use crate::ai_pipeline::finalize::standard::stream::common::*;
-
-#[derive(Default)]
-struct ClaudeProviderToolState {
-    call_id: String,
-    name: String,
-    started_emitted: bool,
-}
-
-#[derive(Default)]
-pub(crate) struct ClaudeProviderState {
-    message_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-    tool_calls: BTreeMap<usize, ClaudeProviderToolState>,
-}
-
-impl ClaudeProviderState {
-    fn identity(&self, report_context: &Value) -> (String, String) {
-        resolve_identity(
-            self.message_id.as_deref(),
-            self.model.as_deref(),
-            report_context,
-            "msg-local-stream",
-        )
-    }
-
-    fn ensure_started(&mut self, report_context: &Value, out: &mut Vec<CanonicalStreamFrame>) {
-        if self.started {
-            return;
-        }
-        let (id, model) = self.identity(report_context);
-        out.push(CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Start,
-        });
-        self.started = true;
-    }
-
-    pub(crate) fn push_line(
-        &mut self,
-        report_context: &Value,
-        line: Vec<u8>,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        let Some(value) = decode_json_data_line(&line) else {
-            return Ok(Vec::new());
-        };
-        let Some(event_object) = value.as_object() else {
-            return Ok(Vec::new());
-        };
-        let mut out = Vec::new();
-        match event_object
-            .get("type")
-            .and_then(Value::as_str)
-            .unwrap_or_default()
-        {
-            "message_start" => {
-                if let Some(message) = event_object.get("message").and_then(Value::as_object) {
-                    self.message_id = message
-                        .get("id")
-                        .and_then(Value::as_str)
-                        .map(ToOwned::to_owned);
-                    self.model = message
-                        .get("model")
-                        .and_then(Value::as_str)
-                        .map(ToOwned::to_owned);
-                }
-                self.ensure_started(report_context, &mut out);
-            }
-            "content_block_delta" => {
-                let index = event_object
-                    .get("index")
-                    .and_then(Value::as_u64)
-                    .map(|value| value as usize)
-                    .unwrap_or(0);
-                let Some(delta) = event_object.get("delta").and_then(Value::as_object) else {
-                    return Ok(out);
-                };
-                match delta
-                    .get("type")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default()
-                {
-                    "text_delta" => {
-                        let Some(piece) = delta.get("text").and_then(Value::as_str) else {
-                            return Ok(out);
-                        };
-                        if piece.is_empty() {
-                            return Ok(out);
-                        }
-                        self.ensure_started(report_context, &mut out);
-                        let (id, model) = self.identity(report_context);
-                        out.push(CanonicalStreamFrame {
-                            id,
-                            model,
-                            event: CanonicalStreamEvent::TextDelta(piece.to_string()),
-                        });
-                    }
-                    "input_json_delta" => {
-                        let Some(partial_json) = delta.get("partial_json").and_then(Value::as_str)
-                        else {
-                            return Ok(out);
-                        };
-                        if partial_json.is_empty() {
-                            return Ok(out);
-                        }
-                        self.ensure_started(report_context, &mut out);
-                        let (id, model) = self.identity(report_context);
-                        let tool_state = self.tool_calls.entry(index).or_default();
-                        if !tool_state.started_emitted {
-                            out.push(CanonicalStreamFrame {
-                                id: id.clone(),
-                                model: model.clone(),
-                                event: CanonicalStreamEvent::ToolCallStart {
-                                    index,
-                                    call_id: if tool_state.call_id.is_empty() {
-                                        build_generated_tool_call_id(index)
-                                    } else {
-                                        tool_state.call_id.clone()
-                                    },
-                                    name: if tool_state.name.is_empty() {
-                                        "unknown".to_string()
-                                    } else {
-                                        tool_state.name.clone()
-                                    },
-                                },
-                            });
-                            tool_state.started_emitted = true;
-                        }
-                        out.push(CanonicalStreamFrame {
-                            id,
-                            model,
-                            event: CanonicalStreamEvent::ToolCallArgumentsDelta {
-                                index,
-                                arguments: partial_json.to_string(),
-                            },
-                        });
-                    }
-                    _ => {}
-                }
-            }
-            "content_block_start" => {
-                let index = event_object
-                    .get("index")
-                    .and_then(Value::as_u64)
-                    .map(|value| value as usize)
-                    .unwrap_or(0);
-                let Some(block) = event_object.get("content_block").and_then(Value::as_object)
-                else {
-                    return Ok(out);
-                };
-                let block_type = block
-                    .get("type")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default();
-                if block_type == "text" {
-                    let Some(text) = block.get("text").and_then(Value::as_str) else {
-                        return Ok(out);
-                    };
-                    if text.is_empty() {
-                        return Ok(out);
-                    }
-                    self.ensure_started(report_context, &mut out);
-                    let (id, model) = self.identity(report_context);
-                    out.push(CanonicalStreamFrame {
-                        id,
-                        model,
-                        event: CanonicalStreamEvent::TextDelta(text.to_string()),
-                    });
-                    return Ok(out);
-                }
-                if block_type != "tool_use" {
-                    return Ok(out);
-                }
-                self.ensure_started(report_context, &mut out);
-                let (id, model) = self.identity(report_context);
-                let tool_state = self.tool_calls.entry(index).or_default();
-                tool_state.call_id = block
-                    .get("id")
-                    .and_then(Value::as_str)
-                    .unwrap_or_else(|| tool_state.call_id.as_str())
-                    .to_string();
-                tool_state.name = block
-                    .get("name")
-                    .and_then(Value::as_str)
-                    .unwrap_or_else(|| tool_state.name.as_str())
-                    .to_string();
-                if !tool_state.started_emitted {
-                    out.push(CanonicalStreamFrame {
-                        id: id.clone(),
-                        model: model.clone(),
-                        event: CanonicalStreamEvent::ToolCallStart {
-                            index,
-                            call_id: if tool_state.call_id.is_empty() {
-                                build_generated_tool_call_id(index)
-                            } else {
-                                tool_state.call_id.clone()
-                            },
-                            name: if tool_state.name.is_empty() {
-                                "unknown".to_string()
-                            } else {
-                                tool_state.name.clone()
-                            },
-                        },
-                    });
-                    tool_state.started_emitted = true;
-                }
-                let arguments = canonicalize_tool_arguments(block.get("input").cloned());
-                if !arguments.is_empty() && arguments != "{}" {
-                    out.push(CanonicalStreamFrame {
-                        id,
-                        model,
-                        event: CanonicalStreamEvent::ToolCallArgumentsDelta { index, arguments },
-                    });
-                }
-            }
-            "message_delta" => {
-                self.ensure_started(report_context, &mut out);
-                let Some(delta) = event_object.get("delta").and_then(Value::as_object) else {
-                    return Ok(out);
-                };
-                let finish_reason = map_claude_stop_reason(
-                    delta.get("stop_reason").and_then(Value::as_str),
-                    delta.get("stop_reason").and_then(Value::as_str) == Some("tool_use"),
-                )
-                .map(ToOwned::to_owned);
-                let (id, model) = self.identity(report_context);
-                out.push(CanonicalStreamFrame {
-                    id,
-                    model,
-                    event: CanonicalStreamEvent::Finish {
-                        finish_reason,
-                        usage: canonical_usage_from_claude_usage(event_object.get("usage")),
-                    },
-                });
-                self.finished = true;
-            }
-            _ => {}
-        }
-        Ok(out)
-    }
-
-    pub(crate) fn finish(
-        &mut self,
-        report_context: &Value,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        self.finished = true;
-        let (id, model) = self.identity(report_context);
-        Ok(vec![CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Finish {
-                finish_reason: None,
-                usage: None,
-            },
-        }])
-    }
-}
-
-enum ClaudeOpenBlock {
-    Text {
-        block_index: usize,
-    },
-    Tool {
-        tool_index: usize,
-        block_index: usize,
-    },
-}
-
-#[derive(Default)]
-pub(crate) struct ClaudeClientEmitter {
-    message_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-    next_block_index: usize,
-    open_block: Option<ClaudeOpenBlock>,
-    tool_block_indices: BTreeMap<usize, usize>,
-}
-
-impl ClaudeClientEmitter {
-    fn update_identity(&mut self, frame: &CanonicalStreamFrame) {
-        self.message_id = Some(frame.id.clone());
-        self.model = Some(frame.model.clone());
-    }
-
-    fn ensure_started(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if self.started {
-            return Ok(Vec::new());
-        }
-        self.started = true;
-        encode_json_sse(
-            Some("message_start"),
-            &json!({
-                "type": "message_start",
-                "message": {
-                    "id": self.message_id.as_deref().unwrap_or("msg-local-stream"),
-                    "type": "message",
-                    "role": "assistant",
-                    "model": self.model.as_deref().unwrap_or("unknown"),
-                    "content": [],
-                    "stop_reason": Value::Null,
-                    "stop_sequence": Value::Null,
-                }
-            }),
-        )
-    }
-
-    fn close_open_block(&mut self) -> Result<Vec<u8>, GatewayError> {
-        let Some(open_block) = self.open_block.take() else {
-            return Ok(Vec::new());
-        };
-        let block_index = match open_block {
-            ClaudeOpenBlock::Text { block_index } => block_index,
-            ClaudeOpenBlock::Tool { block_index, .. } => block_index,
-        };
-        encode_json_sse(
-            Some("content_block_stop"),
-            &json!({
-                "type": "content_block_stop",
-                "index": block_index,
-            }),
-        )
-    }
-
-    fn ensure_text_block(&mut self) -> Result<Vec<u8>, GatewayError> {
-        let mut out = Vec::new();
-        if let Some(ClaudeOpenBlock::Text { .. }) = self.open_block {
-            return Ok(out);
-        }
-        out.extend(self.close_open_block()?);
-        let block_index = self.next_block_index;
-        self.next_block_index += 1;
-        self.open_block = Some(ClaudeOpenBlock::Text { block_index });
-        out.extend(encode_json_sse(
-            Some("content_block_start"),
-            &json!({
-                "type": "content_block_start",
-                "index": block_index,
-                "content_block": {
-                    "type": "text",
-                    "text": "",
-                }
-            }),
-        )?);
-        Ok(out)
-    }
-
-    fn ensure_tool_block(
-        &mut self,
-        tool_index: usize,
-        call_id: &str,
-        name: &str,
-    ) -> Result<Vec<u8>, GatewayError> {
-        let mut out = Vec::new();
-        if let Some(ClaudeOpenBlock::Tool {
-            tool_index: current_tool_index,
-            ..
-        }) = self.open_block
-        {
-            if current_tool_index == tool_index {
-                return Ok(out);
-            }
-        }
-        out.extend(self.close_open_block()?);
-        let block_index = self
-            .tool_block_indices
-            .get(&tool_index)
-            .copied()
-            .unwrap_or_else(|| {
-                let block_index = self.next_block_index;
-                self.next_block_index += 1;
-                self.tool_block_indices.insert(tool_index, block_index);
-                block_index
-            });
-        self.open_block = Some(ClaudeOpenBlock::Tool {
-            tool_index,
-            block_index,
-        });
-        out.extend(encode_json_sse(
-            Some("content_block_start"),
-            &json!({
-                "type": "content_block_start",
-                "index": block_index,
-                "content_block": {
-                    "type": "tool_use",
-                    "id": call_id,
-                    "name": name,
-                    "input": {},
-                }
-            }),
-        )?);
-        Ok(out)
-    }
-
-    pub(crate) fn emit(&mut self, frame: CanonicalStreamFrame) -> Result<Vec<u8>, GatewayError> {
-        self.update_identity(&frame);
-        match frame.event {
-            CanonicalStreamEvent::Start => self.ensure_started(),
-            CanonicalStreamEvent::TextDelta(text) => {
-                let mut out = self.ensure_started()?;
-                out.extend(self.ensure_text_block()?);
-                let block_index = match self.open_block {
-                    Some(ClaudeOpenBlock::Text { block_index }) => block_index,
-                    _ => return Ok(out),
-                };
-                out.extend(encode_json_sse(
-                    Some("content_block_delta"),
-                    &json!({
-                        "type": "content_block_delta",
-                        "index": block_index,
-                        "delta": {
-                            "type": "text_delta",
-                            "text": text,
-                        }
-                    }),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::ToolCallStart {
-                index,
-                call_id,
-                name,
-            } => {
-                let mut out = self.ensure_started()?;
-                out.extend(self.ensure_tool_block(index, &call_id, &name)?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::ToolCallArgumentsDelta { index, arguments } => {
-                let mut out = self.ensure_started()?;
-                let call_id = format!("tool_{index}");
-                out.extend(self.ensure_tool_block(index, &call_id, "unknown")?);
-                let block_index = match self.open_block {
-                    Some(ClaudeOpenBlock::Tool { block_index, .. }) => block_index,
-                    _ => return Ok(out),
-                };
-                out.extend(encode_json_sse(
-                    Some("content_block_delta"),
-                    &json!({
-                        "type": "content_block_delta",
-                        "index": block_index,
-                        "delta": {
-                            "type": "input_json_delta",
-                            "partial_json": arguments,
-                        }
-                    }),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::Finish {
-                finish_reason,
-                usage,
-            } => {
-                if self.finished {
-                    return Ok(Vec::new());
-                }
-                let mut out = self.ensure_started()?;
-                out.extend(self.close_open_block()?);
-                let mut payload = Map::new();
-                payload.insert(
-                    "type".to_string(),
-                    Value::String("message_delta".to_string()),
-                );
-                payload.insert(
-                    "delta".to_string(),
-                    json!({
-                        "stop_reason": map_openai_finish_reason_to_claude(
-                            finish_reason.as_deref()
-                        ),
-                        "stop_sequence": Value::Null,
-                    }),
-                );
-                if let Some(usage) = usage {
-                    payload.insert(
-                        "usage".to_string(),
-                        json!({
-                            "input_tokens": usage.input_tokens,
-                            "output_tokens": usage.output_tokens,
-                        }),
-                    );
-                }
-                out.extend(encode_json_sse(
-                    Some("message_delta"),
-                    &Value::Object(payload),
-                )?);
-                out.extend(encode_json_sse(
-                    Some("message_stop"),
-                    &json!({
-                        "type": "message_stop",
-                    }),
-                )?);
-                self.finished = true;
-                Ok(out)
-            }
-        }
-    }
-
-    pub(crate) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        self.emit(CanonicalStreamFrame {
-            id: self
-                .message_id
-                .clone()
-                .unwrap_or_else(|| "msg-local-stream".to_string()),
-            model: self.model.clone().unwrap_or_else(|| "unknown".to_string()),
-            event: CanonicalStreamEvent::Finish {
-                finish_reason: None,
-                usage: None,
-            },
-        })
-    }
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/sync/chat.rs

@@ -1,490 +0,0 @@
-use base64::Engine as _;
-use std::collections::BTreeMap;
-
-use serde_json::{json, Map, Value};
-
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
-    LocalCoreSyncFinalizeOutcome,
-};
-use crate::control::GatewayControlDecision;
-use crate::{usage::GatewaySyncReportRequest, GatewayError};
-
-#[derive(Debug, Default)]
-struct ClaudeContentBlockState {
-    object: Map<String, Value>,
-    text: String,
-    partial_json: String,
-}
-
-pub(crate) fn maybe_build_local_claude_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "claude_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "claude:chat"
-        || client_api_format != "claude:chat"
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let body_json = match aggregate_claude_stream_sync_response(&body_bytes) {
-        Some(body_json) => body_json,
-        None => return Ok(None),
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn maybe_build_local_claude_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "claude_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "claude:chat"
-        || client_api_format != "claude:chat"
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn convert_claude_chat_response_to_openai_chat(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let content = body.get("content")?.as_array()?;
-    let mut text = String::new();
-    let mut tool_calls = Vec::new();
-    for (index, block) in content.iter().enumerate() {
-        let block = block.as_object()?;
-        match block.get("type")?.as_str()? {
-            "text" => {
-                text.push_str(block.get("text")?.as_str()?);
-            }
-            "tool_use" => {
-                let tool_name = block.get("name")?.as_str()?;
-                let tool_id = block
-                    .get("id")
-                    .and_then(Value::as_str)
-                    .filter(|value| !value.is_empty())
-                    .map(ToOwned::to_owned)
-                    .unwrap_or_else(|| build_generated_tool_call_id(index));
-                let arguments = canonicalize_tool_arguments(block.get("input").cloned());
-                tool_calls.push(json!({
-                    "id": tool_id,
-                    "type": "function",
-                    "function": {
-                        "name": tool_name,
-                        "arguments": arguments,
-                    }
-                }));
-            }
-            _ => return None,
-        }
-    }
-    let mut finish_reason = match body.get("stop_reason").and_then(Value::as_str) {
-        Some("end_turn") | Some("stop_sequence") => Some("stop"),
-        Some("max_tokens") => Some("length"),
-        Some("tool_use") => Some("tool_calls"),
-        Some(other) if !other.is_empty() => Some(other),
-        _ => None,
-    };
-    if !tool_calls.is_empty() && finish_reason.is_none_or(|reason| reason == "stop") {
-        finish_reason = Some("tool_calls");
-    }
-    let usage = body.get("usage").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("input_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let completion_tokens = usage
-        .and_then(|value| value.get("output_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = prompt_tokens + completion_tokens;
-    let model = body
-        .get("model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let id = body
-        .get("id")
-        .and_then(Value::as_str)
-        .unwrap_or("chatcmpl-local-finalize");
-    let message_content = if text.is_empty() && !tool_calls.is_empty() {
-        Value::Null
-    } else {
-        Value::String(text)
-    };
-    let mut message = Map::new();
-    message.insert("role".to_string(), Value::String("assistant".to_string()));
-    message.insert("content".to_string(), message_content);
-    if !tool_calls.is_empty() {
-        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
-    }
-    Some(json!({
-        "id": id,
-        "object": "chat.completion",
-        "model": model,
-        "choices": [{
-            "index": 0,
-            "message": Value::Object(message),
-            "finish_reason": finish_reason,
-        }],
-        "usage": {
-            "prompt_tokens": prompt_tokens,
-            "completion_tokens": completion_tokens,
-            "total_tokens": total_tokens,
-        }
-    }))
-}
-
-pub(crate) fn convert_openai_chat_response_to_claude_chat(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let choices = body.get("choices")?.as_array()?;
-    let first_choice = choices.first()?.as_object()?;
-    let message = first_choice.get("message")?.as_object()?;
-    let mut content = Vec::new();
-
-    if let Some(text) = extract_openai_assistant_text(message.get("content")) {
-        if !text.trim().is_empty() {
-            content.push(json!({
-                "type": "text",
-                "text": text,
-            }));
-        }
-    }
-    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
-        for (index, tool_call) in tool_call_values.iter().enumerate() {
-            let tool_call = tool_call.as_object()?;
-            let function = tool_call.get("function")?.as_object()?;
-            let tool_name = function
-                .get("name")
-                .and_then(Value::as_str)
-                .map(str::trim)
-                .filter(|value| !value.is_empty())?;
-            let tool_id = tool_call
-                .get("id")
-                .and_then(Value::as_str)
-                .filter(|value| !value.is_empty())
-                .map(ToOwned::to_owned)
-                .unwrap_or_else(|| build_generated_tool_call_id(index));
-            let input = parse_openai_function_arguments(function.get("arguments"))?;
-            content.push(json!({
-                "type": "tool_use",
-                "id": tool_id,
-                "name": tool_name,
-                "input": input,
-            }));
-        }
-    }
-    if content.is_empty() {
-        content.push(json!({
-            "type": "text",
-            "text": "",
-        }));
-    }
-
-    let stop_reason = match first_choice.get("finish_reason").and_then(Value::as_str) {
-        Some("stop") | None => "end_turn",
-        Some("length") => "max_tokens",
-        Some("tool_calls") | Some("function_call") => "tool_use",
-        Some("content_filter") => "content_filtered",
-        Some(other) => other,
-    };
-    let usage = body.get("usage").and_then(Value::as_object);
-    let input_tokens = usage
-        .and_then(|value| value.get("prompt_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .and_then(|value| value.get("completion_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let model = body
-        .get("model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let id = body
-        .get("id")
-        .and_then(Value::as_str)
-        .unwrap_or("msg-local-finalize");
-
-    Some(json!({
-        "id": id,
-        "type": "message",
-        "role": "assistant",
-        "model": model,
-        "content": content,
-        "stop_reason": stop_reason,
-        "usage": {
-            "input_tokens": input_tokens,
-            "output_tokens": output_tokens,
-        }
-    }))
-}
-
-fn extract_openai_assistant_text(content: Option<&Value>) -> Option<String> {
-    match content? {
-        Value::Null => Some(String::new()),
-        Value::String(text) => Some(text.clone()),
-        Value::Array(parts) => {
-            let mut text = String::new();
-            for part in parts {
-                let part = part.as_object()?;
-                let part_type = part
-                    .get("type")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default()
-                    .trim()
-                    .to_ascii_lowercase();
-                if matches!(part_type.as_str(), "text" | "output_text") {
-                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
-                        text.push_str(piece);
-                    }
-                }
-            }
-            Some(text)
-        }
-        _ => None,
-    }
-}
-
-fn parse_openai_function_arguments(arguments: Option<&Value>) -> Option<Value> {
-    match arguments.cloned().unwrap_or(Value::Object(Map::new())) {
-        Value::String(text) => serde_json::from_str(&text)
-            .ok()
-            .or(Some(Value::String(text))),
-        other => Some(other),
-    }
-}
-
-pub(crate) fn aggregate_claude_stream_sync_response(body: &[u8]) -> Option<Value> {
-    let events = parse_stream_json_events(body)?;
-    if events.is_empty() {
-        return None;
-    }
-
-    let mut message_object: Option<Map<String, Value>> = None;
-    let mut content_blocks: BTreeMap<usize, ClaudeContentBlockState> = BTreeMap::new();
-    let mut usage: Option<Value> = None;
-    let mut saw_message_start = false;
-
-    for event in events {
-        let event_object = event.as_object()?;
-        let event_type = event_object
-            .get("type")
-            .and_then(Value::as_str)
-            .unwrap_or_default();
-
-        match event_type {
-            "message_start" => {
-                let mut message = event_object.get("message")?.as_object()?.clone();
-                usage = message.remove("usage");
-                message_object = Some(message);
-                saw_message_start = true;
-            }
-            "content_block_start" => {
-                let index = event_object
-                    .get("index")
-                    .and_then(Value::as_u64)
-                    .map(|value| value as usize)
-                    .unwrap_or(0);
-                let object = event_object
-                    .get("content_block")
-                    .and_then(Value::as_object)
-                    .cloned()
-                    .unwrap_or_default();
-                content_blocks.insert(
-                    index,
-                    ClaudeContentBlockState {
-                        object,
-                        ..Default::default()
-                    },
-                );
-            }
-            "content_block_delta" => {
-                let index = event_object
-                    .get("index")
-                    .and_then(Value::as_u64)
-                    .map(|value| value as usize)
-                    .unwrap_or(0);
-                let state = content_blocks.entry(index).or_default();
-                let Some(delta) = event_object.get("delta").and_then(Value::as_object) else {
-                    continue;
-                };
-                match delta
-                    .get("type")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default()
-                {
-                    "text_delta" => {
-                        if let Some(text) = delta.get("text").and_then(Value::as_str) {
-                            state.text.push_str(text);
-                        }
-                    }
-                    "input_json_delta" => {
-                        if let Some(partial_json) =
-                            delta.get("partial_json").and_then(Value::as_str)
-                        {
-                            state.partial_json.push_str(partial_json);
-                        }
-                    }
-                    _ => {}
-                }
-            }
-            "message_delta" => {
-                if let Some(message) = message_object.as_mut() {
-                    if let Some(delta) = event_object.get("delta").and_then(Value::as_object) {
-                        if let Some(stop_reason) = delta.get("stop_reason") {
-                            message.insert("stop_reason".to_string(), stop_reason.clone());
-                        }
-                        if let Some(stop_sequence) = delta.get("stop_sequence") {
-                            message.insert("stop_sequence".to_string(), stop_sequence.clone());
-                        }
-                    }
-                }
-                if let Some(delta_usage) = event_object.get("usage") {
-                    usage = Some(delta_usage.clone());
-                }
-            }
-            "message_stop" => {}
-            _ => {}
-        }
-    }
-
-    if !saw_message_start {
-        return None;
-    }
-
-    let mut message = message_object?;
-    let mut content = Vec::with_capacity(content_blocks.len());
-    for (_index, state) in content_blocks {
-        let mut block = state.object;
-        let block_type = block
-            .get("type")
-            .and_then(Value::as_str)
-            .unwrap_or("text")
-            .to_string();
-        match block_type.as_str() {
-            "text" => {
-                block.insert(
-                    "text".to_string(),
-                    Value::String(if state.text.is_empty() {
-                        block
-                            .get("text")
-                            .and_then(Value::as_str)
-                            .unwrap_or_default()
-                            .to_string()
-                    } else {
-                        state.text
-                    }),
-                );
-            }
-            "tool_use" => {
-                if !state.partial_json.is_empty() {
-                    let input = serde_json::from_str::<Value>(&state.partial_json)
-                        .unwrap_or(Value::String(state.partial_json));
-                    block.insert("input".to_string(), input);
-                }
-            }
-            _ => {
-                if !state.text.is_empty() {
-                    block.insert("text".to_string(), Value::String(state.text));
-                }
-            }
-        }
-        content.push(Value::Object(block));
-    }
-    message.insert("content".to_string(), Value::Array(content));
-    if let Some(usage_value) = usage {
-        message.insert("usage".to_string(), usage_value);
-    }
-
-    Some(Value::Object(message))
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/sync/cli.rs

@@ -1,142 +0,0 @@
-use base64::Engine as _;
-
-use super::chat::aggregate_claude_stream_sync_response;
-use serde_json::{json, Value};
-
-use crate::ai_pipeline::conversion::response::build_openai_cli_response;
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
-    LocalCoreSyncFinalizeOutcome,
-};
-use crate::control::GatewayControlDecision;
-use crate::{usage::GatewaySyncReportRequest, GatewayError};
-
-pub(crate) fn maybe_build_local_claude_cli_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "claude_cli_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "claude:cli" || client_api_format != "claude:cli" || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let body_json =
-        match aggregate_provider_claude_cli_stream_sync_response(&body_bytes, report_context)? {
-            Some(body_json) => body_json,
-            None => return Ok(None),
-        };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-fn aggregate_provider_claude_cli_stream_sync_response(
-    body_bytes: &[u8],
-    _report_context: &Value,
-) -> Result<Option<Value>, GatewayError> {
-    Ok(aggregate_claude_stream_sync_response(body_bytes))
-}
-
-pub(crate) fn convert_claude_cli_response_to_openai_cli(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let content = body.get("content")?.as_array()?;
-    let mut text = String::new();
-    let mut function_calls = Vec::new();
-    for (index, block) in content.iter().enumerate() {
-        let block = block.as_object()?;
-        match block.get("type")?.as_str()? {
-            "text" => {
-                text.push_str(block.get("text")?.as_str()?);
-            }
-            "tool_use" => {
-                let tool_name = block.get("name")?.as_str()?;
-                let call_id = block
-                    .get("id")
-                    .and_then(Value::as_str)
-                    .filter(|value| !value.is_empty())
-                    .map(ToOwned::to_owned)
-                    .unwrap_or_else(|| build_generated_tool_call_id(index));
-                let arguments = canonicalize_tool_arguments(block.get("input").cloned());
-                function_calls.push(json!({
-                    "type": "function_call",
-                    "call_id": call_id,
-                    "name": tool_name,
-                    "arguments": arguments,
-                }));
-            }
-            _ => return None,
-        }
-    }
-
-    let usage = body.get("usage").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("input_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .and_then(|value| value.get("output_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = prompt_tokens + output_tokens;
-    let model = body
-        .get("model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let response_id = body
-        .get("id")
-        .and_then(Value::as_str)
-        .unwrap_or("resp-local-finalize");
-
-    Some(build_openai_cli_response(
-        response_id,
-        model,
-        &text,
-        function_calls,
-        prompt_tokens,
-        output_tokens,
-        total_tokens,
-    ))
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/sync/mod.rs

@@ -1,11 +0,0 @@
-mod chat;
-mod cli;
-
-pub(crate) use chat::{
-    aggregate_claude_stream_sync_response, convert_claude_chat_response_to_openai_chat,
-    convert_openai_chat_response_to_claude_chat, maybe_build_local_claude_stream_sync_response,
-    maybe_build_local_claude_sync_response,
-};
-pub(crate) use cli::{
-    convert_claude_cli_response_to_openai_cli, maybe_build_local_claude_cli_stream_sync_response,
-};

apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/mod.rs

@@ -1,9 +1,6 @@
 pub(super) mod stream;
-pub(super) mod sync;
 
-pub(crate) use sync::{
+pub(crate) use crate::ai_pipeline::conversion::response::{
-    aggregate_gemini_stream_sync_response, convert_gemini_chat_response_to_openai_chat,
+    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
-    convert_gemini_cli_response_to_openai_cli, convert_openai_chat_response_to_gemini_chat,
+    convert_openai_chat_response_to_gemini_chat,
-    maybe_build_local_gemini_cli_stream_sync_response,
-    maybe_build_local_gemini_stream_sync_response, maybe_build_local_gemini_sync_response,
 };

apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/stream.rs

@@ -1,404 +1,3 @@
-use std::collections::BTreeMap;
+pub(crate) use aether_ai_pipeline::finalize::standard::gemini::stream::{
-
+    GeminiClientEmitter, GeminiProviderState,
-use serde_json::{json, Map, Value};
-
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, canonicalize_tool_arguments,
 };
-use crate::ai_pipeline::finalize::sse::encode_json_sse;
-use crate::GatewayError;
-
-use crate::ai_pipeline::finalize::standard::stream::common::*;
-
-#[derive(Default)]
-struct GeminiProviderToolState {
-    call_id: String,
-    name: String,
-    arguments: String,
-    started_emitted: bool,
-}
-
-#[derive(Default)]
-pub(crate) struct GeminiProviderState {
-    response_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-    text_parts: BTreeMap<usize, String>,
-    tool_calls: BTreeMap<usize, GeminiProviderToolState>,
-}
-
-impl GeminiProviderState {
-    fn identity(&self, report_context: &Value) -> (String, String) {
-        resolve_identity(
-            self.response_id.as_deref(),
-            self.model.as_deref(),
-            report_context,
-            "resp-local-stream",
-        )
-    }
-
-    fn ensure_started(&mut self, report_context: &Value, out: &mut Vec<CanonicalStreamFrame>) {
-        if self.started {
-            return;
-        }
-        let (id, model) = self.identity(report_context);
-        out.push(CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Start,
-        });
-        self.started = true;
-    }
-
-    pub(crate) fn push_line(
-        &mut self,
-        report_context: &Value,
-        line: Vec<u8>,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        let Some(value) = decode_json_data_line(&line) else {
-            return Ok(Vec::new());
-        };
-        let Some(raw_event_object) = value.as_object() else {
-            return Ok(Vec::new());
-        };
-        if let Some(id) = raw_event_object.get("responseId").and_then(Value::as_str) {
-            self.response_id = Some(id.to_string());
-        }
-        let event_object = raw_event_object
-            .get("response")
-            .and_then(Value::as_object)
-            .filter(|response| response.contains_key("candidates"))
-            .unwrap_or(raw_event_object);
-        if let Some(id) = event_object.get("responseId").and_then(Value::as_str) {
-            self.response_id = Some(id.to_string());
-        }
-        if let Some(version) = event_object.get("modelVersion").and_then(Value::as_str) {
-            self.model = Some(version.to_string());
-        }
-
-        let mut out = Vec::new();
-        let Some(candidates) = event_object.get("candidates").and_then(Value::as_array) else {
-            return Ok(out);
-        };
-
-        for candidate in candidates {
-            let Some(candidate_object) = candidate.as_object() else {
-                continue;
-            };
-            let Some(content) = candidate_object.get("content").and_then(Value::as_object) else {
-                continue;
-            };
-            let Some(parts) = content.get("parts").and_then(Value::as_array) else {
-                continue;
-            };
-            if !parts.is_empty() {
-                self.ensure_started(report_context, &mut out);
-            }
-            let (id, model) = self.identity(report_context);
-            for (index, part) in parts.iter().enumerate() {
-                let Some(part_object) = part.as_object() else {
-                    continue;
-                };
-                if let Some(text) = part_object.get("text").and_then(Value::as_str) {
-                    let previous = self.text_parts.entry(index).or_default();
-                    let delta = if text.starts_with(previous.as_str()) {
-                        text[previous.len()..].to_string()
-                    } else if previous.as_str() == text {
-                        String::new()
-                    } else {
-                        text.to_string()
-                    };
-                    *previous = text.to_string();
-                    if !delta.is_empty() {
-                        out.push(CanonicalStreamFrame {
-                            id: id.clone(),
-                            model: model.clone(),
-                            event: CanonicalStreamEvent::TextDelta(delta),
-                        });
-                    }
-                    continue;
-                }
-                let Some(function_call) =
-                    part_object.get("functionCall").and_then(Value::as_object)
-                else {
-                    continue;
-                };
-                let tool_state = self.tool_calls.entry(index).or_default();
-                tool_state.call_id = function_call
-                    .get("id")
-                    .and_then(Value::as_str)
-                    .unwrap_or_else(|| tool_state.call_id.as_str())
-                    .to_string();
-                tool_state.name = function_call
-                    .get("name")
-                    .and_then(Value::as_str)
-                    .unwrap_or_else(|| tool_state.name.as_str())
-                    .to_string();
-                if !tool_state.started_emitted {
-                    out.push(CanonicalStreamFrame {
-                        id: id.clone(),
-                        model: model.clone(),
-                        event: CanonicalStreamEvent::ToolCallStart {
-                            index,
-                            call_id: if tool_state.call_id.is_empty() {
-                                build_generated_tool_call_id(index)
-                            } else {
-                                tool_state.call_id.clone()
-                            },
-                            name: if tool_state.name.is_empty() {
-                                "unknown".to_string()
-                            } else {
-                                tool_state.name.clone()
-                            },
-                        },
-                    });
-                    tool_state.started_emitted = true;
-                }
-                let arguments = canonicalize_tool_arguments(function_call.get("args").cloned());
-                let delta = if arguments.starts_with(&tool_state.arguments) {
-                    arguments[tool_state.arguments.len()..].to_string()
-                } else if tool_state.arguments == arguments {
-                    String::new()
-                } else {
-                    arguments.clone()
-                };
-                tool_state.arguments = arguments;
-                if !delta.is_empty() {
-                    out.push(CanonicalStreamFrame {
-                        id: id.clone(),
-                        model: model.clone(),
-                        event: CanonicalStreamEvent::ToolCallArgumentsDelta {
-                            index,
-                            arguments: delta,
-                        },
-                    });
-                }
-            }
-            if let Some(finish_reason) =
-                candidate_object.get("finishReason").and_then(Value::as_str)
-            {
-                let has_tool_calls = !self.tool_calls.is_empty();
-                let mut finish_reason = normalize_openai_finish_reason(match finish_reason {
-                    "STOP" => Some("stop"),
-                    "MAX_TOKENS" => Some("length"),
-                    "SAFETY" => Some("content_filter"),
-                    other => Some(other),
-                });
-                if has_tool_calls && finish_reason.as_deref().is_none_or(|value| value == "stop") {
-                    finish_reason = Some("tool_calls".to_string());
-                }
-                out.push(CanonicalStreamFrame {
-                    id,
-                    model,
-                    event: CanonicalStreamEvent::Finish {
-                        finish_reason,
-                        usage: canonical_usage_from_gemini_usage(event_object.get("usageMetadata")),
-                    },
-                });
-                self.finished = true;
-            }
-        }
-
-        Ok(out)
-    }
-
-    pub(crate) fn finish(
-        &mut self,
-        report_context: &Value,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        self.finished = true;
-        let (id, model) = self.identity(report_context);
-        Ok(vec![CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Finish {
-                finish_reason: None,
-                usage: None,
-            },
-        }])
-    }
-}
-
-#[derive(Default)]
-struct GeminiClientToolState {
-    call_id: String,
-    name: String,
-    arguments: String,
-    emitted: bool,
-}
-
-#[derive(Default)]
-pub(crate) struct GeminiClientEmitter {
-    response_id: Option<String>,
-    model: Option<String>,
-    finished: bool,
-    tool_calls: BTreeMap<usize, GeminiClientToolState>,
-}
-
-impl GeminiClientEmitter {
-    fn update_identity(&mut self, frame: &CanonicalStreamFrame) {
-        self.response_id = Some(frame.id.clone());
-        self.model = Some(frame.model.clone());
-    }
-
-    fn emit_candidate(
-        &self,
-        parts: Vec<Value>,
-        finish_reason: Option<&str>,
-        usage: Option<CanonicalUsage>,
-    ) -> Result<Vec<u8>, GatewayError> {
-        let mut candidate = Map::new();
-        candidate.insert(
-            "content".to_string(),
-            json!({
-                "role": "model",
-                "parts": parts,
-            }),
-        );
-        candidate.insert("index".to_string(), Value::from(0_u64));
-        if let Some(finish_reason) = finish_reason {
-            candidate.insert(
-                "finishReason".to_string(),
-                Value::String(map_openai_finish_reason_to_gemini(Some(finish_reason)).to_string()),
-            );
-        }
-        let mut response = Map::new();
-        response.insert(
-            "responseId".to_string(),
-            Value::String(
-                self.response_id
-                    .clone()
-                    .unwrap_or_else(|| "resp-local-stream".to_string()),
-            ),
-        );
-        response.insert(
-            "modelVersion".to_string(),
-            Value::String(self.model.clone().unwrap_or_else(|| "unknown".to_string())),
-        );
-        response.insert(
-            "candidates".to_string(),
-            Value::Array(vec![Value::Object(candidate)]),
-        );
-        if let Some(usage) = usage {
-            response.insert(
-                "usageMetadata".to_string(),
-                json!({
-                    "promptTokenCount": usage.input_tokens,
-                    "candidatesTokenCount": usage.output_tokens,
-                    "totalTokenCount": usage.total_tokens,
-                }),
-            );
-        }
-        encode_json_sse(None, &Value::Object(response))
-    }
-
-    fn flush_pending_tool_calls(&mut self) -> Result<Vec<u8>, GatewayError> {
-        let mut out = Vec::new();
-        let mut pending = Vec::new();
-        for (index, tool_call) in &mut self.tool_calls {
-            if tool_call.emitted {
-                continue;
-            }
-            let args_value = parse_json_arguments_value(&tool_call.arguments)
-                .unwrap_or_else(|| Value::Object(Map::new()));
-            tool_call.emitted = true;
-            pending.push(json!({
-                "functionCall": {
-                    "id": if tool_call.call_id.is_empty() {
-                        build_generated_tool_call_id(*index)
-                    } else {
-                        tool_call.call_id.clone()
-                    },
-                    "name": if tool_call.name.is_empty() {
-                        "unknown".to_string()
-                    } else {
-                        tool_call.name.clone()
-                    },
-                    "args": args_value,
-                }
-            }));
-        }
-        for part in pending {
-            out.extend(self.emit_candidate(vec![part], None, None)?);
-        }
-        Ok(out)
-    }
-
-    pub(crate) fn emit(&mut self, frame: CanonicalStreamFrame) -> Result<Vec<u8>, GatewayError> {
-        self.update_identity(&frame);
-        match frame.event {
-            CanonicalStreamEvent::Start => Ok(Vec::new()),
-            CanonicalStreamEvent::TextDelta(text) => {
-                self.emit_candidate(vec![json!({ "text": text })], None, None)
-            }
-            CanonicalStreamEvent::ToolCallStart {
-                index,
-                call_id,
-                name,
-            } => {
-                let state = self.tool_calls.entry(index).or_default();
-                state.call_id = call_id;
-                state.name = name;
-                Ok(Vec::new())
-            }
-            CanonicalStreamEvent::ToolCallArgumentsDelta { index, arguments } => {
-                let emitted_part = {
-                    let state = self.tool_calls.entry(index).or_default();
-                    state.arguments.push_str(&arguments);
-                    if state.emitted {
-                        None
-                    } else {
-                        let args_value = parse_json_arguments_value(&state.arguments);
-                        args_value.map(|args_value| {
-                            state.emitted = true;
-                            json!({
-                                "functionCall": {
-                                    "id": if state.call_id.is_empty() {
-                                        build_generated_tool_call_id(index)
-                                    } else {
-                                        state.call_id.clone()
-                                    },
-                                    "name": if state.name.is_empty() {
-                                        "unknown".to_string()
-                                    } else {
-                                        state.name.clone()
-                                    },
-                                    "args": args_value,
-                                }
-                            })
-                        })
-                    }
-                };
-                let Some(part) = emitted_part else {
-                    return Ok(Vec::new());
-                };
-                self.emit_candidate(vec![part], None, None)
-            }
-            CanonicalStreamEvent::Finish {
-                finish_reason,
-                usage,
-            } => {
-                if self.finished {
-                    return Ok(Vec::new());
-                }
-                let mut out = self.flush_pending_tool_calls()?;
-                out.extend(self.emit_candidate(vec![], finish_reason.as_deref(), usage)?);
-                self.finished = true;
-                Ok(out)
-            }
-        }
-    }
-
-    pub(crate) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if self.finished {
-            return Ok(Vec::new());
-        }
-        let out = self.flush_pending_tool_calls()?;
-        self.finished = true;
-        Ok(out)
-    }
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/sync/chat.rs

@@ -1,439 +0,0 @@
-use base64::Engine as _;
-use std::collections::BTreeMap;
-
-use serde_json::{json, Map, Value};
-
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
-    LocalCoreSyncFinalizeOutcome,
-};
-use crate::control::GatewayControlDecision;
-use crate::{usage::GatewaySyncReportRequest, GatewayError};
-
-pub(crate) fn maybe_build_local_gemini_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "gemini_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "gemini:chat"
-        || client_api_format != "gemini:chat"
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let body_json = match aggregate_gemini_stream_sync_response(&body_bytes) {
-        Some(body_json) => body_json,
-        None => return Ok(None),
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn maybe_build_local_gemini_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "gemini_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "gemini:chat"
-        || client_api_format != "gemini:chat"
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn convert_gemini_chat_response_to_openai_chat(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let candidates = body.get("candidates")?.as_array()?;
-    let first_candidate = candidates.first()?.as_object()?;
-    let content = first_candidate.get("content")?.as_object()?;
-    let parts = content.get("parts")?.as_array()?;
-    let mut text = String::new();
-    let mut tool_calls = Vec::new();
-    for (index, part) in parts.iter().enumerate() {
-        let part = part.as_object()?;
-        if let Some(piece) = part.get("text").and_then(Value::as_str) {
-            text.push_str(piece);
-        } else if let Some(function_call) = part.get("functionCall").and_then(Value::as_object) {
-            let tool_name = function_call.get("name")?.as_str()?;
-            let tool_id = function_call
-                .get("id")
-                .and_then(Value::as_str)
-                .filter(|value| !value.is_empty())
-                .map(ToOwned::to_owned)
-                .unwrap_or_else(|| build_generated_tool_call_id(index));
-            let arguments = canonicalize_tool_arguments(function_call.get("args").cloned());
-            tool_calls.push(json!({
-                "id": tool_id,
-                "type": "function",
-                "function": {
-                    "name": tool_name,
-                    "arguments": arguments,
-                }
-            }));
-        } else {
-            return None;
-        }
-    }
-    let mut finish_reason = match first_candidate.get("finishReason").and_then(Value::as_str) {
-        Some("STOP") => Some("stop"),
-        Some("MAX_TOKENS") => Some("length"),
-        Some("SAFETY") => Some("content_filter"),
-        Some(other) if !other.is_empty() => Some(other),
-        _ => None,
-    };
-    if !tool_calls.is_empty() && finish_reason.is_none_or(|reason| reason == "stop") {
-        finish_reason = Some("tool_calls");
-    }
-    let usage = body.get("usageMetadata").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("promptTokenCount"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let completion_tokens = usage
-        .and_then(|value| value.get("candidatesTokenCount"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = usage
-        .and_then(|value| value.get("totalTokenCount"))
-        .and_then(Value::as_u64)
-        .unwrap_or(prompt_tokens + completion_tokens);
-    let model = body
-        .get("modelVersion")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let id = body
-        .get("responseId")
-        .and_then(Value::as_str)
-        .unwrap_or("chatcmpl-local-finalize");
-    let message_content = if text.is_empty() && !tool_calls.is_empty() {
-        Value::Null
-    } else {
-        Value::String(text)
-    };
-    let mut message = Map::new();
-    message.insert("role".to_string(), Value::String("assistant".to_string()));
-    message.insert("content".to_string(), message_content);
-    if !tool_calls.is_empty() {
-        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
-    }
-    Some(json!({
-        "id": id,
-        "object": "chat.completion",
-        "model": model,
-        "choices": [{
-            "index": first_candidate.get("index").and_then(Value::as_u64).unwrap_or(0),
-            "message": Value::Object(message),
-            "finish_reason": finish_reason,
-        }],
-        "usage": {
-            "prompt_tokens": prompt_tokens,
-            "completion_tokens": completion_tokens,
-            "total_tokens": total_tokens,
-        }
-    }))
-}
-
-pub(crate) fn convert_openai_chat_response_to_gemini_chat(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let choices = body.get("choices")?.as_array()?;
-    let first_choice = choices.first()?.as_object()?;
-    let message = first_choice.get("message")?.as_object()?;
-    let mut parts = Vec::new();
-
-    if let Some(text) = extract_openai_assistant_text(message.get("content")) {
-        if !text.trim().is_empty() {
-            parts.push(json!({ "text": text }));
-        }
-    }
-    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
-        for (index, tool_call) in tool_call_values.iter().enumerate() {
-            let tool_call = tool_call.as_object()?;
-            let function = tool_call.get("function")?.as_object()?;
-            let tool_name = function
-                .get("name")
-                .and_then(Value::as_str)
-                .map(str::trim)
-                .filter(|value| !value.is_empty())?;
-            let call_id = tool_call
-                .get("id")
-                .and_then(Value::as_str)
-                .filter(|value| !value.is_empty())
-                .map(ToOwned::to_owned)
-                .unwrap_or_else(|| build_generated_tool_call_id(index));
-            parts.push(json!({
-                "functionCall": {
-                    "id": call_id,
-                    "name": tool_name,
-                    "args": parse_openai_function_arguments(function.get("arguments"))?,
-                }
-            }));
-        }
-    }
-    if parts.is_empty() {
-        parts.push(json!({ "text": "" }));
-    }
-
-    let usage = body.get("usage").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("prompt_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let completion_tokens = usage
-        .and_then(|value| value.get("completion_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = usage
-        .and_then(|value| value.get("total_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(prompt_tokens + completion_tokens);
-    let mut finish_reason = match first_choice.get("finish_reason").and_then(Value::as_str) {
-        Some("stop") | None => "STOP",
-        Some("length") => "MAX_TOKENS",
-        Some("content_filter") => "SAFETY",
-        Some("tool_calls") | Some("function_call") => "STOP",
-        Some(other) => other,
-    };
-    if parts.iter().any(|part| part.get("functionCall").is_some()) {
-        finish_reason = "STOP";
-    }
-    let model = body
-        .get("model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let response_id = body
-        .get("id")
-        .and_then(Value::as_str)
-        .unwrap_or("resp-local-finalize");
-
-    Some(json!({
-        "responseId": response_id,
-        "modelVersion": model,
-        "candidates": [{
-            "content": {
-                "role": "model",
-                "parts": parts,
-            },
-            "finishReason": finish_reason,
-            "index": 0,
-        }],
-        "usageMetadata": {
-            "promptTokenCount": prompt_tokens,
-            "candidatesTokenCount": completion_tokens,
-            "totalTokenCount": total_tokens,
-        }
-    }))
-}
-
-fn extract_openai_assistant_text(content: Option<&Value>) -> Option<String> {
-    match content? {
-        Value::Null => Some(String::new()),
-        Value::String(text) => Some(text.clone()),
-        Value::Array(parts) => {
-            let mut text = String::new();
-            for part in parts {
-                let part = part.as_object()?;
-                let part_type = part
-                    .get("type")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default()
-                    .trim()
-                    .to_ascii_lowercase();
-                if matches!(part_type.as_str(), "text" | "output_text") {
-                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
-                        text.push_str(piece);
-                    }
-                }
-            }
-            Some(text)
-        }
-        _ => None,
-    }
-}
-
-fn parse_openai_function_arguments(arguments: Option<&Value>) -> Option<Value> {
-    match arguments.cloned().unwrap_or(Value::Object(Map::new())) {
-        Value::String(text) => serde_json::from_str(&text)
-            .ok()
-            .or(Some(Value::String(text))),
-        other => Some(other),
-    }
-}
-
-pub(crate) fn aggregate_gemini_stream_sync_response(body: &[u8]) -> Option<Value> {
-    let events = parse_stream_json_events(body)?;
-    if events.is_empty() {
-        return None;
-    }
-
-    let mut candidates: BTreeMap<usize, Value> = BTreeMap::new();
-    let mut response_id: Option<Value> = None;
-    let mut private_response_id: Option<Value> = None;
-    let mut model_version: Option<Value> = None;
-    let mut usage_metadata: Option<Value> = None;
-    let mut prompt_feedback: Option<Value> = None;
-    let mut saw_candidate = false;
-
-    for event in events {
-        let raw_event_object = event.as_object()?;
-        if let Some(id) = raw_event_object.get("responseId") {
-            response_id = Some(id.clone());
-        }
-        if let Some(id) = raw_event_object.get("_v1internal_response_id") {
-            private_response_id = Some(id.clone());
-        }
-        let event_object = if let Some(response) = raw_event_object
-            .get("response")
-            .and_then(Value::as_object)
-            .filter(|response| response.contains_key("candidates"))
-        {
-            response
-        } else {
-            raw_event_object
-        };
-        if let Some(id) = event_object.get("responseId") {
-            response_id = Some(id.clone());
-        }
-        if let Some(id) = event_object.get("_v1internal_response_id") {
-            private_response_id = Some(id.clone());
-        }
-        if let Some(version) = event_object.get("modelVersion") {
-            model_version = Some(version.clone());
-        }
-        if let Some(usage) = event_object.get("usageMetadata") {
-            usage_metadata = Some(usage.clone());
-        }
-        if let Some(prompt) = event_object.get("promptFeedback") {
-            prompt_feedback = Some(prompt.clone());
-        }
-        let Some(event_candidates) = event_object.get("candidates").and_then(Value::as_array)
-        else {
-            continue;
-        };
-        for candidate in event_candidates {
-            let Some(candidate_object) = candidate.as_object() else {
-                continue;
-            };
-            let index = candidate_object
-                .get("index")
-                .and_then(Value::as_u64)
-                .map(|value| value as usize)
-                .unwrap_or(0);
-            candidates.insert(index, Value::Object(candidate_object.clone()));
-            saw_candidate = true;
-        }
-    }
-
-    if !saw_candidate {
-        return None;
-    }
-
-    let mut response = Map::new();
-    if let Some(response_id) = response_id {
-        response.insert("responseId".to_string(), response_id);
-    }
-    if let Some(private_response_id) = private_response_id {
-        response.insert("_v1internal_response_id".to_string(), private_response_id);
-    }
-    response.insert(
-        "candidates".to_string(),
-        Value::Array(candidates.into_values().collect()),
-    );
-    if let Some(version) = model_version {
-        response.insert("modelVersion".to_string(), version);
-    }
-    if let Some(usage) = usage_metadata {
-        response.insert("usageMetadata".to_string(), usage);
-    }
-    if let Some(prompt) = prompt_feedback {
-        response.insert("promptFeedback".to_string(), prompt);
-    }
-    Some(Value::Object(response))
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/sync/cli.rs

@@ -1,147 +0,0 @@
-use base64::Engine as _;
-
-use super::chat::aggregate_gemini_stream_sync_response;
-use serde_json::{json, Value};
-
-use crate::ai_pipeline::conversion::response::build_openai_cli_response;
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
-    LocalCoreSyncFinalizeOutcome,
-};
-use crate::control::GatewayControlDecision;
-use crate::{usage::GatewaySyncReportRequest, GatewayError};
-
-pub(crate) fn maybe_build_local_gemini_cli_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "gemini_cli_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "gemini:cli" || client_api_format != "gemini:cli" || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let body_json = match aggregate_gemini_stream_sync_response(&body_bytes) {
-        Some(body_json) => body_json,
-        None => return Ok(None),
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn convert_gemini_cli_response_to_openai_cli(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let candidates = body.get("candidates")?.as_array()?;
-    let first_candidate = candidates.first()?.as_object()?;
-    let content = first_candidate.get("content")?.as_object()?;
-    let parts = content.get("parts")?.as_array()?;
-    let mut text = String::new();
-    let mut function_calls = Vec::new();
-    for (index, part) in parts.iter().enumerate() {
-        let part = part.as_object()?;
-        if let Some(piece) = part.get("text").and_then(Value::as_str) {
-            text.push_str(piece);
-        } else if let Some(function_call) = part.get("functionCall").and_then(Value::as_object) {
-            let tool_name = function_call.get("name")?.as_str()?;
-            let call_id = function_call
-                .get("id")
-                .and_then(Value::as_str)
-                .filter(|value| !value.is_empty())
-                .map(ToOwned::to_owned)
-                .unwrap_or_else(|| build_generated_tool_call_id(index));
-            let arguments = canonicalize_tool_arguments(function_call.get("args").cloned());
-            function_calls.push(json!({
-                "type": "function_call",
-                "call_id": call_id,
-                "name": tool_name,
-                "arguments": arguments,
-            }));
-        } else {
-            return None;
-        }
-    }
-
-    let usage = body.get("usageMetadata").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("promptTokenCount"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .map(|value| {
-            value
-                .get("candidatesTokenCount")
-                .and_then(Value::as_u64)
-                .unwrap_or(0)
-                + value
-                    .get("thoughtsTokenCount")
-                    .and_then(Value::as_u64)
-                    .unwrap_or(0)
-        })
-        .unwrap_or(0);
-    let total_tokens = usage
-        .and_then(|value| value.get("totalTokenCount"))
-        .and_then(Value::as_u64)
-        .unwrap_or(prompt_tokens + output_tokens);
-    let model = body
-        .get("modelVersion")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let response_id = body
-        .get("responseId")
-        .or_else(|| body.get("_v1internal_response_id"))
-        .and_then(Value::as_str)
-        .unwrap_or("resp-local-finalize");
-
-    Some(build_openai_cli_response(
-        response_id,
-        model,
-        &text,
-        function_calls,
-        prompt_tokens,
-        output_tokens,
-        total_tokens,
-    ))
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/sync/mod.rs

@@ -1,11 +0,0 @@
-mod chat;
-mod cli;
-
-pub(crate) use chat::{
-    aggregate_gemini_stream_sync_response, convert_gemini_chat_response_to_openai_chat,
-    convert_openai_chat_response_to_gemini_chat, maybe_build_local_gemini_stream_sync_response,
-    maybe_build_local_gemini_sync_response,
-};
-pub(crate) use cli::{
-    convert_gemini_cli_response_to_openai_cli, maybe_build_local_gemini_cli_stream_sync_response,
-};

apps/aether-gateway/src/ai_pipeline/finalize/standard/mod.rs

@@ -1,7 +1,5 @@
 //! Standard finalize surface for standard contract sync/stream compilation.
 
-use serde_json::Value;
-
 mod claude;
 mod gemini;
 mod openai;
@@ -10,87 +8,18 @@ mod stream;
 
 pub(crate) use crate::ai_pipeline::conversion::response::{
     build_openai_cli_response, convert_openai_chat_response_to_openai_cli,
+    convert_openai_cli_response_to_openai_chat,
+};
+pub(crate) use aether_ai_pipeline::finalize::sync_products::{
+    aggregate_standard_chat_stream_sync_response, aggregate_standard_cli_stream_sync_response,
+    convert_standard_chat_response, convert_standard_cli_response,
+    maybe_build_openai_chat_cross_format_sync_product_from_normalized_payload,
+    maybe_build_openai_cli_cross_format_sync_product_from_normalized_payload,
+    maybe_build_openai_cli_same_family_sync_body_from_normalized_payload,
+    maybe_build_standard_cross_format_sync_product,
+    maybe_build_standard_cross_format_sync_product_from_normalized_payload,
+    maybe_build_standard_same_format_sync_body_from_normalized_payload,
+    maybe_build_standard_sync_finalize_product_from_normalized_payload,
+    StandardCrossFormatSyncProduct, StandardSyncFinalizeNormalizedProduct,
 };
-pub(crate) use claude::*;
-pub(crate) use gemini::*;
-pub(crate) use openai::*;
 pub(crate) use stream::*;
-
-pub(crate) fn aggregate_standard_chat_stream_sync_response(
-    body: &[u8],
-    provider_api_format: &str,
-) -> Option<Value> {
-    match provider_api_format.trim().to_ascii_lowercase().as_str() {
-        "openai:chat" => aggregate_openai_chat_stream_sync_response(body),
-        "openai:cli" | "openai:compact" => aggregate_openai_cli_stream_sync_response(body),
-        "claude:chat" | "claude:cli" => aggregate_claude_stream_sync_response(body),
-        "gemini:chat" | "gemini:cli" => aggregate_gemini_stream_sync_response(body),
-        _ => None,
-    }
-}
-
-pub(crate) fn convert_standard_chat_response(
-    body_json: &Value,
-    provider_api_format: &str,
-    client_api_format: &str,
-    report_context: &Value,
-) -> Option<Value> {
-    let canonical = match provider_api_format.trim().to_ascii_lowercase().as_str() {
-        "openai:chat" => body_json.clone(),
-        "openai:cli" | "openai:compact" => {
-            convert_openai_cli_response_to_openai_chat(body_json, report_context)?
-        }
-        "claude:chat" | "claude:cli" => {
-            convert_claude_chat_response_to_openai_chat(body_json, report_context)?
-        }
-        "gemini:chat" | "gemini:cli" => {
-            convert_gemini_chat_response_to_openai_chat(body_json, report_context)?
-        }
-        _ => return None,
-    };
-
-    match client_api_format.trim().to_ascii_lowercase().as_str() {
-        "openai:chat" => Some(canonical),
-        "claude:chat" => convert_openai_chat_response_to_claude_chat(&canonical, report_context),
-        "gemini:chat" => convert_openai_chat_response_to_gemini_chat(&canonical, report_context),
-        _ => None,
-    }
-}
-
-pub(crate) fn aggregate_standard_cli_stream_sync_response(
-    body: &[u8],
-    provider_api_format: &str,
-) -> Option<Value> {
-    aggregate_standard_chat_stream_sync_response(body, provider_api_format)
-}
-
-pub(crate) fn convert_standard_cli_response(
-    body_json: &Value,
-    provider_api_format: &str,
-    client_api_format: &str,
-    report_context: &Value,
-) -> Option<Value> {
-    let canonical = match provider_api_format.trim().to_ascii_lowercase().as_str() {
-        "openai:cli" | "openai:compact" => {
-            convert_openai_cli_response_to_openai_chat(body_json, report_context)?
-        }
-        _ => convert_standard_chat_response(
-            body_json,
-            provider_api_format,
-            "openai:chat",
-            report_context,
-        )?,
-    };
-
-    match client_api_format.trim().to_ascii_lowercase().as_str() {
-        "openai:cli" => {
-            convert_openai_chat_response_to_openai_cli(&canonical, report_context, false)
-        }
-        "openai:compact" => {
-            convert_openai_chat_response_to_openai_cli(&canonical, report_context, true)
-        }
-        "claude:cli" => convert_openai_chat_response_to_claude_chat(&canonical, report_context),
-        "gemini:cli" => convert_openai_chat_response_to_gemini_chat(&canonical, report_context),
-        _ => None,
-    }
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/mod.rs

@@ -1,14 +1 @@
 pub(super) mod stream;
-pub(super) mod sync;
-
-pub(crate) use sync::{
-    aggregate_openai_chat_stream_sync_response, aggregate_openai_cli_stream_sync_response,
-    build_openai_cli_response, convert_openai_cli_response_to_openai_chat,
-    maybe_build_local_openai_chat_cross_format_stream_sync_response,
-    maybe_build_local_openai_chat_cross_format_sync_response,
-    maybe_build_local_openai_chat_stream_sync_response,
-    maybe_build_local_openai_chat_sync_response,
-    maybe_build_local_openai_cli_cross_format_stream_sync_response,
-    maybe_build_local_openai_cli_cross_format_sync_response,
-    maybe_build_local_openai_cli_stream_sync_response,
-};

apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/stream.rs

@@ -1,962 +1,4 @@
-use std::collections::BTreeMap;
+pub(crate) use aether_ai_pipeline::finalize::standard::openai::stream::{
-
+    OpenAIChatClientEmitter, OpenAIChatProviderState, OpenAICliClientEmitter,
-use serde_json::{json, Value};
+    OpenAICliProviderState,
-
-use crate::ai_pipeline::conversion::response::build_openai_cli_response;
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, canonicalize_tool_arguments,
 };
-use crate::ai_pipeline::finalize::sse::{encode_done_sse, encode_json_sse};
-use crate::GatewayError;
-
-use crate::ai_pipeline::finalize::standard::stream::common::*;
-
-#[derive(Default)]
-struct OpenAIChatProviderToolState {
-    id: Option<String>,
-    name: Option<String>,
-    started_emitted: bool,
-}
-
-#[derive(Default)]
-pub(crate) struct OpenAIChatProviderState {
-    response_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-    tool_calls: BTreeMap<usize, OpenAIChatProviderToolState>,
-}
-
-#[derive(Default)]
-struct OpenAICliProviderToolState {
-    call_id: String,
-    name: String,
-    arguments: String,
-    started_emitted: bool,
-}
-
-#[derive(Default)]
-pub(crate) struct OpenAICliProviderState {
-    response_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-    text: String,
-    tool_calls: BTreeMap<usize, OpenAICliProviderToolState>,
-    tool_index_by_key: BTreeMap<String, usize>,
-    last_tool_index: Option<usize>,
-}
-
-impl OpenAIChatProviderState {
-    fn identity(&self, report_context: &Value) -> (String, String) {
-        resolve_identity(
-            self.response_id.as_deref(),
-            self.model.as_deref(),
-            report_context,
-            "chatcmpl-local-stream",
-        )
-    }
-
-    fn ensure_started(&mut self, report_context: &Value, out: &mut Vec<CanonicalStreamFrame>) {
-        if self.started {
-            return;
-        }
-        let (id, model) = self.identity(report_context);
-        out.push(CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Start,
-        });
-        self.started = true;
-    }
-
-    pub(crate) fn push_line(
-        &mut self,
-        report_context: &Value,
-        line: Vec<u8>,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        let Some(value) = decode_json_data_line(&line) else {
-            return Ok(Vec::new());
-        };
-        let Some(chunk_object) = value.as_object() else {
-            return Ok(Vec::new());
-        };
-        self.response_id = chunk_object
-            .get("id")
-            .and_then(Value::as_str)
-            .map(ToOwned::to_owned)
-            .or_else(|| self.response_id.clone());
-        self.model = chunk_object
-            .get("model")
-            .and_then(Value::as_str)
-            .map(ToOwned::to_owned)
-            .or_else(|| self.model.clone());
-
-        let mut out = Vec::new();
-        let Some(chunk_choices) = chunk_object.get("choices").and_then(Value::as_array) else {
-            return Ok(out);
-        };
-        for chunk_choice in chunk_choices {
-            let Some(choice_object) = chunk_choice.as_object() else {
-                continue;
-            };
-            let Some(delta) = choice_object.get("delta").and_then(Value::as_object) else {
-                if let Some(finish_reason) = normalize_openai_finish_reason(
-                    choice_object.get("finish_reason").and_then(Value::as_str),
-                ) {
-                    self.ensure_started(report_context, &mut out);
-                    let (id, model) = self.identity(report_context);
-                    out.push(CanonicalStreamFrame {
-                        id,
-                        model,
-                        event: CanonicalStreamEvent::Finish {
-                            finish_reason: Some(finish_reason),
-                            usage: canonical_usage_from_openai_usage(chunk_object.get("usage")),
-                        },
-                    });
-                    self.finished = true;
-                }
-                continue;
-            };
-
-            if delta.get("role").and_then(Value::as_str) == Some("assistant") {
-                self.ensure_started(report_context, &mut out);
-            }
-
-            if let Some(content) = delta.get("content").and_then(Value::as_str) {
-                if !content.is_empty() {
-                    self.ensure_started(report_context, &mut out);
-                    let (id, model) = self.identity(report_context);
-                    out.push(CanonicalStreamFrame {
-                        id,
-                        model,
-                        event: CanonicalStreamEvent::TextDelta(content.to_string()),
-                    });
-                }
-            }
-
-            if let Some(tool_calls) = delta.get("tool_calls").and_then(Value::as_array) {
-                self.ensure_started(report_context, &mut out);
-                let (id, model) = self.identity(report_context);
-                for tool_call in tool_calls {
-                    let Some(tool_call_object) = tool_call.as_object() else {
-                        continue;
-                    };
-                    let index = tool_call_object
-                        .get("index")
-                        .and_then(Value::as_u64)
-                        .map(|value| value as usize)
-                        .unwrap_or(0);
-                    let state = self.tool_calls.entry(index).or_default();
-                    if let Some(call_id) = tool_call_object.get("id").and_then(Value::as_str) {
-                        state.id = Some(call_id.to_string());
-                    }
-                    if let Some(function) =
-                        tool_call_object.get("function").and_then(Value::as_object)
-                    {
-                        if let Some(name) = function.get("name").and_then(Value::as_str) {
-                            state.name = Some(name.to_string());
-                        }
-                        if !state.started_emitted && (state.id.is_some() || state.name.is_some()) {
-                            out.push(CanonicalStreamFrame {
-                                id: id.clone(),
-                                model: model.clone(),
-                                event: CanonicalStreamEvent::ToolCallStart {
-                                    index,
-                                    call_id: state
-                                        .id
-                                        .clone()
-                                        .unwrap_or_else(|| build_generated_tool_call_id(index)),
-                                    name: state
-                                        .name
-                                        .clone()
-                                        .unwrap_or_else(|| "unknown".to_string()),
-                                },
-                            });
-                            state.started_emitted = true;
-                        }
-                        if let Some(arguments) = function.get("arguments").and_then(Value::as_str) {
-                            if !arguments.is_empty() {
-                                if !state.started_emitted {
-                                    out.push(CanonicalStreamFrame {
-                                        id: id.clone(),
-                                        model: model.clone(),
-                                        event: CanonicalStreamEvent::ToolCallStart {
-                                            index,
-                                            call_id: state.id.clone().unwrap_or_else(|| {
-                                                build_generated_tool_call_id(index)
-                                            }),
-                                            name: state
-                                                .name
-                                                .clone()
-                                                .unwrap_or_else(|| "unknown".to_string()),
-                                        },
-                                    });
-                                    state.started_emitted = true;
-                                }
-                                out.push(CanonicalStreamFrame {
-                                    id: id.clone(),
-                                    model: model.clone(),
-                                    event: CanonicalStreamEvent::ToolCallArgumentsDelta {
-                                        index,
-                                        arguments: arguments.to_string(),
-                                    },
-                                });
-                            }
-                        }
-                    }
-                }
-            }
-
-            if let Some(finish_reason) = normalize_openai_finish_reason(
-                choice_object.get("finish_reason").and_then(Value::as_str),
-            ) {
-                self.ensure_started(report_context, &mut out);
-                let (id, model) = self.identity(report_context);
-                out.push(CanonicalStreamFrame {
-                    id,
-                    model,
-                    event: CanonicalStreamEvent::Finish {
-                        finish_reason: Some(finish_reason),
-                        usage: canonical_usage_from_openai_usage(chunk_object.get("usage")),
-                    },
-                });
-                self.finished = true;
-            }
-        }
-
-        Ok(out)
-    }
-
-    pub(crate) fn finish(
-        &mut self,
-        report_context: &Value,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        self.finished = true;
-        let (id, model) = self.identity(report_context);
-        Ok(vec![CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Finish {
-                finish_reason: None,
-                usage: None,
-            },
-        }])
-    }
-}
-
-impl OpenAICliProviderState {
-    fn identity(&self, report_context: &Value) -> (String, String) {
-        resolve_identity(
-            self.response_id.as_deref(),
-            self.model.as_deref(),
-            report_context,
-            "resp-local-stream",
-        )
-    }
-
-    fn ensure_started(&mut self, report_context: &Value, out: &mut Vec<CanonicalStreamFrame>) {
-        if self.started {
-            return;
-        }
-        let (id, model) = self.identity(report_context);
-        out.push(CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Start,
-        });
-        self.started = true;
-    }
-
-    fn tool_index_for_key(&mut self, key: Option<String>, output_index: Option<usize>) -> usize {
-        if let Some(output_index) = output_index {
-            if let Some(key) = key.as_ref() {
-                self.tool_index_by_key
-                    .entry(key.clone())
-                    .or_insert(output_index);
-            }
-            self.last_tool_index = Some(output_index);
-            return output_index;
-        }
-        if let Some(key) = key.as_ref() {
-            if let Some(index) = self.tool_index_by_key.get(key).copied() {
-                self.last_tool_index = Some(index);
-                return index;
-            }
-        }
-        let index = self.last_tool_index.unwrap_or(self.tool_calls.len());
-        if let Some(key) = key {
-            self.tool_index_by_key.insert(key, index);
-        }
-        self.last_tool_index = Some(index);
-        index
-    }
-
-    pub(crate) fn push_line(
-        &mut self,
-        report_context: &Value,
-        line: Vec<u8>,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        let Some(value) = decode_json_data_line(&line) else {
-            return Ok(Vec::new());
-        };
-        let mut out = Vec::new();
-        if let Some(response) = value.get("response").and_then(Value::as_object) {
-            self.response_id = response
-                .get("id")
-                .and_then(Value::as_str)
-                .map(ToOwned::to_owned)
-                .or_else(|| self.response_id.clone());
-            self.model = response
-                .get("model")
-                .and_then(Value::as_str)
-                .map(ToOwned::to_owned)
-                .or_else(|| self.model.clone());
-        }
-
-        match value
-            .get("type")
-            .and_then(Value::as_str)
-            .unwrap_or_default()
-        {
-            "response.created" => {
-                self.ensure_started(report_context, &mut out);
-            }
-            "response.output_text.delta" => {
-                let piece = match value.get("delta") {
-                    Some(Value::String(text)) => text.clone(),
-                    Some(Value::Object(delta)) => delta
-                        .get("text")
-                        .and_then(Value::as_str)
-                        .unwrap_or_default()
-                        .to_string(),
-                    _ => String::new(),
-                };
-                if !piece.is_empty() {
-                    self.ensure_started(report_context, &mut out);
-                    self.text.push_str(&piece);
-                    let (id, model) = self.identity(report_context);
-                    out.push(CanonicalStreamFrame {
-                        id,
-                        model,
-                        event: CanonicalStreamEvent::TextDelta(piece),
-                    });
-                }
-            }
-            "response.output_item.added" => {
-                let Some(item) = value.get("item").and_then(Value::as_object) else {
-                    return Ok(out);
-                };
-                if item.get("type").and_then(Value::as_str) != Some("function_call") {
-                    return Ok(out);
-                }
-                self.ensure_started(report_context, &mut out);
-                let key = item
-                    .get("call_id")
-                    .or_else(|| item.get("id"))
-                    .and_then(Value::as_str)
-                    .map(ToOwned::to_owned);
-                let output_index = value
-                    .get("output_index")
-                    .and_then(Value::as_u64)
-                    .map(|value| value as usize);
-                let index = self.tool_index_for_key(key.clone(), output_index);
-                let (id, model) = self.identity(report_context);
-                let state = self.tool_calls.entry(index).or_default();
-                state.call_id = item
-                    .get("call_id")
-                    .or_else(|| item.get("id"))
-                    .and_then(Value::as_str)
-                    .unwrap_or_else(|| state.call_id.as_str())
-                    .to_string();
-                state.name = item
-                    .get("name")
-                    .and_then(Value::as_str)
-                    .unwrap_or_else(|| state.name.as_str())
-                    .to_string();
-                if !state.started_emitted {
-                    out.push(CanonicalStreamFrame {
-                        id: id.clone(),
-                        model: model.clone(),
-                        event: CanonicalStreamEvent::ToolCallStart {
-                            index,
-                            call_id: if state.call_id.is_empty() {
-                                build_generated_tool_call_id(index)
-                            } else {
-                                state.call_id.clone()
-                            },
-                            name: if state.name.is_empty() {
-                                "unknown".to_string()
-                            } else {
-                                state.name.clone()
-                            },
-                        },
-                    });
-                    state.started_emitted = true;
-                }
-                if let Some(arguments) = item.get("arguments").and_then(Value::as_str) {
-                    if !arguments.is_empty() {
-                        state.arguments.push_str(arguments);
-                        out.push(CanonicalStreamFrame {
-                            id,
-                            model,
-                            event: CanonicalStreamEvent::ToolCallArgumentsDelta {
-                                index,
-                                arguments: arguments.to_string(),
-                            },
-                        });
-                    }
-                }
-            }
-            "response.function_call_arguments.delta" => {
-                let delta = value
-                    .get("delta")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default();
-                if delta.is_empty() {
-                    return Ok(out);
-                }
-                self.ensure_started(report_context, &mut out);
-                let key = value
-                    .get("item_id")
-                    .or_else(|| value.get("call_id"))
-                    .or_else(|| value.get("id"))
-                    .and_then(Value::as_str)
-                    .map(ToOwned::to_owned);
-                let output_index = value
-                    .get("output_index")
-                    .and_then(Value::as_u64)
-                    .map(|value| value as usize);
-                let index = self.tool_index_for_key(key, output_index);
-                let (id, model) = self.identity(report_context);
-                let state = self.tool_calls.entry(index).or_default();
-                if !state.started_emitted {
-                    out.push(CanonicalStreamFrame {
-                        id: id.clone(),
-                        model: model.clone(),
-                        event: CanonicalStreamEvent::ToolCallStart {
-                            index,
-                            call_id: if state.call_id.is_empty() {
-                                build_generated_tool_call_id(index)
-                            } else {
-                                state.call_id.clone()
-                            },
-                            name: if state.name.is_empty() {
-                                "unknown".to_string()
-                            } else {
-                                state.name.clone()
-                            },
-                        },
-                    });
-                    state.started_emitted = true;
-                }
-                state.arguments.push_str(delta);
-                out.push(CanonicalStreamFrame {
-                    id,
-                    model,
-                    event: CanonicalStreamEvent::ToolCallArgumentsDelta {
-                        index,
-                        arguments: delta.to_string(),
-                    },
-                });
-            }
-            "response.completed" => {
-                let Some(response) = value.get("response").and_then(Value::as_object) else {
-                    return Ok(out);
-                };
-                self.ensure_started(report_context, &mut out);
-                let (id, model) = self.identity(report_context);
-
-                for raw_item in response
-                    .get("output")
-                    .and_then(Value::as_array)
-                    .into_iter()
-                    .flatten()
-                {
-                    let Some(item) = raw_item.as_object() else {
-                        continue;
-                    };
-                    match item.get("type").and_then(Value::as_str).unwrap_or_default() {
-                        "message" => {
-                            let mut completed_text = String::new();
-                            for raw_content in item
-                                .get("content")
-                                .and_then(Value::as_array)
-                                .into_iter()
-                                .flatten()
-                            {
-                                let Some(content) = raw_content.as_object() else {
-                                    continue;
-                                };
-                                if content.get("type").and_then(Value::as_str)
-                                    == Some("output_text")
-                                {
-                                    if let Some(text) = content.get("text").and_then(Value::as_str)
-                                    {
-                                        completed_text.push_str(text);
-                                    }
-                                }
-                            }
-                            let missing = if completed_text.starts_with(&self.text) {
-                                completed_text[self.text.len()..].to_string()
-                            } else if self.text == completed_text {
-                                String::new()
-                            } else {
-                                completed_text.clone()
-                            };
-                            if !missing.is_empty() {
-                                self.text.push_str(&missing);
-                                out.push(CanonicalStreamFrame {
-                                    id: id.clone(),
-                                    model: model.clone(),
-                                    event: CanonicalStreamEvent::TextDelta(missing),
-                                });
-                            }
-                        }
-                        "function_call" => {
-                            let key = item
-                                .get("call_id")
-                                .or_else(|| item.get("id"))
-                                .and_then(Value::as_str)
-                                .map(ToOwned::to_owned);
-                            let index = self.tool_index_for_key(key, None);
-                            let state = self.tool_calls.entry(index).or_default();
-                            state.call_id = item
-                                .get("call_id")
-                                .or_else(|| item.get("id"))
-                                .and_then(Value::as_str)
-                                .unwrap_or_else(|| state.call_id.as_str())
-                                .to_string();
-                            state.name = item
-                                .get("name")
-                                .and_then(Value::as_str)
-                                .unwrap_or_else(|| state.name.as_str())
-                                .to_string();
-                            if !state.started_emitted {
-                                out.push(CanonicalStreamFrame {
-                                    id: id.clone(),
-                                    model: model.clone(),
-                                    event: CanonicalStreamEvent::ToolCallStart {
-                                        index,
-                                        call_id: if state.call_id.is_empty() {
-                                            build_generated_tool_call_id(index)
-                                        } else {
-                                            state.call_id.clone()
-                                        },
-                                        name: if state.name.is_empty() {
-                                            "unknown".to_string()
-                                        } else {
-                                            state.name.clone()
-                                        },
-                                    },
-                                });
-                                state.started_emitted = true;
-                            }
-                            let completed_arguments = item
-                                .get("arguments")
-                                .and_then(Value::as_str)
-                                .unwrap_or_default()
-                                .to_string();
-                            let missing = if completed_arguments.starts_with(&state.arguments) {
-                                completed_arguments[state.arguments.len()..].to_string()
-                            } else if state.arguments == completed_arguments {
-                                String::new()
-                            } else {
-                                completed_arguments.clone()
-                            };
-                            if !missing.is_empty() {
-                                state.arguments.push_str(&missing);
-                                out.push(CanonicalStreamFrame {
-                                    id: id.clone(),
-                                    model: model.clone(),
-                                    event: CanonicalStreamEvent::ToolCallArgumentsDelta {
-                                        index,
-                                        arguments: missing,
-                                    },
-                                });
-                            }
-                        }
-                        _ => {}
-                    }
-                }
-
-                let finish_reason = if self.tool_calls.is_empty() {
-                    Some("stop".to_string())
-                } else {
-                    Some("tool_calls".to_string())
-                };
-                out.push(CanonicalStreamFrame {
-                    id,
-                    model,
-                    event: CanonicalStreamEvent::Finish {
-                        finish_reason,
-                        usage: canonical_usage_from_openai_usage(response.get("usage")),
-                    },
-                });
-                self.finished = true;
-            }
-            _ => {}
-        }
-
-        Ok(out)
-    }
-
-    pub(crate) fn finish(
-        &mut self,
-        report_context: &Value,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        self.finished = true;
-        let (id, model) = self.identity(report_context);
-        let finish_reason = if self.tool_calls.is_empty() {
-            Some("stop".to_string())
-        } else {
-            Some("tool_calls".to_string())
-        };
-        Ok(vec![CanonicalStreamFrame {
-            id,
-            model,
-            event: CanonicalStreamEvent::Finish {
-                finish_reason,
-                usage: None,
-            },
-        }])
-    }
-}
-
-#[derive(Default)]
-pub(crate) struct OpenAIChatClientEmitter {
-    response_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-}
-
-#[derive(Default)]
-struct OpenAICliClientToolState {
-    call_id: String,
-    name: String,
-    arguments: String,
-}
-
-#[derive(Default)]
-pub(crate) struct OpenAICliClientEmitter {
-    response_id: Option<String>,
-    model: Option<String>,
-    started: bool,
-    finished: bool,
-    text: String,
-    tool_calls: BTreeMap<usize, OpenAICliClientToolState>,
-}
-
-impl OpenAIChatClientEmitter {
-    fn update_identity(&mut self, frame: &CanonicalStreamFrame) {
-        self.response_id = Some(frame.id.clone());
-        self.model = Some(frame.model.clone());
-    }
-
-    fn ensure_started(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if self.started {
-            return Ok(Vec::new());
-        }
-        self.started = true;
-        Ok(encode_json_sse(
-            None,
-            &build_openai_chat_role_chunk(
-                self.response_id
-                    .as_deref()
-                    .unwrap_or("chatcmpl-local-stream"),
-                self.model.as_deref().unwrap_or("unknown"),
-            ),
-        )?)
-    }
-
-    pub(crate) fn emit(&mut self, frame: CanonicalStreamFrame) -> Result<Vec<u8>, GatewayError> {
-        self.update_identity(&frame);
-        match frame.event {
-            CanonicalStreamEvent::Start => self.ensure_started(),
-            CanonicalStreamEvent::TextDelta(text) => {
-                let mut out = self.ensure_started()?;
-                out.extend(encode_json_sse(
-                    None,
-                    &build_openai_chat_chunk(
-                        self.response_id
-                            .as_deref()
-                            .unwrap_or("chatcmpl-local-stream"),
-                        self.model.as_deref().unwrap_or("unknown"),
-                        text,
-                        None,
-                        None,
-                    ),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::ToolCallStart {
-                index,
-                call_id,
-                name,
-            } => {
-                let mut out = self.ensure_started()?;
-                out.extend(encode_json_sse(
-                    None,
-                    &build_openai_chat_chunk(
-                        self.response_id
-                            .as_deref()
-                            .unwrap_or("chatcmpl-local-stream"),
-                        self.model.as_deref().unwrap_or("unknown"),
-                        String::new(),
-                        Some(vec![json!({
-                            "index": index,
-                            "id": call_id,
-                            "type": "function",
-                            "function": {
-                                "name": name,
-                                "arguments": "",
-                            }
-                        })]),
-                        None,
-                    ),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::ToolCallArgumentsDelta { index, arguments } => {
-                let mut out = self.ensure_started()?;
-                out.extend(encode_json_sse(
-                    None,
-                    &json!({
-                        "id": self.response_id
-                            .as_deref()
-                            .unwrap_or("chatcmpl-local-stream"),
-                        "object": "chat.completion.chunk",
-                        "model": self.model.as_deref().unwrap_or("unknown"),
-                        "choices": [{
-                            "index": 0,
-                            "delta": {
-                                "tool_calls": [{
-                                    "index": index,
-                                    "function": {
-                                        "arguments": arguments,
-                                    }
-                                }]
-                            },
-                            "finish_reason": Value::Null
-                        }]
-                    }),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::Finish { finish_reason, .. } => {
-                if self.finished {
-                    return Ok(Vec::new());
-                }
-                let mut out = self.ensure_started()?;
-                out.extend(encode_json_sse(
-                    None,
-                    &build_openai_chat_finish_chunk(
-                        self.response_id
-                            .as_deref()
-                            .unwrap_or("chatcmpl-local-stream"),
-                        self.model.as_deref().unwrap_or("unknown"),
-                        finish_reason.as_deref(),
-                    ),
-                )?);
-                out.extend(encode_done_sse());
-                self.finished = true;
-                Ok(out)
-            }
-        }
-    }
-
-    pub(crate) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        let out = encode_json_sse(
-            None,
-            &build_openai_chat_finish_chunk(
-                self.response_id
-                    .as_deref()
-                    .unwrap_or("chatcmpl-local-stream"),
-                self.model.as_deref().unwrap_or("unknown"),
-                None,
-            ),
-        )?;
-        self.finished = true;
-        let mut bytes = out;
-        bytes.extend(encode_done_sse());
-        Ok(bytes)
-    }
-}
-
-impl OpenAICliClientEmitter {
-    fn update_identity(&mut self, frame: &CanonicalStreamFrame) {
-        self.response_id = Some(frame.id.clone().replace("chatcmpl", "resp"));
-        self.model = Some(frame.model.clone());
-    }
-
-    fn ensure_started(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if self.started {
-            return Ok(Vec::new());
-        }
-        self.started = true;
-        encode_json_sse(
-            Some("response.created"),
-            &json!({
-                "type": "response.created",
-                "response": {
-                    "id": self.response_id.as_deref().unwrap_or("resp-local-stream"),
-                    "object": "response",
-                    "model": self.model.as_deref().unwrap_or("unknown"),
-                    "status": "in_progress",
-                    "output": [],
-                }
-            }),
-        )
-    }
-
-    fn function_output_index(&self, index: usize) -> usize {
-        if self.text.is_empty() {
-            index
-        } else {
-            index + 1
-        }
-    }
-
-    pub(crate) fn emit(&mut self, frame: CanonicalStreamFrame) -> Result<Vec<u8>, GatewayError> {
-        self.update_identity(&frame);
-        match frame.event {
-            CanonicalStreamEvent::Start => self.ensure_started(),
-            CanonicalStreamEvent::TextDelta(text) => {
-                let mut out = self.ensure_started()?;
-                self.text.push_str(&text);
-                out.extend(encode_json_sse(
-                    Some("response.output_text.delta"),
-                    &json!({
-                        "type": "response.output_text.delta",
-                        "output_index": 0,
-                        "content_index": 0,
-                        "delta": text,
-                    }),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::ToolCallStart {
-                index,
-                call_id,
-                name,
-            } => {
-                let mut out = self.ensure_started()?;
-                let output_index = self.function_output_index(index);
-                let state = self.tool_calls.entry(index).or_default();
-                state.call_id = call_id.clone();
-                state.name = name.clone();
-                out.extend(encode_json_sse(
-                    Some("response.output_item.added"),
-                    &json!({
-                        "type": "response.output_item.added",
-                        "output_index": output_index,
-                        "item": {
-                            "type": "function_call",
-                            "id": call_id,
-                            "call_id": state.call_id,
-                            "name": state.name,
-                            "arguments": "",
-                        }
-                    }),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::ToolCallArgumentsDelta { index, arguments } => {
-                let mut out = self.ensure_started()?;
-                let output_index = self.function_output_index(index);
-                let state = self.tool_calls.entry(index).or_default();
-                state.arguments.push_str(&arguments);
-                out.extend(encode_json_sse(
-                    Some("response.function_call_arguments.delta"),
-                    &json!({
-                        "type": "response.function_call_arguments.delta",
-                        "output_index": output_index,
-                        "item_id": if state.call_id.is_empty() {
-                            build_generated_tool_call_id(index)
-                        } else {
-                            state.call_id.clone()
-                        },
-                        "delta": arguments,
-                    }),
-                )?);
-                Ok(out)
-            }
-            CanonicalStreamEvent::Finish { usage, .. } => {
-                if self.finished {
-                    return Ok(Vec::new());
-                }
-                let mut out = self.ensure_started()?;
-                let usage = usage.unwrap_or_default();
-                let function_calls = self
-                    .tool_calls
-                    .iter()
-                    .map(|(index, state)| {
-                        json!({
-                            "type": "function_call",
-                            "id": if state.call_id.is_empty() {
-                                build_generated_tool_call_id(*index)
-                            } else {
-                                state.call_id.clone()
-                            },
-                            "call_id": if state.call_id.is_empty() {
-                                build_generated_tool_call_id(*index)
-                            } else {
-                                state.call_id.clone()
-                            },
-                            "name": if state.name.is_empty() {
-                                "unknown".to_string()
-                            } else {
-                                state.name.clone()
-                            },
-                            "arguments": state.arguments.clone(),
-                        })
-                    })
-                    .collect::<Vec<_>>();
-                out.extend(encode_json_sse(
-                    Some("response.completed"),
-                    &json!({
-                        "type": "response.completed",
-                        "response": build_openai_cli_response(
-                            self.response_id.as_deref().unwrap_or("resp-local-stream"),
-                            self.model.as_deref().unwrap_or("unknown"),
-                            &self.text,
-                            function_calls,
-                            usage.input_tokens,
-                            usage.output_tokens,
-                            usage.total_tokens,
-                        ),
-                    }),
-                )?);
-                self.finished = true;
-                Ok(out)
-            }
-        }
-    }
-
-    pub(crate) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
-        if !self.started || self.finished {
-            return Ok(Vec::new());
-        }
-        self.emit(CanonicalStreamFrame {
-            id: self
-                .response_id
-                .clone()
-                .unwrap_or_else(|| "resp-local-stream".to_string()),
-            model: self.model.clone().unwrap_or_else(|| "unknown".to_string()),
-            event: CanonicalStreamEvent::Finish {
-                finish_reason: None,
-                usage: None,
-            },
-        })
-    }
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/sync/chat.rs

@@ -1,583 +0,0 @@
-use base64::Engine as _;
-use std::collections::BTreeMap;
-
-use serde_json::{json, Map, Value};
-
-use super::cli::aggregate_openai_cli_stream_sync_response;
-use crate::ai_pipeline::conversion::response::{
-    convert_claude_chat_response_to_openai_chat, convert_gemini_chat_response_to_openai_chat,
-};
-use crate::ai_pipeline::conversion::sync_chat_response_conversion_kind;
-use crate::ai_pipeline::finalize::common::{
-    build_generated_tool_call_id, build_local_success_outcome,
-    build_local_success_outcome_with_conversion_report, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
-    LocalCoreSyncFinalizeOutcome,
-};
-use crate::ai_pipeline::finalize::standard::claude::aggregate_claude_stream_sync_response;
-use crate::ai_pipeline::finalize::standard::gemini::aggregate_gemini_stream_sync_response;
-use crate::control::GatewayControlDecision;
-use crate::{usage::GatewaySyncReportRequest, GatewayError};
-
-#[derive(Debug, Default)]
-struct OpenAIChatChoiceState {
-    role: Option<String>,
-    content: String,
-    finish_reason: Option<String>,
-    tool_calls: BTreeMap<usize, OpenAIChatToolCallState>,
-}
-
-#[derive(Debug, Default)]
-struct OpenAIChatToolCallState {
-    id: Option<String>,
-    tool_type: Option<String>,
-    function_name: Option<String>,
-    function_arguments: String,
-}
-
-pub(crate) fn maybe_build_local_openai_chat_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "openai_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "openai:chat"
-        || client_api_format != "openai:chat"
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let body_json = match aggregate_openai_chat_stream_sync_response(&body_bytes) {
-        Some(body_json) => body_json,
-        None => return Ok(None),
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn maybe_build_local_openai_chat_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "openai_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if provider_api_format != "openai:chat"
-        || client_api_format != "openai:chat"
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-pub(crate) fn maybe_build_local_openai_chat_cross_format_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "openai_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if client_api_format != "openai:chat" || !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    let Some(conversion_kind) =
-        sync_chat_response_conversion_kind(&provider_api_format, &client_api_format)
-    else {
-        return Ok(None);
-    };
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let aggregated = match provider_api_format.as_str() {
-        "claude:chat" | "claude:cli" => aggregate_claude_stream_sync_response(&body_bytes),
-        "gemini:chat" | "gemini:cli" => aggregate_gemini_stream_sync_response(&body_bytes),
-        "openai:cli" | "openai:compact" => aggregate_openai_cli_stream_sync_response(&body_bytes),
-        _ => None,
-    };
-    let Some(aggregated) = aggregated else {
-        return Ok(None);
-    };
-    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
-        return Ok(None);
-    };
-    let converted = match provider_api_format.as_str() {
-        "claude:chat" | "claude:cli" => {
-            convert_claude_chat_response_to_openai_chat(&aggregated, report_context)
-        }
-        "gemini:chat" | "gemini:cli" => {
-            convert_gemini_chat_response_to_openai_chat(&aggregated, report_context)
-        }
-        "openai:cli" | "openai:compact" => {
-            convert_openai_cli_response_to_openai_chat(&aggregated, report_context)
-        }
-        _ => None,
-    };
-    let Some(converted) = converted else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, aggregated,
-    )?))
-}
-
-pub(crate) fn maybe_build_local_openai_chat_cross_format_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if payload.report_kind != "openai_chat_sync_finalize" || payload.status_code >= 400 {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if client_api_format != "openai:chat" || !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    let Some(conversion_kind) =
-        sync_chat_response_conversion_kind(&provider_api_format, &client_api_format)
-    else {
-        return Ok(None);
-    };
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-    let converted = match provider_api_format.as_str() {
-        "claude:chat" | "claude:cli" => {
-            convert_claude_chat_response_to_openai_chat(&body_json, report_context)
-        }
-        "gemini:chat" | "gemini:cli" => {
-            convert_gemini_chat_response_to_openai_chat(&body_json, report_context)
-        }
-        "openai:cli" | "openai:compact" => {
-            convert_openai_cli_response_to_openai_chat(&body_json, report_context)
-        }
-        _ => None,
-    };
-    let Some(converted) = converted else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, body_json,
-    )?))
-}
-
-pub(crate) fn convert_openai_cli_response_to_openai_chat(
-    body_json: &Value,
-    report_context: &Value,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let mut text = String::new();
-    let mut tool_calls = Vec::new();
-
-    if let Some(output_items) = body.get("output").and_then(Value::as_array) {
-        for (index, item) in output_items.iter().enumerate() {
-            let item_object = item.as_object()?;
-            let item_type = item_object
-                .get("type")
-                .and_then(Value::as_str)
-                .unwrap_or_default()
-                .trim()
-                .to_ascii_lowercase();
-            match item_type.as_str() {
-                "message" => {
-                    if let Some(content) = item_object.get("content").and_then(Value::as_array) {
-                        for part in content {
-                            let part_object = part.as_object()?;
-                            let part_type = part_object
-                                .get("type")
-                                .and_then(Value::as_str)
-                                .unwrap_or_default()
-                                .trim()
-                                .to_ascii_lowercase();
-                            if matches!(part_type.as_str(), "output_text" | "text") {
-                                if let Some(piece) = part_object.get("text").and_then(Value::as_str)
-                                {
-                                    text.push_str(piece);
-                                }
-                            }
-                        }
-                    }
-                }
-                "function_call" => {
-                    let tool_name = item_object
-                        .get("name")
-                        .and_then(Value::as_str)
-                        .map(str::trim)
-                        .filter(|value| !value.is_empty())?;
-                    let tool_id = item_object
-                        .get("call_id")
-                        .and_then(Value::as_str)
-                        .filter(|value| !value.is_empty())
-                        .or_else(|| {
-                            item_object
-                                .get("id")
-                                .and_then(Value::as_str)
-                                .filter(|value| !value.is_empty())
-                        })
-                        .map(ToOwned::to_owned)
-                        .unwrap_or_else(|| build_generated_tool_call_id(index));
-                    tool_calls.push(json!({
-                        "id": tool_id,
-                        "type": "function",
-                        "function": {
-                            "name": tool_name,
-                            "arguments": canonicalize_tool_arguments(item_object.get("arguments").cloned()),
-                        }
-                    }));
-                }
-                "output_text" | "text" => {
-                    if let Some(piece) = item_object.get("text").and_then(Value::as_str) {
-                        text.push_str(piece);
-                    }
-                }
-                _ => {}
-            }
-        }
-    }
-
-    let finish_reason = if tool_calls.is_empty() {
-        Some("stop")
-    } else {
-        Some("tool_calls")
-    };
-    let model = body
-        .get("model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-    let id = body
-        .get("id")
-        .and_then(Value::as_str)
-        .unwrap_or("chatcmpl-local-openai-cli");
-
-    let usage = body.get("usage").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("input_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let completion_tokens = usage
-        .and_then(|value| value.get("output_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = usage
-        .and_then(|value| value.get("total_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(prompt_tokens + completion_tokens);
-
-    let mut message = Map::new();
-    message.insert("role".to_string(), Value::String("assistant".to_string()));
-    if text.is_empty() && !tool_calls.is_empty() {
-        message.insert("content".to_string(), Value::Null);
-    } else {
-        message.insert("content".to_string(), Value::String(text));
-    }
-    if !tool_calls.is_empty() {
-        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
-    }
-
-    Some(json!({
-        "id": id,
-        "object": "chat.completion",
-        "model": model,
-        "choices": [{
-            "index": 0,
-            "message": Value::Object(message),
-            "finish_reason": finish_reason,
-        }],
-        "usage": {
-            "prompt_tokens": prompt_tokens,
-            "completion_tokens": completion_tokens,
-            "total_tokens": total_tokens,
-        }
-    }))
-}
-
-pub(crate) fn aggregate_openai_chat_stream_sync_response(body: &[u8]) -> Option<Value> {
-    let text = std::str::from_utf8(body).ok()?;
-    let mut response_id: Option<String> = None;
-    let mut model: Option<String> = None;
-    let mut created: Option<u64> = None;
-    let mut usage: Option<Value> = None;
-    let mut choices: BTreeMap<usize, OpenAIChatChoiceState> = BTreeMap::new();
-    let mut saw_chunk = false;
-
-    for raw_line in text.lines() {
-        let line = raw_line.trim_matches('\r').trim();
-        if line.is_empty() || line.starts_with(':') || line.starts_with("event:") {
-            continue;
-        }
-
-        let Some(data_line) = line.strip_prefix("data:") else {
-            continue;
-        };
-        let data_line = data_line.trim();
-        if data_line.is_empty() || data_line == "[DONE]" {
-            continue;
-        }
-
-        let chunk: Value = serde_json::from_str(data_line).ok()?;
-        let chunk_object = chunk.as_object()?;
-        saw_chunk = true;
-
-        if response_id.is_none() {
-            response_id = chunk_object
-                .get("id")
-                .and_then(Value::as_str)
-                .map(ToOwned::to_owned);
-        }
-        if model.is_none() {
-            model = chunk_object
-                .get("model")
-                .and_then(Value::as_str)
-                .map(ToOwned::to_owned);
-        }
-        if created.is_none() {
-            created = chunk_object.get("created").and_then(Value::as_u64);
-        }
-        if let Some(u) = chunk_object.get("usage") {
-            usage = Some(u.clone());
-        }
-
-        let Some(chunk_choices) = chunk_object.get("choices").and_then(Value::as_array) else {
-            continue;
-        };
-        for chunk_choice in chunk_choices {
-            let Some(choice_object) = chunk_choice.as_object() else {
-                continue;
-            };
-            let Some(index) = choice_object
-                .get("index")
-                .and_then(Value::as_u64)
-                .map(|value| value as usize)
-            else {
-                continue;
-            };
-            let state = choices.entry(index).or_default();
-            if let Some(finish_reason) = choice_object.get("finish_reason").and_then(Value::as_str)
-            {
-                state.finish_reason = Some(finish_reason.to_string());
-            }
-
-            let Some(delta) = choice_object.get("delta").and_then(Value::as_object) else {
-                continue;
-            };
-            if let Some(role) = delta.get("role").and_then(Value::as_str) {
-                state.role = Some(role.to_string());
-            }
-            if let Some(content) = delta.get("content").and_then(Value::as_str) {
-                state.content.push_str(content);
-            }
-            if let Some(tool_calls) = delta.get("tool_calls").and_then(Value::as_array) {
-                for tool_call in tool_calls {
-                    let Some(tool_call_object) = tool_call.as_object() else {
-                        continue;
-                    };
-                    let tool_index = tool_call_object
-                        .get("index")
-                        .and_then(Value::as_u64)
-                        .map(|value| value as usize)
-                        .unwrap_or(0);
-                    let tool_state = state.tool_calls.entry(tool_index).or_default();
-                    if let Some(id) = tool_call_object.get("id").and_then(Value::as_str) {
-                        tool_state.id = Some(id.to_string());
-                    }
-                    if let Some(tool_type) = tool_call_object.get("type").and_then(Value::as_str) {
-                        tool_state.tool_type = Some(tool_type.to_string());
-                    }
-                    if let Some(function) =
-                        tool_call_object.get("function").and_then(Value::as_object)
-                    {
-                        if let Some(name) = function.get("name").and_then(Value::as_str) {
-                            tool_state.function_name = Some(name.to_string());
-                        }
-                        if let Some(arguments) = function.get("arguments").and_then(Value::as_str) {
-                            tool_state.function_arguments.push_str(arguments);
-                        }
-                    }
-                }
-            }
-        }
-    }
-
-    if !saw_chunk {
-        return None;
-    }
-
-    let mut response_object = Map::new();
-    response_object.insert(
-        "id".to_string(),
-        Value::String(response_id.unwrap_or_else(|| "chatcmpl-local-finalize".to_string())),
-    );
-    response_object.insert(
-        "object".to_string(),
-        Value::String("chat.completion".to_string()),
-    );
-    if let Some(created) = created {
-        response_object.insert("created".to_string(), Value::Number(created.into()));
-    }
-    if let Some(model) = model {
-        response_object.insert("model".to_string(), Value::String(model));
-    }
-
-    let mut response_choices = Vec::with_capacity(choices.len());
-    for (index, state) in choices {
-        let mut message = Map::new();
-        message.insert(
-            "role".to_string(),
-            Value::String(state.role.unwrap_or_else(|| "assistant".to_string())),
-        );
-        if state.tool_calls.is_empty() {
-            message.insert("content".to_string(), Value::String(state.content));
-        } else {
-            if state.content.is_empty() {
-                message.insert("content".to_string(), Value::Null);
-            } else {
-                message.insert("content".to_string(), Value::String(state.content));
-            }
-            let tool_calls = state
-                .tool_calls
-                .into_iter()
-                .map(|(tool_index, tool_state)| {
-                    json!({
-                        "index": tool_index,
-                        "id": tool_state.id,
-                        "type": tool_state.tool_type.unwrap_or_else(|| "function".to_string()),
-                        "function": {
-                            "name": tool_state.function_name,
-                            "arguments": tool_state.function_arguments,
-                        },
-                    })
-                })
-                .collect::<Vec<_>>();
-            message.insert("tool_calls".to_string(), Value::Array(tool_calls));
-        }
-
-        response_choices.push(json!({
-            "index": index,
-            "message": Value::Object(message),
-            "finish_reason": state.finish_reason,
-        }));
-    }
-    response_object.insert("choices".to_string(), Value::Array(response_choices));
-    if let Some(usage) = usage {
-        response_object.insert("usage".to_string(), usage);
-    }
-
-    Some(Value::Object(response_object))
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/sync/cli.rs

@@ -1,696 +0,0 @@
-use base64::Engine as _;
-
-use serde_json::{json, Value};
-
-use crate::ai_pipeline::conversion::response::{
-    convert_claude_cli_response_to_openai_cli, convert_gemini_cli_response_to_openai_cli,
-};
-use crate::ai_pipeline::conversion::sync_cli_response_conversion_kind;
-use crate::ai_pipeline::finalize::common::{
-    build_local_success_outcome, build_local_success_outcome_with_conversion_report,
-    canonicalize_tool_arguments, local_finalize_allows_envelope,
-    unwrap_local_finalize_response_value, LocalCoreSyncFinalizeOutcome,
-};
-use crate::ai_pipeline::finalize::standard::claude::aggregate_claude_stream_sync_response;
-use crate::ai_pipeline::finalize::standard::gemini::aggregate_gemini_stream_sync_response;
-use crate::control::GatewayControlDecision;
-use crate::{usage::GatewaySyncReportRequest, GatewayError};
-
-pub(crate) fn maybe_build_local_openai_cli_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if let Some(response) =
-        maybe_build_local_openai_cli_direct_stream_sync_response(trace_id, decision, payload)?
-    {
-        return Ok(Some(response));
-    }
-    if let Some(response) = maybe_build_local_openai_cli_openai_family_stream_sync_response(
-        trace_id, decision, payload,
-    )? {
-        return Ok(Some(response));
-    }
-    maybe_build_local_openai_cli_direct_sync_response(trace_id, decision, payload)
-}
-
-pub(crate) fn maybe_build_local_openai_cli_cross_format_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if let Some(response) =
-        maybe_build_local_openai_cli_antigravity_cross_format_stream_sync_response(
-            trace_id, decision, payload,
-        )?
-    {
-        return Ok(Some(response));
-    }
-
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let _has_envelope = report_context
-        .get("has_envelope")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !matches!(client_api_format.as_str(), "openai:cli" | "openai:compact")
-        || !local_finalize_allows_envelope(report_context)
-    {
-        return Ok(None);
-    }
-    let Some(conversion_kind) =
-        sync_cli_response_conversion_kind(&provider_api_format, &client_api_format)
-    else {
-        return Ok(None);
-    };
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let aggregated = match provider_api_format.as_str() {
-        "openai:cli" | "openai:compact" => aggregate_openai_cli_stream_sync_response(&body_bytes),
-        "claude:chat" | "claude:cli" => aggregate_claude_stream_sync_response(&body_bytes),
-        "gemini:chat" | "gemini:cli" => aggregate_gemini_stream_sync_response(&body_bytes),
-        _ => None,
-    };
-    let Some(aggregated) = aggregated else {
-        return Ok(None);
-    };
-    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
-        return Ok(None);
-    };
-    let converted = match provider_api_format.as_str() {
-        "openai:cli" | "openai:compact" => Some(aggregated.clone()),
-        "claude:chat" | "claude:cli" => {
-            convert_claude_cli_response_to_openai_cli(&aggregated, report_context)
-        }
-        "gemini:chat" | "gemini:cli" => {
-            convert_gemini_cli_response_to_openai_cli(&aggregated, report_context)
-        }
-        _ => None,
-    };
-    let Some(converted) = converted else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, aggregated,
-    )?))
-}
-
-pub(crate) fn maybe_build_local_openai_cli_cross_format_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if let Some(response) = maybe_build_local_openai_cli_antigravity_cross_format_sync_response(
-        trace_id, decision, payload,
-    )? {
-        return Ok(Some(response));
-    }
-
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if !matches!(client_api_format.as_str(), "openai:cli" | "openai:compact")
-        || !local_finalize_allows_envelope(report_context)
-    {
-        return Ok(None);
-    }
-    let Some(conversion_kind) =
-        sync_cli_response_conversion_kind(&provider_api_format, &client_api_format)
-    else {
-        return Ok(None);
-    };
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-    let converted = match provider_api_format.as_str() {
-        "openai:cli" | "openai:compact" => Some(body_json.clone()),
-        "claude:chat" | "claude:cli" => {
-            convert_claude_cli_response_to_openai_cli(&body_json, report_context)
-        }
-        "gemini:chat" | "gemini:cli" => {
-            convert_gemini_cli_response_to_openai_cli(&body_json, report_context)
-        }
-        _ => None,
-    };
-    let Some(converted) = converted else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id, decision, payload, converted, body_json,
-    )?))
-}
-
-fn maybe_build_local_openai_cli_direct_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-
-    if !local_finalize_allows_envelope(report_context)
-        || !is_openai_cli_family_api_format(provider_api_format.as_str())
-        || !is_openai_cli_family_api_format(client_api_format.as_str())
-    {
-        return Ok(None);
-    }
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-fn maybe_build_local_openai_cli_direct_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let needs_conversion = report_context
-        .get("needs_conversion")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !local_finalize_allows_envelope(report_context) {
-        return Ok(None);
-    }
-    if !matches!(
-        provider_api_format.as_str(),
-        "openai:cli" | "openai:compact"
-    ) || provider_api_format != client_api_format
-        || needs_conversion
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let body_json = match aggregate_openai_cli_stream_sync_response(&body_bytes) {
-        Some(body_json) => body_json,
-        None => return Ok(None),
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-fn maybe_build_local_openai_cli_openai_family_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if !local_finalize_allows_envelope(report_context)
-        || !is_openai_cli_family_api_format(provider_api_format.as_str())
-        || !is_openai_cli_family_api_format(client_api_format.as_str())
-        || provider_api_format == client_api_format
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let Some(body_json) = aggregate_openai_cli_stream_sync_response(&body_bytes) else {
-        return Ok(None);
-    };
-    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome(
-        trace_id, decision, payload, body_json,
-    )?))
-}
-
-fn maybe_build_local_openai_cli_antigravity_cross_format_stream_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if provider_api_format != "gemini:cli"
-        || !is_openai_cli_family_api_format(client_api_format.as_str())
-        || !is_antigravity_v1internal_envelope(report_context)
-        || !local_finalize_allows_envelope(report_context)
-    {
-        return Ok(None);
-    }
-
-    let Some(body_base64) = payload.body_base64.as_deref() else {
-        return Ok(None);
-    };
-    let body_bytes = base64::engine::general_purpose::STANDARD
-        .decode(body_base64)
-        .map_err(|err| GatewayError::Internal(err.to_string()))?;
-    let Some(aggregated) = aggregate_gemini_stream_sync_response(&body_bytes) else {
-        return Ok(None);
-    };
-    let Some(provider_body_json) =
-        unwrap_cli_conversion_response_value(aggregated, report_context)?
-    else {
-        return Ok(None);
-    };
-    let Some(converted) =
-        convert_gemini_cli_response_to_openai_cli(&provider_body_json, report_context)
-    else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id,
-        decision,
-        payload,
-        converted,
-        provider_body_json,
-    )?))
-}
-
-fn maybe_build_local_openai_cli_antigravity_cross_format_sync_response(
-    trace_id: &str,
-    decision: &GatewayControlDecision,
-    payload: &GatewaySyncReportRequest,
-) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
-    if !matches!(
-        payload.report_kind.as_str(),
-        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
-    ) || payload.status_code >= 400
-    {
-        return Ok(None);
-    }
-
-    let Some(report_context) = payload.report_context.as_ref() else {
-        return Ok(None);
-    };
-    let provider_api_format = report_context
-        .get("provider_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    let client_api_format = report_context
-        .get("client_api_format")
-        .and_then(Value::as_str)
-        .unwrap_or_default()
-        .trim()
-        .to_ascii_lowercase();
-    if provider_api_format != "gemini:cli"
-        || !is_openai_cli_family_api_format(client_api_format.as_str())
-        || !is_antigravity_v1internal_envelope(report_context)
-        || !local_finalize_allows_envelope(report_context)
-    {
-        return Ok(None);
-    }
-
-    let Some(body_json) = payload.body_json.as_ref() else {
-        return Ok(None);
-    };
-    let Some(provider_body_json) =
-        unwrap_cli_conversion_response_value(body_json.clone(), report_context)?
-    else {
-        return Ok(None);
-    };
-    let Some(converted) =
-        convert_gemini_cli_response_to_openai_cli(&provider_body_json, report_context)
-    else {
-        return Ok(None);
-    };
-
-    Ok(Some(build_local_success_outcome_with_conversion_report(
-        trace_id,
-        decision,
-        payload,
-        converted,
-        provider_body_json,
-    )?))
-}
-
-fn unwrap_cli_conversion_response_value(
-    data: Value,
-    report_context: &Value,
-) -> Result<Option<Value>, GatewayError> {
-    if !is_antigravity_v1internal_envelope(report_context) {
-        return unwrap_local_finalize_response_value(data, report_context);
-    }
-
-    let mut unwrapped = if let Some(response) = data
-        .get("response")
-        .and_then(Value::as_object)
-        .filter(|response| !response.contains_key("response"))
-    {
-        let mut response = response.clone();
-        if let Some(response_id) = data.get("responseId").cloned() {
-            response
-                .entry("responseId".to_string())
-                .or_insert(response_id);
-        }
-        Value::Object(response)
-    } else {
-        data
-    };
-
-    if let Some(object) = unwrapped.as_object_mut() {
-        if !object.contains_key("responseId") {
-            if let Some(response_id) = object.get("_v1internal_response_id").cloned() {
-                object.insert("responseId".to_string(), response_id);
-            }
-        }
-    }
-
-    Ok(Some(unwrapped))
-}
-
-fn is_antigravity_v1internal_envelope(report_context: &Value) -> bool {
-    report_context
-        .get("has_envelope")
-        .and_then(Value::as_bool)
-        .unwrap_or(false)
-        && report_context
-            .get("envelope_name")
-            .and_then(Value::as_str)
-            .is_some_and(|value| value.eq_ignore_ascii_case("antigravity:v1internal"))
-}
-
-fn is_openai_cli_family_api_format(api_format: &str) -> bool {
-    matches!(api_format, "openai:cli" | "openai:compact")
-}
-
-pub(crate) fn build_openai_cli_response(
-    response_id: &str,
-    model: &str,
-    text: &str,
-    function_calls: Vec<Value>,
-    prompt_tokens: u64,
-    output_tokens: u64,
-    total_tokens: u64,
-) -> Value {
-    let mut output = Vec::new();
-    if !text.is_empty() {
-        output.push(json!({
-            "type": "message",
-            "id": format!("{response_id}_msg"),
-            "role": "assistant",
-            "status": "completed",
-            "content": [{
-                "type": "output_text",
-                "text": text,
-                "annotations": []
-            }]
-        }));
-    }
-    output.extend(function_calls);
-    json!({
-        "id": response_id,
-        "object": "response",
-        "status": "completed",
-        "model": model,
-        "output": output,
-        "usage": {
-            "input_tokens": prompt_tokens,
-            "output_tokens": output_tokens,
-            "total_tokens": total_tokens,
-        }
-    })
-}
-
-pub(crate) fn convert_openai_chat_response_to_openai_cli(
-    body_json: &Value,
-    report_context: &Value,
-    compact: bool,
-) -> Option<Value> {
-    let body = body_json.as_object()?;
-    let choices = body.get("choices")?.as_array()?;
-    let first_choice = choices.first()?.as_object()?;
-    let message = first_choice.get("message")?.as_object()?;
-    let mut text = String::new();
-    match message.get("content") {
-        Some(Value::String(value)) => text.push_str(value),
-        Some(Value::Array(parts)) => {
-            for part in parts {
-                let part = part.as_object()?;
-                let part_type = part
-                    .get("type")
-                    .and_then(Value::as_str)
-                    .unwrap_or_default()
-                    .trim()
-                    .to_ascii_lowercase();
-                if matches!(part_type.as_str(), "text" | "output_text") {
-                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
-                        text.push_str(piece);
-                    }
-                }
-            }
-        }
-        Some(Value::Null) | None => {}
-        _ => return None,
-    }
-
-    let mut function_calls = Vec::new();
-    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
-        for tool_call in tool_call_values {
-            let tool_call = tool_call.as_object()?;
-            let function = tool_call.get("function")?.as_object()?;
-            let tool_name = function
-                .get("name")
-                .and_then(Value::as_str)
-                .map(str::trim)
-                .filter(|value| !value.is_empty())?;
-            function_calls.push(json!({
-                "type": "function_call",
-                "id": tool_call.get("id").cloned().unwrap_or(Value::Null),
-                "call_id": tool_call.get("id").cloned().unwrap_or(Value::Null),
-                "name": tool_name,
-                "arguments": canonicalize_tool_arguments(function.get("arguments").cloned()),
-            }));
-        }
-    }
-
-    let usage = body.get("usage").and_then(Value::as_object);
-    let prompt_tokens = usage
-        .and_then(|value| value.get("prompt_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .and_then(|value| value.get("completion_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = usage
-        .and_then(|value| value.get("total_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(prompt_tokens + output_tokens);
-    let response_id = if compact {
-        body.get("id")
-            .and_then(Value::as_str)
-            .map(|value| value.replace("chatcmpl", "resp"))
-            .unwrap_or_else(|| "resp-local-finalize".to_string())
-    } else {
-        body.get("id")
-            .and_then(Value::as_str)
-            .map(|value| value.replace("chatcmpl", "resp"))
-            .unwrap_or_else(|| "resp-local-finalize".to_string())
-    };
-    let model = body
-        .get("model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown");
-
-    Some(build_openai_cli_response(
-        &response_id,
-        model,
-        &text,
-        function_calls,
-        prompt_tokens,
-        output_tokens,
-        total_tokens,
-    ))
-}
-
-pub(crate) fn aggregate_openai_cli_stream_sync_response(body: &[u8]) -> Option<Value> {
-    let text = std::str::from_utf8(body).ok()?;
-
-    for raw_line in text.lines() {
-        let line = raw_line.trim_matches('\r').trim();
-        if line.is_empty() || line.starts_with(':') || line.starts_with("event:") {
-            continue;
-        }
-        let Some(data_line) = line.strip_prefix("data:") else {
-            continue;
-        };
-        let data_line = data_line.trim();
-        if data_line.is_empty() || data_line == "[DONE]" {
-            continue;
-        }
-
-        let event: Value = serde_json::from_str(data_line).ok()?;
-        let event_type = event
-            .get("type")
-            .and_then(Value::as_str)
-            .unwrap_or_default();
-        if event_type == "response.completed" {
-            let response = event.get("response")?.as_object()?.clone();
-            return Some(Value::Object(response));
-        }
-    }
-
-    None
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/sync/mod.rs

@@ -1,16 +0,0 @@
-mod chat;
-mod cli;
-
-pub(crate) use chat::{
-    aggregate_openai_chat_stream_sync_response, convert_openai_cli_response_to_openai_chat,
-    maybe_build_local_openai_chat_cross_format_stream_sync_response,
-    maybe_build_local_openai_chat_cross_format_sync_response,
-    maybe_build_local_openai_chat_stream_sync_response,
-    maybe_build_local_openai_chat_sync_response,
-};
-pub(crate) use cli::{
-    aggregate_openai_cli_stream_sync_response, build_openai_cli_response,
-    maybe_build_local_openai_cli_cross_format_stream_sync_response,
-    maybe_build_local_openai_cli_cross_format_sync_response,
-    maybe_build_local_openai_cli_stream_sync_response,
-};

apps/aether-gateway/src/ai_pipeline/finalize/standard/stream_core/common.rs

@@ -1,222 +1 @@
-use serde_json::{json, Map, Value};
+pub(crate) use aether_ai_pipeline::finalize::standard::stream_core::common::*;
-
-#[derive(Clone, Debug, Default)]
-pub(crate) struct CanonicalUsage {
-    pub(crate) input_tokens: u64,
-    pub(crate) output_tokens: u64,
-    pub(crate) total_tokens: u64,
-}
-
-#[derive(Clone, Debug)]
-pub(crate) enum CanonicalStreamEvent {
-    Start,
-    TextDelta(String),
-    ToolCallStart {
-        index: usize,
-        call_id: String,
-        name: String,
-    },
-    ToolCallArgumentsDelta {
-        index: usize,
-        arguments: String,
-    },
-    Finish {
-        finish_reason: Option<String>,
-        usage: Option<CanonicalUsage>,
-    },
-}
-
-#[derive(Clone, Debug)]
-pub(crate) struct CanonicalStreamFrame {
-    pub(crate) id: String,
-    pub(crate) model: String,
-    pub(crate) event: CanonicalStreamEvent,
-}
-
-pub(crate) fn decode_json_data_line(line: &[u8]) -> Option<Value> {
-    let text = std::str::from_utf8(line).ok()?;
-    let trimmed = text.trim_matches('\r').trim();
-    if trimmed.is_empty() || trimmed.starts_with(':') || trimmed.starts_with("event:") {
-        return None;
-    }
-    let data_line = trimmed.strip_prefix("data:")?.trim();
-    if data_line.is_empty() || data_line == "[DONE]" {
-        return None;
-    }
-    serde_json::from_str(data_line).ok()
-}
-
-pub(crate) fn resolve_identity(
-    response_id: Option<&str>,
-    model: Option<&str>,
-    report_context: &Value,
-    default_id: &str,
-) -> (String, String) {
-    let id = response_id
-        .filter(|value| !value.is_empty())
-        .unwrap_or(default_id)
-        .to_string();
-    let model = model
-        .filter(|value| !value.is_empty())
-        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or("unknown")
-        .to_string();
-    (id, model)
-}
-
-pub(crate) fn canonical_usage_from_openai_usage(value: Option<&Value>) -> Option<CanonicalUsage> {
-    let usage = value?.as_object()?;
-    let input_tokens = usage
-        .get("input_tokens")
-        .or_else(|| usage.get("prompt_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .get("output_tokens")
-        .or_else(|| usage.get("completion_tokens"))
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = usage
-        .get("total_tokens")
-        .and_then(Value::as_u64)
-        .unwrap_or(input_tokens + output_tokens);
-    Some(CanonicalUsage {
-        input_tokens,
-        output_tokens,
-        total_tokens,
-    })
-}
-
-pub(crate) fn canonical_usage_from_claude_usage(value: Option<&Value>) -> Option<CanonicalUsage> {
-    let usage = value?.as_object()?;
-    let input_tokens = usage
-        .get("input_tokens")
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .get("output_tokens")
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    Some(CanonicalUsage {
-        input_tokens,
-        output_tokens,
-        total_tokens: input_tokens + output_tokens,
-    })
-}
-
-pub(crate) fn canonical_usage_from_gemini_usage(value: Option<&Value>) -> Option<CanonicalUsage> {
-    let usage = value?.as_object()?;
-    let input_tokens = usage
-        .get("promptTokenCount")
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let output_tokens = usage
-        .get("candidatesTokenCount")
-        .and_then(Value::as_u64)
-        .unwrap_or(0);
-    let total_tokens = usage
-        .get("totalTokenCount")
-        .and_then(Value::as_u64)
-        .unwrap_or(input_tokens + output_tokens);
-    Some(CanonicalUsage {
-        input_tokens,
-        output_tokens,
-        total_tokens,
-    })
-}
-
-pub(crate) fn normalize_openai_finish_reason(value: Option<&str>) -> Option<String> {
-    match value {
-        Some("function_call") => Some("tool_calls".to_string()),
-        Some(other) if !other.trim().is_empty() => Some(other.to_string()),
-        _ => None,
-    }
-}
-
-pub(crate) fn map_openai_finish_reason_to_claude(value: Option<&str>) -> &'static str {
-    match value {
-        Some("length") => "max_tokens",
-        Some("tool_calls") | Some("function_call") => "tool_use",
-        Some("content_filter") => "content_filtered",
-        _ => "end_turn",
-    }
-}
-
-pub(crate) fn map_openai_finish_reason_to_gemini(value: Option<&str>) -> &'static str {
-    match value {
-        Some("length") => "MAX_TOKENS",
-        Some("content_filter") => "SAFETY",
-        _ => "STOP",
-    }
-}
-
-pub(crate) fn parse_json_arguments_value(arguments: &str) -> Option<Value> {
-    let trimmed = arguments.trim();
-    if trimmed.is_empty() {
-        return Some(Value::Object(Map::new()));
-    }
-    serde_json::from_str(trimmed).ok()
-}
-
-pub(crate) fn build_openai_chat_chunk(
-    id: &str,
-    model: &str,
-    text: String,
-    tool_calls: Option<Vec<Value>>,
-    finish_reason: Option<&str>,
-) -> Value {
-    let mut delta = Map::new();
-    delta.insert("role".to_string(), Value::String("assistant".to_string()));
-    if !text.is_empty() {
-        delta.insert("content".to_string(), Value::String(text));
-    } else if tool_calls.is_none() {
-        delta.insert("content".to_string(), Value::String(String::new()));
-    }
-    if let Some(tool_calls) = tool_calls {
-        delta.insert("tool_calls".to_string(), Value::Array(tool_calls));
-    }
-
-    json!({
-        "id": id,
-        "object": "chat.completion.chunk",
-        "model": model,
-        "choices": [{
-            "index": 0,
-            "delta": Value::Object(delta),
-            "finish_reason": finish_reason,
-        }]
-    })
-}
-
-pub(crate) fn build_openai_chat_role_chunk(id: &str, model: &str) -> Value {
-    json!({
-        "id": id,
-        "object": "chat.completion.chunk",
-        "model": model,
-        "choices": [{
-            "index": 0,
-            "delta": {
-                "role": "assistant"
-            },
-            "finish_reason": Value::Null
-        }]
-    })
-}
-
-pub(crate) fn build_openai_chat_finish_chunk(
-    id: &str,
-    model: &str,
-    finish_reason: Option<&str>,
-) -> Value {
-    json!({
-        "id": id,
-        "object": "chat.completion.chunk",
-        "model": model,
-        "choices": [{
-            "index": 0,
-            "delta": {},
-            "finish_reason": finish_reason,
-        }]
-    })
-}

apps/aether-gateway/src/ai_pipeline/finalize/standard/stream_core/mod.rs

@@ -1,100 +1,7 @@
 //! Standard finalize streaming conversion helpers.
-
+pub use aether_ai_pipeline::finalize::standard::stream_core::CanonicalStreamFrame;
-use serde_json::Value;
-
-use crate::GatewayError;
-
-use super::claude::stream::{ClaudeClientEmitter, ClaudeProviderState};
-use super::gemini::stream::{GeminiClientEmitter, GeminiProviderState};
-use super::openai::stream::{
-    OpenAIChatClientEmitter, OpenAIChatProviderState, OpenAICliClientEmitter,
-    OpenAICliProviderState,
-};
 
 pub(crate) mod common;
 mod orchestrator;
 
-use common::CanonicalStreamFrame;
-
-pub(crate) enum ProviderStreamParser {
-    OpenAIChat(OpenAIChatProviderState),
-    OpenAICli(OpenAICliProviderState),
-    Claude(ClaudeProviderState),
-    Gemini(GeminiProviderState),
-}
-
-impl ProviderStreamParser {
-    pub(crate) fn for_api_format(provider_api_format: &str) -> Option<Self> {
-        Some(match provider_api_format {
-            "openai:chat" => Self::OpenAIChat(OpenAIChatProviderState::default()),
-            "openai:cli" | "openai:compact" => Self::OpenAICli(OpenAICliProviderState::default()),
-            "claude:chat" | "claude:cli" => Self::Claude(ClaudeProviderState::default()),
-            "gemini:chat" | "gemini:cli" => Self::Gemini(GeminiProviderState::default()),
-            _ => return None,
-        })
-    }
-
-    pub(crate) fn push_line(
-        &mut self,
-        report_context: &Value,
-        line: Vec<u8>,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        match self {
-            ProviderStreamParser::OpenAIChat(state) => state.push_line(report_context, line),
-            ProviderStreamParser::OpenAICli(state) => state.push_line(report_context, line),
-            ProviderStreamParser::Claude(state) => state.push_line(report_context, line),
-            ProviderStreamParser::Gemini(state) => state.push_line(report_context, line),
-        }
-    }
-
-    pub(crate) fn finish(
-        &mut self,
-        report_context: &Value,
-    ) -> Result<Vec<CanonicalStreamFrame>, GatewayError> {
-        match self {
-            ProviderStreamParser::OpenAIChat(state) => state.finish(report_context),
-            ProviderStreamParser::OpenAICli(state) => state.finish(report_context),
-            ProviderStreamParser::Claude(state) => state.finish(report_context),
-            ProviderStreamParser::Gemini(state) => state.finish(report_context),
-        }
-    }
-}
-
-pub(crate) enum ClientStreamEmitter {
-    OpenAIChat(OpenAIChatClientEmitter),
-    OpenAICli(OpenAICliClientEmitter),
-    Claude(ClaudeClientEmitter),
-    Gemini(GeminiClientEmitter),
-}
-
-impl ClientStreamEmitter {
-    pub(crate) fn for_api_format(client_api_format: &str) -> Option<Self> {
-        Some(match client_api_format {
-            "openai:chat" => Self::OpenAIChat(OpenAIChatClientEmitter::default()),
-            "openai:cli" | "openai:compact" => Self::OpenAICli(OpenAICliClientEmitter::default()),
-            "claude:chat" | "claude:cli" => Self::Claude(ClaudeClientEmitter::default()),
-            "gemini:chat" | "gemini:cli" => Self::Gemini(GeminiClientEmitter::default()),
-            _ => return None,
-        })
-    }
-
-    pub(crate) fn emit(&mut self, frame: CanonicalStreamFrame) -> Result<Vec<u8>, GatewayError> {
-        match self {
-            ClientStreamEmitter::OpenAIChat(state) => state.emit(frame),
-            ClientStreamEmitter::OpenAICli(state) => state.emit(frame),
-            ClientStreamEmitter::Claude(state) => state.emit(frame),
-            ClientStreamEmitter::Gemini(state) => state.emit(frame),
-        }
-    }
-
-    pub(crate) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
-        match self {
-            ClientStreamEmitter::OpenAIChat(state) => state.finish(),
-            ClientStreamEmitter::OpenAICli(state) => state.finish(),
-            ClientStreamEmitter::Claude(state) => state.finish(),
-            ClientStreamEmitter::Gemini(state) => state.finish(),
-        }
-    }
-}
-
 pub(crate) use orchestrator::StreamingStandardConversionState;

apps/aether-gateway/src/ai_pipeline/finalize/standard/stream_core/orchestrator.rs

@@ -3,14 +3,11 @@ use serde_json::Value;
 use crate::ai_pipeline::adaptation::private_envelope::transform_provider_private_stream_line as transform_envelope_line;
 use crate::ai_pipeline::adaptation::surfaces::provider_adaptation_should_unwrap_stream_envelope;
 use crate::GatewayError;
-
+use aether_ai_pipeline::finalize::standard::stream_core::StreamingStandardFormatMatrix;
-use super::common::CanonicalStreamFrame;
-use super::{ClientStreamEmitter, ProviderStreamParser};
 
 #[derive(Default)]
 pub(crate) struct StreamingStandardConversionState {
-    provider: Option<ProviderStreamParser>,
+    matrix: StreamingStandardFormatMatrix,
-    client: Option<ClientStreamEmitter>,
 }
 
 impl StreamingStandardConversionState {
@@ -19,7 +16,6 @@ impl StreamingStandardConversionState {
         report_context: &Value,
         line: Vec<u8>,
     ) -> Result<Vec<u8>, GatewayError> {
-        self.ensure_initialized(report_context)?;
         let line = if should_unwrap_envelope(report_context) {
             transform_envelope_line(report_context, line)?
         } else {
@@ -28,59 +24,13 @@ impl StreamingStandardConversionState {
         if line.is_empty() {
             return Ok(Vec::new());
         }
-        let Some(provider) = self.provider.as_mut() else {
+        self.matrix
-            return Ok(Vec::new());
+            .transform_line(report_context, line)
-        };
+            .map_err(Into::into)
-        let frames = provider.push_line(report_context, line)?;
-        self.emit_frames(frames)
     }
 
     pub(crate) fn finish(&mut self, report_context: &Value) -> Result<Vec<u8>, GatewayError> {
-        self.ensure_initialized(report_context)?;
+        self.matrix.finish(report_context).map_err(Into::into)
-        let Some(provider) = self.provider.as_mut() else {
-            return Ok(Vec::new());
-        };
-        let frames = provider.finish(report_context)?;
-        let mut out = self.emit_frames(frames)?;
-        if let Some(client) = self.client.as_mut() {
-            out.extend(client.finish()?);
-        }
-        Ok(out)
-    }
-
-    fn ensure_initialized(&mut self, report_context: &Value) -> Result<(), GatewayError> {
-        if self.provider.is_some() && self.client.is_some() {
-            return Ok(());
-        }
-
-        let provider_api_format = report_context
-            .get("provider_api_format")
-            .and_then(Value::as_str)
-            .unwrap_or_default()
-            .trim()
-            .to_ascii_lowercase();
-        let client_api_format = report_context
-            .get("client_api_format")
-            .and_then(Value::as_str)
-            .unwrap_or_default()
-            .trim()
-            .to_ascii_lowercase();
-
-        self.provider = ProviderStreamParser::for_api_format(provider_api_format.as_str());
-        self.client = ClientStreamEmitter::for_api_format(client_api_format.as_str());
-
-        Ok(())
-    }
-
-    fn emit_frames(&mut self, frames: Vec<CanonicalStreamFrame>) -> Result<Vec<u8>, GatewayError> {
-        let Some(client) = self.client.as_mut() else {
-            return Ok(Vec::new());
-        };
-        let mut out = Vec::new();
-        for frame in frames {
-            out.extend(client.emit(frame)?);
-        }
-        Ok(out)
     }
 }
 

apps/aether-gateway/src/ai_pipeline/finalize/tests_sync.rs

@@ -10,7 +10,10 @@ use super::{
     convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
     maybe_build_local_core_sync_finalize_response,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
+use crate::ai_pipeline::conversion::response::{
+    convert_openai_chat_response_to_openai_cli, convert_openai_cli_response_to_openai_chat,
+};
 use crate::usage::GatewaySyncReportRequest;
 
 fn test_decision() -> GatewayControlDecision {
@@ -368,6 +371,39 @@ fn converts_claude_chat_tool_use_to_openai_chat_tool_calls() {
     );
 }
 
+#[test]
+fn converts_claude_chat_thinking_block_to_openai_reasoning_content() {
+    let result = convert_claude_chat_response_to_openai_chat(
+        &json!({
+            "id": "msg_think_123",
+            "type": "message",
+            "role": "assistant",
+            "model": "claude-sonnet-4-upstream",
+            "content": [
+                {"type": "thinking", "thinking": "Need to reason first."},
+                {"type": "text", "text": "Final answer"}
+            ],
+            "stop_reason": "end_turn",
+            "usage": {
+                "input_tokens": 5,
+                "output_tokens": 7
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "claude:chat",
+            "model": "gpt-5"
+        }),
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["choices"][0]["message"]["reasoning_content"],
+        "Need to reason first."
+    );
+    assert_eq!(result["choices"][0]["message"]["content"], "Final answer");
+}
+
 #[test]
 fn converts_gemini_chat_response_to_openai_chat_response() {
     let result = convert_gemini_chat_response_to_openai_chat(
@@ -481,6 +517,223 @@ fn converts_gemini_chat_function_call_to_openai_chat_tool_calls() {
     );
 }
 
+#[test]
+fn converts_gemini_chat_thought_part_to_openai_reasoning_content() {
+    let result = convert_gemini_chat_response_to_openai_chat(
+        &json!({
+            "responseId": "resp_think_123",
+            "candidates": [{
+                "content": {
+                    "parts": [
+                        {"text": "Internal reasoning.", "thought": true},
+                        {"text": "Visible answer"}
+                    ],
+                    "role": "model"
+                },
+                "finishReason": "STOP",
+                "index": 0
+            }],
+            "modelVersion": "gemini-2.5-pro-upstream",
+            "usageMetadata": {
+                "promptTokenCount": 1,
+                "candidatesTokenCount": 2,
+                "totalTokenCount": 3
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "gemini:chat",
+            "model": "gpt-5"
+        }),
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["choices"][0]["message"]["reasoning_content"],
+        "Internal reasoning."
+    );
+    assert_eq!(result["choices"][0]["message"]["content"], "Visible answer");
+}
+
+#[test]
+fn converts_gemini_chat_inline_data_to_openai_chat_image_part() {
+    let result = convert_gemini_chat_response_to_openai_chat(
+        &json!({
+            "responseId": "resp_img_123",
+            "candidates": [{
+                "content": {
+                    "parts": [{
+                        "inlineData": {
+                            "mimeType": "image/png",
+                            "data": "iVBORw0KGgo="
+                        }
+                    }],
+                    "role": "model"
+                },
+                "finishReason": "STOP",
+                "index": 0
+            }],
+            "modelVersion": "gemini-2.5-pro-upstream",
+            "usageMetadata": {
+                "promptTokenCount": 1,
+                "candidatesTokenCount": 2,
+                "totalTokenCount": 3
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "gemini:chat",
+            "model": "gpt-5"
+        }),
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["choices"][0]["message"]["content"],
+        json!([{
+            "type": "image_url",
+            "image_url": {
+                "url": "data:image/png;base64,iVBORw0KGgo="
+            }
+        }])
+    );
+}
+
+#[test]
+fn converts_openai_cli_reasoning_item_to_openai_chat_reasoning_content() {
+    let result = convert_openai_cli_response_to_openai_chat(
+        &json!({
+            "id": "resp_reason_123",
+            "object": "response",
+            "model": "gpt-5",
+            "output": [
+                {
+                    "type": "reasoning",
+                    "id": "rs_1",
+                    "summary": [{
+                        "type": "summary_text",
+                        "text": "Thinking summary."
+                    }]
+                },
+                {
+                    "type": "message",
+                    "id": "msg_1",
+                    "role": "assistant",
+                    "status": "completed",
+                    "content": [{
+                        "type": "output_text",
+                        "text": "Final answer",
+                        "annotations": []
+                    }]
+                }
+            ],
+            "usage": {
+                "input_tokens": 3,
+                "output_tokens": 5,
+                "total_tokens": 8
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "openai:cli",
+            "model": "gpt-5"
+        }),
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["choices"][0]["message"]["reasoning_content"],
+        "Thinking summary."
+    );
+    assert_eq!(result["choices"][0]["message"]["content"], "Final answer");
+}
+
+#[test]
+fn converts_openai_cli_output_image_to_openai_chat_image_part() {
+    let result = convert_openai_cli_response_to_openai_chat(
+        &json!({
+            "id": "resp_img_cli_123",
+            "object": "response",
+            "model": "gpt-5",
+            "output": [{
+                "type": "message",
+                "id": "msg_1",
+                "role": "assistant",
+                "status": "completed",
+                "content": [{
+                    "type": "output_image",
+                    "image_url": "data:image/png;base64,iVBORw0KGgo="
+                }]
+            }],
+            "usage": {
+                "input_tokens": 3,
+                "output_tokens": 5,
+                "total_tokens": 8
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "openai:cli",
+            "model": "gpt-5"
+        }),
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["choices"][0]["message"]["content"],
+        json!([{
+            "type": "image_url",
+            "image_url": {
+                "url": "data:image/png;base64,iVBORw0KGgo="
+            }
+        }])
+    );
+}
+
+#[test]
+fn converts_openai_chat_image_part_to_openai_cli_output_image() {
+    let result = convert_openai_chat_response_to_openai_cli(
+        &json!({
+            "id": "chatcmpl_img_123",
+            "object": "chat.completion",
+            "model": "gpt-5",
+            "choices": [{
+                "index": 0,
+                "message": {
+                    "role": "assistant",
+                    "content": [{
+                        "type": "image_url",
+                        "image_url": {
+                            "url": "data:image/png;base64,iVBORw0KGgo="
+                        }
+                    }]
+                },
+                "finish_reason": "stop"
+            }],
+            "usage": {
+                "prompt_tokens": 3,
+                "completion_tokens": 5,
+                "total_tokens": 8
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:cli",
+            "provider_api_format": "openai:chat",
+            "model": "gpt-5"
+        }),
+        false,
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["output"][0]["content"],
+        json!([{
+            "type": "output_image",
+            "image_url": "data:image/png;base64,iVBORw0KGgo="
+        }])
+    );
+}
+
 #[test]
 fn converts_claude_cli_response_to_openai_cli_response() {
     let result = convert_claude_cli_response_to_openai_cli(
@@ -721,6 +974,49 @@ fn converts_gemini_cli_function_call_to_openai_cli_function_call() {
     );
 }
 
+#[test]
+fn converts_gemini_cli_inline_data_to_openai_cli_output_image() {
+    let result = convert_gemini_cli_response_to_openai_cli(
+        &json!({
+            "responseId": "resp_cli_img_123",
+            "candidates": [{
+                "content": {
+                    "parts": [{
+                        "inlineData": {
+                            "mimeType": "image/png",
+                            "data": "iVBORw0KGgo="
+                        }
+                    }],
+                    "role": "model"
+                },
+                "finishReason": "STOP",
+                "index": 0
+            }],
+            "modelVersion": "gemini-cli-upstream",
+            "usageMetadata": {
+                "promptTokenCount": 3,
+                "candidatesTokenCount": 5,
+                "thoughtsTokenCount": 2,
+                "totalTokenCount": 10
+            }
+        }),
+        &json!({
+            "client_api_format": "openai:cli",
+            "provider_api_format": "gemini:cli",
+            "model": "gpt-5"
+        }),
+    )
+    .expect("result should exist");
+
+    assert_eq!(
+        result["output"][0]["content"],
+        json!([{
+            "type": "output_image",
+            "image_url": "data:image/png;base64,iVBORw0KGgo="
+        }])
+    );
+}
+
 #[test]
 fn local_finalize_handles_openai_compact_cross_format_sync_response() {
     let payload = GatewaySyncReportRequest {

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

@@ -1,6 +1,9 @@
 pub(crate) mod adaptation;
 pub(crate) mod contracts;
+pub(crate) mod control_facade;
 pub(crate) mod conversion;
+pub(crate) mod execution_facade;
 pub(crate) mod finalize;
 pub(crate) mod planner;
+pub(crate) mod provider_transport_facade;
 pub(crate) mod runtime;

apps/aether-gateway/src/ai_pipeline/planner/auth_snapshot_facade.rs

@@ -0,0 +1,15 @@
+pub(crate) use crate::data::auth::GatewayAuthApiKeySnapshot;
+use crate::{AppState, GatewayError};
+
+pub(crate) async fn read_auth_api_key_snapshot(
+    state: &AppState,
+    user_id: &str,
+    api_key_id: &str,
+    now_unix_secs: u64,
+) -> Result<Option<GatewayAuthApiKeySnapshot>, GatewayError> {
+    state
+        .data
+        .read_auth_api_key_snapshot(user_id, api_key_id, now_unix_secs)
+        .await
+        .map_err(|err| GatewayError::Internal(err.to_string()))
+}

apps/aether-gateway/src/ai_pipeline/planner/candidate_affinity.rs

@@ -1,8 +1,9 @@
 use tracing::warn;
 
-use crate::provider_transport::resolve_transport_proxy_snapshot;
+use crate::ai_pipeline::planner::transport_facade::read_provider_transport_snapshot;
-use crate::scheduler::GatewayMinimalCandidateSelectionCandidate;
+use crate::ai_pipeline::provider_transport_facade::resolve_transport_proxy_snapshot;
 use crate::AppState;
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
 enum TunnelOwnerAffinityBucket {
@@ -13,8 +14,8 @@ enum TunnelOwnerAffinityBucket {
 
 pub(crate) async fn prefer_local_tunnel_owner_candidates(
     state: &AppState,
-    candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
+    candidates: Vec<SchedulerMinimalCandidateSelectionCandidate>,
-) -> Vec<GatewayMinimalCandidateSelectionCandidate> {
+) -> Vec<SchedulerMinimalCandidateSelectionCandidate> {
     let mut ranked = Vec::with_capacity(candidates.len());
     for (original_index, candidate) in candidates.into_iter().enumerate() {
         let bucket = resolve_candidate_tunnel_owner_affinity(state, &candidate).await;
@@ -29,10 +30,10 @@ pub(crate) async fn prefer_local_tunnel_owner_candidates(
 
 async fn resolve_candidate_tunnel_owner_affinity(
     state: &AppState,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
 ) -> TunnelOwnerAffinityBucket {
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -101,14 +102,14 @@ async fn resolve_candidate_tunnel_owner_affinity(
 mod tests {
     use std::time::{SystemTime, UNIX_EPOCH};
 
-    use aether_data::repository::provider_catalog::{
+    use aether_data::repository::provider_catalog::InMemoryProviderCatalogReadRepository;
-        InMemoryProviderCatalogReadRepository, StoredProviderCatalogEndpoint,
+    use aether_data_contracts::repository::provider_catalog::{
-        StoredProviderCatalogKey, StoredProviderCatalogProvider,
+        StoredProviderCatalogEndpoint, StoredProviderCatalogKey, StoredProviderCatalogProvider,
     };
     use serde_json::json;
 
     use super::{
-        prefer_local_tunnel_owner_candidates, AppState, GatewayMinimalCandidateSelectionCandidate,
+        prefer_local_tunnel_owner_candidates, AppState, SchedulerMinimalCandidateSelectionCandidate,
     };
     use crate::data::GatewayDataState;
     use crate::tunnel::TunnelAttachmentRecord;
@@ -116,8 +117,8 @@ mod tests {
     fn sample_candidate(
         endpoint_id: &str,
         key_id: &str,
-    ) -> GatewayMinimalCandidateSelectionCandidate {
+    ) -> SchedulerMinimalCandidateSelectionCandidate {
-        GatewayMinimalCandidateSelectionCandidate {
+        SchedulerMinimalCandidateSelectionCandidate {
             provider_id: "provider-1".to_string(),
             provider_name: "provider-1".to_string(),
             provider_type: "custom".to_string(),

apps/aether-gateway/src/ai_pipeline/planner/candidate_runtime_facade.rs

@@ -0,0 +1,60 @@
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
+use serde_json::Value;
+
+use crate::{AppState, GatewayError};
+
+#[allow(clippy::too_many_arguments)]
+pub(crate) async fn persist_available_local_candidate(
+    state: &AppState,
+    trace_id: &str,
+    user_id: &str,
+    api_key_id: &str,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
+    candidate_index: u32,
+    candidate_id: &str,
+    extra_data: Option<Value>,
+    created_at_unix_secs: u64,
+    error_context: &'static str,
+) -> String {
+    crate::request_candidate_runtime::persist_available_local_candidate(
+        state,
+        trace_id,
+        user_id,
+        api_key_id,
+        candidate,
+        candidate_index,
+        candidate_id,
+        extra_data,
+        created_at_unix_secs,
+        error_context,
+    )
+    .await
+}
+
+#[allow(clippy::too_many_arguments)]
+pub(crate) async fn persist_skipped_local_candidate(
+    state: &AppState,
+    trace_id: &str,
+    user_id: &str,
+    api_key_id: &str,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
+    candidate_index: u32,
+    candidate_id: &str,
+    skip_reason: &str,
+    finished_at_unix_secs: u64,
+    error_context: &'static str,
+) {
+    crate::request_candidate_runtime::persist_skipped_local_candidate(
+        state,
+        trace_id,
+        user_id,
+        api_key_id,
+        candidate,
+        candidate_index,
+        candidate_id,
+        skip_reason,
+        finished_at_unix_secs,
+        error_context,
+    )
+    .await
+}

apps/aether-gateway/src/ai_pipeline/planner/common.rs

@@ -1,7 +1,7 @@
 use axum::body::Bytes;
-use base64::Engine as _;
 
-pub(crate) use crate::ai_pipeline::contracts::{
+use crate::ai_pipeline::control_facade::is_json_request;
+pub(crate) use aether_ai_pipeline::contracts::{
     CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
     CLAUDE_CLI_SYNC_PLAN_KIND, EXECUTION_RUNTIME_STREAM_ACTION,
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
@@ -15,25 +15,45 @@ pub(crate) use crate::ai_pipeline::contracts::{
     OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
     OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND, OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
 };
-use crate::headers::is_json_request;
 
 pub(crate) fn parse_direct_request_body(
     parts: &http::request::Parts,
     body_bytes: &Bytes,
 ) -> Option<(serde_json::Value, Option<String>)> {
-    if is_json_request(&parts.headers) {
+    aether_ai_pipeline::planner::common::parse_direct_request_body(
-        if body_bytes.is_empty() {
+        is_json_request(&parts.headers),
-            Some((serde_json::json!({}), None))
+        body_bytes.as_ref(),
-        } else {
+    )
-            serde_json::from_slice::<serde_json::Value>(body_bytes)
-                .ok()
-                .map(|value| (value, None))
 }
-    } else {
+
-        Some((
+pub(crate) fn force_upstream_streaming_for_provider(
-            serde_json::json!({}),
+    provider_type: &str,
-            (!body_bytes.is_empty())
+    provider_api_format: &str,
-                .then(|| base64::engine::general_purpose::STANDARD.encode(body_bytes)),
+) -> bool {
-        ))
+    provider_type.trim().eq_ignore_ascii_case("codex")
+        && provider_api_format
+            .trim()
+            .eq_ignore_ascii_case("openai:cli")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::force_upstream_streaming_for_provider;
+
+    #[test]
+    fn forces_streaming_for_codex_openai_cli() {
+        assert!(force_upstream_streaming_for_provider("codex", "openai:cli"));
+    }
+
+    #[test]
+    fn does_not_force_streaming_for_compact_or_other_provider_types() {
+        assert!(!force_upstream_streaming_for_provider(
+            "codex",
+            "openai:compact"
+        ));
+        assert!(!force_upstream_streaming_for_provider(
+            "openai",
+            "openai:cli"
+        ));
     }
 }

apps/aether-gateway/src/ai_pipeline/planner/decision/control_plan.rs

@@ -1,3 +1,4 @@
+use crate::ai_pipeline::control_facade::{GatewayControlAuthContext, GatewayControlDecision};
 use crate::ai_pipeline::planner::common::{
     CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
     CLAUDE_CLI_SYNC_PLAN_KIND, EXECUTION_RUNTIME_STREAM_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
@@ -18,8 +19,6 @@ use crate::ai_pipeline::planner::plan_builders::{
     build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlAuthContext;
-use crate::control::GatewayControlDecision;
 use crate::{
     AppState, GatewayControlPlanResponse, GatewayControlSyncDecisionResponse, GatewayError,
 };

apps/aether-gateway/src/ai_pipeline/planner/decision/mod.rs

@@ -23,7 +23,7 @@ pub(crate) use super::standard::{
     maybe_build_sync_local_openai_cli_decision_payload,
     maybe_build_sync_local_standard_decision_payload,
 };
-pub(crate) use crate::scheduler::{
+pub(crate) use crate::ai_pipeline::planner::{
     resolve_execution_runtime_stream_plan_kind as resolve_stream_plan_kind,
     resolve_execution_runtime_sync_plan_kind as resolve_sync_plan_kind,
 };

apps/aether-gateway/src/ai_pipeline/planner/decision/stream.rs

@@ -1,11 +1,11 @@
 use std::collections::BTreeMap;
 
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
 use crate::ai_pipeline::planner::common::{
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, OPENAI_VIDEO_CONTENT_PLAN_KIND,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::planner::{
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
-use crate::scheduler::{
     is_matching_stream_request, resolve_execution_runtime_stream_plan_kind,
 };
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};

apps/aether-gateway/src/ai_pipeline/planner/decision/sync.rs

@@ -2,16 +2,17 @@ use std::collections::BTreeMap;
 
 use url::Url;
 
+use crate::ai_pipeline::control_facade::{
+    resolve_execution_runtime_auth_context, GatewayControlDecision,
+};
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
 use crate::ai_pipeline::planner::common::{
     EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_FILES_DELETE_PLAN_KIND,
     GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
     OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND, OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND,
     OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
 };
-use crate::control::resolve_execution_runtime_auth_context;
+use crate::ai_pipeline::planner::resolve_execution_runtime_sync_plan_kind;
-use crate::control::GatewayControlDecision;
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
-use crate::scheduler::resolve_execution_runtime_sync_plan_kind;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
 pub(crate) async fn maybe_build_sync_decision_payload(

apps/aether-gateway/src/ai_pipeline/planner/executor_facade.rs

@@ -0,0 +1,16 @@
+use aether_contracts::ExecutionPlan;
+use serde_json::Value;
+
+use crate::AppState;
+
+pub(crate) async fn mark_unused_local_candidate_items<T, FPlan, FContext>(
+    state: &AppState,
+    remaining: Vec<T>,
+    plan: FPlan,
+    report_context: FContext,
+) where
+    FPlan: Fn(&T) -> &ExecutionPlan,
+    FContext: Fn(&T) -> Option<&Value>,
+{
+    crate::executor::mark_unused_local_candidate_items(state, remaining, plan, report_context).await
+}

apps/aether-gateway/src/ai_pipeline/planner/mod.rs

@@ -1,16 +1,28 @@
 use crate::ai_pipeline::contracts::{
     GatewayControlPlanResponse, GatewayControlSyncDecisionResponse,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::{AppState, GatewayError};
 
+pub(crate) mod auth_snapshot_facade;
 pub(crate) mod candidate_affinity;
+pub(crate) mod candidate_runtime_facade;
 pub(crate) mod common;
 mod decision;
+pub(crate) mod executor_facade;
 pub(crate) mod passthrough;
 pub(crate) mod plan_builders;
+mod route;
+pub(crate) mod scheduler_facade;
 pub(crate) mod specialized;
 pub(crate) mod standard;
+pub(crate) mod transport_facade;
+
+pub(crate) use self::route::{
+    is_matching_stream_request, resolve_execution_runtime_stream_plan_kind,
+    resolve_execution_runtime_sync_plan_kind, supports_stream_scheduler_decision_kind,
+    supports_sync_scheduler_decision_kind,
+};
 
 pub(crate) async fn maybe_build_sync_decision_payload(
     state: &AppState,

apps/aether-gateway/src/ai_pipeline/planner/passthrough/mod.rs

@@ -6,4 +6,4 @@ pub(crate) use self::provider::{
     maybe_build_stream_local_same_format_provider_decision_payload,
     maybe_build_sync_local_same_format_provider_decision_payload,
 };
-pub(crate) use crate::provider_transport::provider_types::provider_type_supports_local_same_format_transport;
+pub(crate) use crate::ai_pipeline::provider_transport_facade::provider_types::provider_type_supports_local_same_format_transport;

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider.rs

@@ -3,11 +3,16 @@ use axum::http::Response;
 use std::collections::BTreeMap;
 use url::form_urlencoded;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use aether_data_contracts::repository::candidates::{
+    RequestCandidateStatus, UpsertRequestCandidateRecord,
+};
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 use serde_json::{json, Value};
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::ai_pipeline::control_facade::{collect_control_headers, GatewayControlDecision};
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
 use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
 use crate::ai_pipeline::planner::common::{
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
@@ -15,47 +20,41 @@ use crate::ai_pipeline::planner::common::{
 use crate::ai_pipeline::planner::plan_builders::{
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::provider_transport_facade::antigravity::{
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
-use crate::headers::collect_control_headers;
-use crate::provider_transport::antigravity::{
     build_antigravity_safe_v1internal_request, build_antigravity_static_identity_headers,
     build_antigravity_v1internal_url, classify_local_antigravity_request_support,
     AntigravityEnvelopeRequestType, AntigravityRequestEnvelopeSupport,
     AntigravityRequestSideSupport, AntigravityRequestUrlAction,
 };
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_openai_passthrough_headers, resolve_local_gemini_auth, resolve_local_standard_auth,
 };
-use crate::provider_transport::claude_code::{
+use crate::ai_pipeline::provider_transport_facade::claude_code::{
     build_claude_code_messages_url, build_claude_code_passthrough_headers,
     sanitize_claude_code_request_body, supports_local_claude_code_transport_with_network,
 };
-use crate::provider_transport::kiro::{
+use crate::ai_pipeline::provider_transport_facade::kiro::{
     build_kiro_generate_assistant_response_url, build_kiro_provider_headers,
     build_kiro_provider_request_body, supports_local_kiro_request_transport_with_network,
     KIRO_ENVELOPE_NAME,
 };
-use crate::provider_transport::policy::{
+use crate::ai_pipeline::provider_transport_facade::policy::{
     supports_local_gemini_transport_with_network, supports_local_standard_transport_with_network,
 };
-use crate::provider_transport::url::{
+use crate::ai_pipeline::provider_transport_facade::url::{
     build_claude_messages_url, build_gemini_content_url, build_passthrough_path_url,
 };
-use crate::provider_transport::vertex::{
+use crate::ai_pipeline::provider_transport_facade::vertex::{
     build_vertex_api_key_gemini_content_url, resolve_local_vertex_api_key_query_auth,
     supports_local_vertex_api_key_gemini_transport_with_network,
 };
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     apply_local_body_rules, apply_local_header_rules, build_passthrough_headers,
     ensure_upstream_auth_header, resolve_transport_execution_timeouts,
     resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
     LocalResolvedOAuthRequestAuth,
 };
-use crate::scheduler::{
+use crate::clock::current_unix_secs;
-    current_unix_secs, list_selectable_candidates, record_local_request_candidate_status,
-    GatewayMinimalCandidateSelectionCandidate,
-};
 use crate::{
     append_execution_contract_fields_to_value, AppState, GatewayControlSyncDecisionResponse,
     GatewayError,

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/build.rs

@@ -1,4 +1,4 @@
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
 use super::super::plans::{resolve_stream_spec, resolve_sync_spec};

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/candidates.rs

@@ -1,12 +1,14 @@
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
 use serde_json::json;
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::auth_snapshot_facade::read_auth_api_key_snapshot;
 use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::planner::candidate_runtime_facade::persist_available_local_candidate;
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::scheduler_facade::list_selectable_candidates;
-use crate::scheduler::{current_unix_secs, list_selectable_candidates};
+use crate::clock::current_unix_secs;
 use crate::{append_execution_contract_fields_to_value, AppState, GatewayError};
 
 use super::types::{
@@ -40,8 +42,8 @@ pub(crate) async fn resolve_local_same_format_provider_decision_input(
         }
     };
 
-    let auth_snapshot = match state
+    let auth_snapshot = match read_auth_api_key_snapshot(
-        .read_auth_api_key_snapshot(
+        state,
         &auth_context.user_id,
         &auth_context.api_key_id,
         current_unix_secs(),
@@ -107,47 +109,19 @@ pub(crate) async fn materialize_local_same_format_provider_candidate_attempts(
             spec.api_format,
         );
 
-        let candidate_id = match state
+        let candidate_id = persist_available_local_candidate(
-            .upsert_request_candidate(UpsertRequestCandidateRecord {
+            state,
-                id: generated_candidate_id.clone(),
+            trace_id,
-                request_id: trace_id.to_string(),
+            &input.auth_context.user_id,
-                user_id: Some(input.auth_context.user_id.clone()),
+            &input.auth_context.api_key_id,
-                api_key_id: Some(input.auth_context.api_key_id.clone()),
+            &candidate,
-                username: None,
+            candidate_index as u32,
-                api_key_name: None,
+            &generated_candidate_id,
-                candidate_index: candidate_index as u32,
+            Some(extra_data),
-                retry_index: 0,
+            created_at_unix_secs,
-                provider_id: Some(candidate.provider_id.clone()),
+            "gateway local same-format decision request candidate upsert failed",
-                endpoint_id: Some(candidate.endpoint_id.clone()),
+        )
-                key_id: Some(candidate.key_id.clone()),
+        .await;
-                status: RequestCandidateStatus::Available,
-                skip_reason: None,
-                is_cached: Some(false),
-                status_code: None,
-                error_type: None,
-                error_message: None,
-                latency_ms: None,
-                concurrent_requests: None,
-                extra_data: Some(extra_data),
-                required_capabilities: candidate.key_capabilities.clone(),
-                created_at_unix_secs: Some(created_at_unix_secs),
-                started_at_unix_secs: None,
-                finished_at_unix_secs: None,
-            })
-            .await
-        {
-            Ok(Some(stored)) => stored.id,
-            Ok(None) => generated_candidate_id.clone(),
-            Err(err) => {
-                warn!(
-                    trace_id = %trace_id,
-                    api_format = spec.api_format,
-                    error = ?err,
-                    "gateway local same-format decision request candidate upsert failed"
-                );
-                generated_candidate_id.clone()
-            }
-        };
 
         attempts.push(LocalSameFormatProviderCandidateAttempt {
             candidate,

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/payload.rs

@@ -1,43 +1,48 @@
 use std::collections::BTreeMap;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
 use serde_json::json;
 use tracing::warn;
 
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::control_facade::collect_control_headers;
-use crate::headers::collect_control_headers;
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
-use crate::provider_transport::antigravity::{
+use crate::ai_pipeline::planner::candidate_runtime_facade::persist_skipped_local_candidate;
+use crate::ai_pipeline::planner::transport_facade::{
+    read_provider_transport_snapshot, resolve_local_oauth_request_auth,
+    LocalResolvedOAuthRequestAuth,
+};
+use crate::ai_pipeline::provider_transport_facade::antigravity::{
     build_antigravity_safe_v1internal_request, build_antigravity_static_identity_headers,
     classify_local_antigravity_request_support, AntigravityEnvelopeRequestType,
     AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport,
 };
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_openai_passthrough_headers, resolve_local_gemini_auth, resolve_local_standard_auth,
 };
-use crate::provider_transport::claude_code::{
+use crate::ai_pipeline::provider_transport_facade::claude_code::{
     build_claude_code_passthrough_headers, supports_local_claude_code_transport_with_network,
 };
-use crate::provider_transport::kiro::{
+use crate::ai_pipeline::provider_transport_facade::kiro::{
     build_kiro_provider_headers, supports_local_kiro_request_transport_with_network,
     KIRO_ENVELOPE_NAME,
 };
-use crate::provider_transport::policy::{
+use crate::ai_pipeline::provider_transport_facade::policy::{
     supports_local_gemini_transport_with_network, supports_local_standard_transport_with_network,
 };
-use crate::provider_transport::vertex::{
+use crate::ai_pipeline::provider_transport_facade::vertex::{
     resolve_local_vertex_api_key_query_auth,
     supports_local_vertex_api_key_gemini_transport_with_network,
 };
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     apply_local_header_rules, build_passthrough_headers, ensure_upstream_auth_header,
     resolve_transport_execution_timeouts, resolve_transport_proxy_snapshot_with_tunnel_affinity,
-    resolve_transport_tls_profile, LocalResolvedOAuthRequestAuth,
+    resolve_transport_tls_profile,
 };
-use crate::scheduler::{current_unix_secs, GatewayMinimalCandidateSelectionCandidate};
+use crate::clock::current_unix_secs;
 use crate::{
     append_execution_contract_fields_to_value, AppState, GatewayControlSyncDecisionResponse,
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
 };
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 
 use super::types::{
     LocalSameFormatProviderCandidateAttempt, LocalSameFormatProviderDecisionInput,
@@ -59,8 +64,8 @@ pub(crate) async fn maybe_build_local_same_format_provider_decision_payload_for_
         candidate_id,
     } = attempt;
 
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -169,7 +174,7 @@ pub(crate) async fn maybe_build_local_same_format_provider_decision_payload_for_
             && !is_vertex
             && resolve_local_gemini_auth(&transport).is_none();
     let oauth_auth = if should_try_oauth_auth {
-        match state.resolve_local_oauth_request_auth(&transport).await {
+        match resolve_local_oauth_request_auth(state, &transport).await {
             Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(auth))) => {
                 Some(LocalResolvedOAuthRequestAuth::Kiro(auth))
             }
@@ -528,46 +533,22 @@ pub(super) async fn mark_skipped_local_same_format_provider_candidate(
     state: &AppState,
     input: &LocalSameFormatProviderDecisionInput,
     trace_id: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     skip_reason: &'static str,
 ) {
-    if let Err(err) = state
+    persist_skipped_local_candidate(
-        .upsert_request_candidate(UpsertRequestCandidateRecord {
+        state,
-            id: candidate_id.to_string(),
+        trace_id,
-            request_id: trace_id.to_string(),
+        &input.auth_context.user_id,
-            user_id: Some(input.auth_context.user_id.clone()),
+        &input.auth_context.api_key_id,
-            api_key_id: Some(input.auth_context.api_key_id.clone()),
+        candidate,
-            username: None,
-            api_key_name: None,
         candidate_index,
-            retry_index: 0,
+        candidate_id,
-            provider_id: Some(candidate.provider_id.clone()),
-            endpoint_id: Some(candidate.endpoint_id.clone()),
-            key_id: Some(candidate.key_id.clone()),
-            status: RequestCandidateStatus::Skipped,
-            skip_reason: Some(skip_reason.to_string()),
-            is_cached: Some(false),
-            status_code: None,
-            error_type: None,
-            error_message: None,
-            latency_ms: None,
-            concurrent_requests: None,
-            extra_data: None,
-            required_capabilities: candidate.key_capabilities.clone(),
-            created_at_unix_secs: None,
-            started_at_unix_secs: None,
-            finished_at_unix_secs: Some(current_unix_secs()),
-        })
-        .await
-    {
-        warn!(
-            trace_id = %trace_id,
-            candidate_id = %candidate_id,
         skip_reason,
-            error = ?err,
+        current_unix_secs(),
-            "gateway local same-format decision failed to persist skipped candidate"
+        "gateway local same-format decision failed to persist skipped candidate",
-        );
+    )
-    }
+    .await;
 }

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/types.rs

@@ -1,28 +1,20 @@
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+use crate::ai_pipeline::control_facade::GatewayControlAuthContext;
-pub(crate) enum LocalSameFormatProviderFamily {
+use crate::ai_pipeline::planner::auth_snapshot_facade::GatewayAuthApiKeySnapshot;
-    Standard,
-    Gemini,
-}
 
-#[derive(Debug, Clone, Copy)]
+pub(crate) use aether_ai_pipeline::planner::passthrough::provider::{
-pub(crate) struct LocalSameFormatProviderSpec {
+    LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
-    pub(crate) api_format: &'static str,
+};
-    pub(crate) decision_kind: &'static str,
-    pub(crate) report_kind: &'static str,
-    pub(crate) family: LocalSameFormatProviderFamily,
-    pub(crate) require_streaming: bool,
-}
 
 #[derive(Debug, Clone)]
 pub(crate) struct LocalSameFormatProviderDecisionInput {
-    pub(crate) auth_context: crate::control::GatewayControlAuthContext,
+    pub(crate) auth_context: GatewayControlAuthContext,
     pub(crate) requested_model: String,
-    pub(crate) auth_snapshot: crate::data::auth::GatewayAuthApiKeySnapshot,
+    pub(crate) auth_snapshot: GatewayAuthApiKeySnapshot,
 }
 
 #[derive(Debug, Clone)]
 pub(crate) struct LocalSameFormatProviderCandidateAttempt {
-    pub(crate) candidate: crate::scheduler::GatewayMinimalCandidateSelectionCandidate,
+    pub(crate) candidate: aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate,
     pub(crate) candidate_index: u32,
     pub(crate) candidate_id: String,
 }

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/plans.rs

@@ -1,5 +1,9 @@
 use tracing::warn;
 
+pub(crate) use aether_ai_pipeline::planner::passthrough::provider::{
+    resolve_stream_spec, resolve_sync_spec,
+};
+
 use super::{
     materialize_local_same_format_provider_candidate_attempts,
     maybe_build_local_same_format_provider_decision_payload_for_candidate,
@@ -7,84 +11,11 @@ use super::{
     GatewayError, LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::ai_pipeline::planner::common::{
-    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
-    CLAUDE_CLI_SYNC_PLAN_KIND, GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
-    GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
-};
 use crate::ai_pipeline::planner::plan_builders::{
     build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
     build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
 };
 
-pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalSameFormatProviderSpec> {
-    match plan_kind {
-        CLAUDE_CHAT_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "claude:chat",
-            decision_kind: CLAUDE_CHAT_SYNC_PLAN_KIND,
-            report_kind: "claude_chat_sync_success",
-            family: LocalSameFormatProviderFamily::Standard,
-            require_streaming: false,
-        }),
-        CLAUDE_CLI_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "claude:cli",
-            decision_kind: CLAUDE_CLI_SYNC_PLAN_KIND,
-            report_kind: "claude_cli_sync_success",
-            family: LocalSameFormatProviderFamily::Standard,
-            require_streaming: false,
-        }),
-        GEMINI_CHAT_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "gemini:chat",
-            decision_kind: GEMINI_CHAT_SYNC_PLAN_KIND,
-            report_kind: "gemini_chat_sync_success",
-            family: LocalSameFormatProviderFamily::Gemini,
-            require_streaming: false,
-        }),
-        GEMINI_CLI_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "gemini:cli",
-            decision_kind: GEMINI_CLI_SYNC_PLAN_KIND,
-            report_kind: "gemini_cli_sync_success",
-            family: LocalSameFormatProviderFamily::Gemini,
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalSameFormatProviderSpec> {
-    match plan_kind {
-        CLAUDE_CHAT_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "claude:chat",
-            decision_kind: CLAUDE_CHAT_STREAM_PLAN_KIND,
-            report_kind: "claude_chat_stream_success",
-            family: LocalSameFormatProviderFamily::Standard,
-            require_streaming: true,
-        }),
-        CLAUDE_CLI_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "claude:cli",
-            decision_kind: CLAUDE_CLI_STREAM_PLAN_KIND,
-            report_kind: "claude_cli_stream_success",
-            family: LocalSameFormatProviderFamily::Standard,
-            require_streaming: true,
-        }),
-        GEMINI_CHAT_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "gemini:chat",
-            decision_kind: GEMINI_CHAT_STREAM_PLAN_KIND,
-            report_kind: "gemini_chat_stream_success",
-            family: LocalSameFormatProviderFamily::Gemini,
-            require_streaming: true,
-        }),
-        GEMINI_CLI_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
-            api_format: "gemini:cli",
-            decision_kind: GEMINI_CLI_STREAM_PLAN_KIND,
-            report_kind: "gemini_cli_stream_success",
-            family: LocalSameFormatProviderFamily::Gemini,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}
-
 pub(crate) async fn build_local_sync_plan_and_reports(
     state: &AppState,
     parts: &http::request::Parts,

apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/request.rs

@@ -3,6 +3,8 @@ use std::collections::BTreeMap;
 use serde_json::Value;
 use url::form_urlencoded;
 
+use crate::ai_pipeline::planner::transport_facade::GatewayProviderTransportSnapshot;
+
 use super::{
     apply_local_body_rules, build_antigravity_v1internal_url, build_claude_code_messages_url,
     build_claude_messages_url, build_gemini_content_url,
@@ -18,7 +20,7 @@ pub(super) fn build_same_format_provider_request_body(
     spec: LocalSameFormatProviderSpec,
     body_rules: Option<&Value>,
     upstream_is_stream: bool,
-    kiro_auth: Option<&crate::provider_transport::kiro::KiroRequestAuth>,
+    kiro_auth: Option<&crate::ai_pipeline::provider_transport_facade::kiro::KiroRequestAuth>,
     is_claude_code: bool,
 ) -> Option<Value> {
     if let Some(kiro_auth) = kiro_auth {
@@ -60,11 +62,11 @@ pub(super) fn build_same_format_provider_request_body(
 
 pub(super) fn build_same_format_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     mapped_model: &str,
     spec: LocalSameFormatProviderSpec,
     upstream_is_stream: bool,
-    kiro_auth: Option<&crate::provider_transport::kiro::KiroRequestAuth>,
+    kiro_auth: Option<&crate::ai_pipeline::provider_transport_facade::kiro::KiroRequestAuth>,
 ) -> Option<String> {
     if let Some(kiro_auth) = kiro_auth {
         return build_kiro_generate_assistant_response_url(

apps/aether-gateway/src/ai_pipeline/planner/route.rs

@@ -0,0 +1,119 @@
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
+
+pub(crate) fn resolve_execution_runtime_stream_plan_kind(
+    parts: &http::request::Parts,
+    decision: &GatewayControlDecision,
+) -> Option<&'static str> {
+    aether_ai_pipeline::planner::route::resolve_execution_runtime_stream_plan_kind(
+        decision.route_class.as_deref(),
+        decision.route_family.as_deref(),
+        decision.route_kind.as_deref(),
+        &parts.method,
+        parts.uri.path(),
+    )
+}
+
+pub(crate) fn resolve_execution_runtime_sync_plan_kind(
+    parts: &http::request::Parts,
+    decision: &GatewayControlDecision,
+) -> Option<&'static str> {
+    aether_ai_pipeline::planner::route::resolve_execution_runtime_sync_plan_kind(
+        decision.route_class.as_deref(),
+        decision.route_family.as_deref(),
+        decision.route_kind.as_deref(),
+        &parts.method,
+        parts.uri.path(),
+    )
+}
+
+pub(crate) fn is_matching_stream_request(
+    plan_kind: &str,
+    parts: &http::request::Parts,
+    body_json: &serde_json::Value,
+) -> bool {
+    aether_ai_pipeline::planner::route::is_matching_stream_request(
+        plan_kind,
+        parts.uri.path(),
+        body_json,
+    )
+}
+
+pub(crate) fn supports_sync_scheduler_decision_kind(plan_kind: &str) -> bool {
+    aether_ai_pipeline::planner::route::supports_sync_scheduler_decision_kind(plan_kind)
+}
+
+pub(crate) fn supports_stream_scheduler_decision_kind(plan_kind: &str) -> bool {
+    aether_ai_pipeline::planner::route::supports_stream_scheduler_decision_kind(plan_kind)
+}
+
+#[cfg(test)]
+mod tests {
+    use axum::http::{Method, Request};
+
+    use super::{
+        is_matching_stream_request, resolve_execution_runtime_stream_plan_kind,
+        resolve_execution_runtime_sync_plan_kind, supports_stream_scheduler_decision_kind,
+        supports_sync_scheduler_decision_kind,
+    };
+    use crate::ai_pipeline::control_facade::GatewayControlDecision;
+
+    fn sample_decision(route_family: &str, route_kind: &str) -> GatewayControlDecision {
+        GatewayControlDecision {
+            public_path: "/".to_string(),
+            public_query_string: None,
+            route_class: Some("ai_public".to_string()),
+            route_family: Some(route_family.to_string()),
+            route_kind: Some(route_kind.to_string()),
+            auth_context: None,
+            admin_principal: None,
+            auth_endpoint_signature: None,
+            execution_runtime_candidate: true,
+            local_auth_rejection: None,
+        }
+    }
+
+    #[test]
+    fn resolves_openai_chat_plan_kinds_via_pipeline_crate() {
+        let request = Request::builder()
+            .method(Method::POST)
+            .uri("/v1/chat/completions")
+            .body(())
+            .expect("request should build");
+        let (parts, _) = request.into_parts();
+        let decision = sample_decision("openai", "chat");
+
+        assert_eq!(
+            resolve_execution_runtime_sync_plan_kind(&parts, &decision),
+            Some("openai_chat_sync")
+        );
+        assert_eq!(
+            resolve_execution_runtime_stream_plan_kind(&parts, &decision),
+            Some("openai_chat_stream")
+        );
+    }
+
+    #[test]
+    fn stream_matching_uses_pipeline_route_logic() {
+        let request = Request::builder()
+            .method(Method::POST)
+            .uri("/v1/chat/completions")
+            .body(())
+            .expect("request should build");
+        let (parts, _) = request.into_parts();
+
+        assert!(!is_matching_stream_request(
+            "openai_chat_stream",
+            &parts,
+            &serde_json::json!({"stream": false}),
+        ));
+        assert!(is_matching_stream_request(
+            "openai_chat_stream",
+            &parts,
+            &serde_json::json!({"stream": true}),
+        ));
+        assert!(supports_sync_scheduler_decision_kind("openai_chat_sync"));
+        assert!(supports_stream_scheduler_decision_kind(
+            "openai_chat_stream"
+        ));
+    }
+}

apps/aether-gateway/src/ai_pipeline/planner/scheduler_facade.rs

@@ -0,0 +1,44 @@
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
+
+use crate::ai_pipeline::planner::auth_snapshot_facade::GatewayAuthApiKeySnapshot;
+use crate::{AppState, GatewayError};
+
+pub(crate) async fn list_selectable_candidates(
+    state: &AppState,
+    api_format: &str,
+    global_model_name: &str,
+    require_streaming: bool,
+    auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
+    now_unix_secs: u64,
+) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, GatewayError> {
+    crate::scheduler::candidate::list_selectable_candidates(
+        state.data.as_ref(),
+        state,
+        api_format,
+        global_model_name,
+        require_streaming,
+        auth_snapshot,
+        now_unix_secs,
+    )
+    .await
+}
+
+pub(crate) async fn list_selectable_candidates_for_required_capability_without_requested_model(
+    state: &AppState,
+    candidate_api_format: &str,
+    required_capability: &str,
+    require_streaming: bool,
+    auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
+    now_unix_secs: u64,
+) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, GatewayError> {
+    crate::scheduler::candidate::list_selectable_candidates_for_required_capability_without_requested_model(
+        state.data.as_ref(),
+        state,
+        candidate_api_format,
+        required_capability,
+        require_streaming,
+        auth_snapshot,
+        now_unix_secs,
+    )
+    .await
+}

apps/aether-gateway/src/ai_pipeline/planner/specialized/files.rs

@@ -1,57 +1,60 @@
 use std::collections::BTreeMap;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 use serde_json::json;
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::ai_pipeline::control_facade::{
+    collect_control_headers, GatewayControlAuthContext, GatewayControlDecision,
+};
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::auth_snapshot_facade::{
+    read_auth_api_key_snapshot, GatewayAuthApiKeySnapshot,
+};
 use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::planner::candidate_runtime_facade::{
+    persist_available_local_candidate, persist_skipped_local_candidate,
+};
 use crate::ai_pipeline::planner::common::{
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
-    GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND,
-    GEMINI_FILES_LIST_PLAN_KIND, GEMINI_FILES_UPLOAD_PLAN_KIND,
 };
+use crate::ai_pipeline::planner::executor_facade::mark_unused_local_candidate_items;
 use crate::ai_pipeline::planner::plan_builders::{
     build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::planner::scheduler_facade::list_selectable_candidates_for_required_capability_without_requested_model;
-use crate::headers::collect_control_headers;
+use crate::ai_pipeline::planner::transport_facade::read_provider_transport_snapshot;
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_passthrough_headers_with_auth, resolve_local_gemini_auth,
 };
-use crate::provider_transport::policy::supports_local_gemini_transport_with_network;
+use crate::ai_pipeline::provider_transport_facade::policy::supports_local_gemini_transport_with_network;
-use crate::provider_transport::url::build_gemini_files_passthrough_url;
+use crate::ai_pipeline::provider_transport_facade::url::build_gemini_files_passthrough_url;
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     apply_local_body_rules, apply_local_header_rules, resolve_transport_execution_timeouts,
     resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
 };
-use crate::scheduler::{
+use crate::clock::current_unix_secs;
-    current_unix_secs, list_selectable_candidates_for_required_capability_without_requested_model,
-    record_local_request_candidate_status, GatewayMinimalCandidateSelectionCandidate,
-};
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
+use aether_ai_pipeline::contracts::GEMINI_FILES_UPLOAD_PLAN_KIND;
+use aether_ai_pipeline::planner::specialized::files::{
+    resolve_stream_spec, resolve_sync_spec, LocalGeminiFilesSpec,
+};
 
 const GEMINI_FILES_CANDIDATE_API_FORMAT: &str = "gemini:chat";
 const GEMINI_FILES_CLIENT_API_FORMAT: &str = "gemini:files";
 const GEMINI_FILES_REQUIRED_CAPABILITY: &str = "gemini_files";
 
-#[derive(Debug, Clone, Copy)]
-struct LocalGeminiFilesSpec {
-    decision_kind: &'static str,
-    report_kind: Option<&'static str>,
-    require_streaming: bool,
-}
-
 #[derive(Debug, Clone)]
 struct LocalGeminiFilesDecisionInput {
-    auth_context: crate::control::GatewayControlAuthContext,
+    auth_context: GatewayControlAuthContext,
-    auth_snapshot: crate::data::auth::GatewayAuthApiKeySnapshot,
+    auth_snapshot: GatewayAuthApiKeySnapshot,
 }
 
 #[derive(Debug, Clone)]
 struct LocalGeminiFilesCandidateAttempt {
-    candidate: GatewayMinimalCandidateSelectionCandidate,
+    candidate: SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: String,
 }
@@ -181,43 +184,6 @@ pub(crate) async fn maybe_build_stream_local_gemini_files_decision_payload(
     Ok(None)
 }
 
-fn resolve_sync_spec(plan_kind: &str) -> Option<LocalGeminiFilesSpec> {
-    match plan_kind {
-        GEMINI_FILES_UPLOAD_PLAN_KIND => Some(LocalGeminiFilesSpec {
-            decision_kind: GEMINI_FILES_UPLOAD_PLAN_KIND,
-            report_kind: Some("gemini_files_store_mapping"),
-            require_streaming: false,
-        }),
-        GEMINI_FILES_LIST_PLAN_KIND => Some(LocalGeminiFilesSpec {
-            decision_kind: GEMINI_FILES_LIST_PLAN_KIND,
-            report_kind: Some("gemini_files_store_mapping"),
-            require_streaming: false,
-        }),
-        GEMINI_FILES_GET_PLAN_KIND => Some(LocalGeminiFilesSpec {
-            decision_kind: GEMINI_FILES_GET_PLAN_KIND,
-            report_kind: Some("gemini_files_store_mapping"),
-            require_streaming: false,
-        }),
-        GEMINI_FILES_DELETE_PLAN_KIND => Some(LocalGeminiFilesSpec {
-            decision_kind: GEMINI_FILES_DELETE_PLAN_KIND,
-            report_kind: Some("gemini_files_delete_mapping"),
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-fn resolve_stream_spec(plan_kind: &str) -> Option<LocalGeminiFilesSpec> {
-    match plan_kind {
-        GEMINI_FILES_DOWNLOAD_PLAN_KIND => Some(LocalGeminiFilesSpec {
-            decision_kind: GEMINI_FILES_DOWNLOAD_PLAN_KIND,
-            report_kind: None,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}
-
 async fn build_local_sync_plan_and_reports(
     state: &AppState,
     parts: &http::request::Parts,
@@ -333,8 +299,8 @@ async fn resolve_local_gemini_files_decision_input(
         return None;
     };
 
-    let auth_snapshot = match state
+    let auth_snapshot = match read_auth_api_key_snapshot(
-        .read_auth_api_key_snapshot(
+        state,
         &auth_context.user_id,
         &auth_context.api_key_id,
         current_unix_secs(),
@@ -392,46 +358,19 @@ async fn materialize_local_gemini_files_candidate_attempts(
             "key_name": candidate.key_name.clone(),
         });
 
-        let candidate_id = match state
+        let candidate_id = persist_available_local_candidate(
-            .upsert_request_candidate(UpsertRequestCandidateRecord {
+            state,
-                id: generated_candidate_id.clone(),
+            trace_id,
-                request_id: trace_id.to_string(),
+            &input.auth_context.user_id,
-                user_id: Some(input.auth_context.user_id.clone()),
+            &input.auth_context.api_key_id,
-                api_key_id: Some(input.auth_context.api_key_id.clone()),
+            &candidate,
-                username: None,
+            candidate_index as u32,
-                api_key_name: None,
+            &generated_candidate_id,
-                candidate_index: candidate_index as u32,
+            Some(extra_data),
-                retry_index: 0,
+            created_at_unix_secs,
-                provider_id: Some(candidate.provider_id.clone()),
+            "gateway local gemini files request candidate upsert failed",
-                endpoint_id: Some(candidate.endpoint_id.clone()),
+        )
-                key_id: Some(candidate.key_id.clone()),
+        .await;
-                status: RequestCandidateStatus::Available,
-                skip_reason: None,
-                is_cached: Some(false),
-                status_code: None,
-                error_type: None,
-                error_message: None,
-                latency_ms: None,
-                concurrent_requests: None,
-                extra_data: Some(extra_data),
-                required_capabilities: candidate.key_capabilities.clone(),
-                created_at_unix_secs: Some(created_at_unix_secs),
-                started_at_unix_secs: None,
-                finished_at_unix_secs: None,
-            })
-            .await
-        {
-            Ok(Some(stored)) => stored.id,
-            Ok(None) => generated_candidate_id.clone(),
-            Err(err) => {
-                warn!(
-                    trace_id = %trace_id,
-                    error = ?err,
-                    "gateway local gemini files request candidate upsert failed"
-                );
-                generated_candidate_id.clone()
-            }
-        };
 
         attempts.push(LocalGeminiFilesCandidateAttempt {
             candidate,
@@ -460,8 +399,8 @@ async fn maybe_build_local_gemini_files_decision_payload_for_candidate(
         candidate_id,
     } = attempt;
 
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -664,16 +603,8 @@ async fn maybe_build_local_gemini_files_decision_payload_for_candidate(
             EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
         },
         decision_kind: Some(spec.decision_kind.to_string()),
-        execution_strategy: Some(
+        execution_strategy: Some(ExecutionStrategy::LocalSameFormat.as_str().to_string()),
-            crate::execution_runtime::ExecutionStrategy::LocalSameFormat
+        conversion_mode: Some(ConversionMode::None.as_str().to_string()),
-                .as_str()
-                .to_string(),
-        ),
-        conversion_mode: Some(
-            crate::execution_runtime::ConversionMode::None
-                .as_str()
-                .to_string(),
-        ),
         request_id: Some(trace_id.to_string()),
         candidate_id: Some(candidate_id.clone()),
         provider_name: Some(transport.provider.name.clone()),
@@ -737,70 +668,38 @@ async fn mark_skipped_local_gemini_files_candidate(
     state: &AppState,
     input: &LocalGeminiFilesDecisionInput,
     trace_id: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     skip_reason: &'static str,
 ) {
-    if let Err(err) = state
+    persist_skipped_local_candidate(
-        .upsert_request_candidate(UpsertRequestCandidateRecord {
+        state,
-            id: candidate_id.to_string(),
+        trace_id,
-            request_id: trace_id.to_string(),
+        &input.auth_context.user_id,
-            user_id: Some(input.auth_context.user_id.clone()),
+        &input.auth_context.api_key_id,
-            api_key_id: Some(input.auth_context.api_key_id.clone()),
+        candidate,
-            username: None,
-            api_key_name: None,
         candidate_index,
-            retry_index: 0,
+        candidate_id,
-            provider_id: Some(candidate.provider_id.clone()),
-            endpoint_id: Some(candidate.endpoint_id.clone()),
-            key_id: Some(candidate.key_id.clone()),
-            status: RequestCandidateStatus::Skipped,
-            skip_reason: Some(skip_reason.to_string()),
-            is_cached: Some(false),
-            status_code: None,
-            error_type: None,
-            error_message: None,
-            latency_ms: None,
-            concurrent_requests: None,
-            extra_data: None,
-            required_capabilities: candidate.key_capabilities.clone(),
-            created_at_unix_secs: None,
-            started_at_unix_secs: None,
-            finished_at_unix_secs: Some(current_unix_secs()),
-        })
-        .await
-    {
-        warn!(
-            trace_id = %trace_id,
-            candidate_id = %candidate_id,
         skip_reason,
-            error = ?err,
+        current_unix_secs(),
-            "gateway local gemini files failed to persist skipped candidate"
+        "gateway local gemini files failed to persist skipped candidate",
-        );
+    )
-    }
+    .await;
 }
 
 async fn mark_unused_local_files_candidates<T>(state: &AppState, remaining: Vec<T>)
 where
     T: LocalGeminiFilesPlanAndReport,
 {
-    for plan_and_report in remaining {
+    mark_unused_local_candidate_items(
-        record_local_request_candidate_status(
         state,
-            plan_and_report.plan(),
+        remaining,
-            plan_and_report.report_context(),
+        |item| item.plan(),
-            RequestCandidateStatus::Unused,
+        |item| item.report_context(),
-            None,
-            None,
-            None,
-            None,
-            None,
-            None,
     )
     .await;
 }
-}
 
 trait LocalGeminiFilesPlanAndReport {
     fn plan(&self) -> &aether_contracts::ExecutionPlan;

apps/aether-gateway/src/ai_pipeline/planner/specialized/video.rs

@@ -1,63 +1,59 @@
 use std::collections::BTreeMap;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 use serde_json::{json, Value};
 use tracing::warn;
 use uuid::Uuid;
 
-use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::control_facade::{
-use crate::ai_pipeline::planner::common::{
+    collect_control_headers, GatewayControlAuthContext, GatewayControlDecision,
-    EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND,
-    OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
 };
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::auth_snapshot_facade::{
+    read_auth_api_key_snapshot, GatewayAuthApiKeySnapshot,
+};
+use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::planner::candidate_runtime_facade::{
+    persist_available_local_candidate, persist_skipped_local_candidate,
+};
+use crate::ai_pipeline::planner::common::EXECUTION_RUNTIME_SYNC_DECISION_ACTION;
+use crate::ai_pipeline::planner::executor_facade::mark_unused_local_candidate_items;
 use crate::ai_pipeline::planner::plan_builders::{
     build_passthrough_sync_plan_from_decision, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::planner::scheduler_facade::list_selectable_candidates;
-use crate::headers::collect_control_headers;
+use crate::ai_pipeline::planner::transport_facade::{
-use crate::provider_transport::auth::{
+    read_provider_transport_snapshot, GatewayProviderTransportSnapshot,
+};
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_passthrough_headers_with_auth, resolve_local_gemini_auth, resolve_local_openai_chat_auth,
 };
-use crate::provider_transport::policy::{
+use crate::ai_pipeline::provider_transport_facade::policy::{
     supports_local_gemini_transport_with_network, supports_local_standard_transport_with_network,
 };
-use crate::provider_transport::url::{
+use crate::ai_pipeline::provider_transport_facade::url::{
     build_gemini_video_predict_long_running_url, build_passthrough_path_url,
 };
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     apply_local_body_rules, apply_local_header_rules, resolve_transport_execution_timeouts,
     resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
 };
-use crate::scheduler::{
+use crate::clock::current_unix_secs;
-    current_unix_secs, list_selectable_candidates, record_local_request_candidate_status,
-    GatewayMinimalCandidateSelectionCandidate,
-};
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
-
+use aether_ai_pipeline::planner::specialized::video::{
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+    resolve_sync_spec, LocalVideoCreateFamily, LocalVideoCreateSpec,
-enum LocalVideoCreateFamily {
+};
-    OpenAi,
-    Gemini,
-}
-
-#[derive(Debug, Clone, Copy)]
-struct LocalVideoCreateSpec {
-    api_format: &'static str,
-    decision_kind: &'static str,
-    report_kind: &'static str,
-    family: LocalVideoCreateFamily,
-}
 
 #[derive(Debug, Clone)]
 struct LocalVideoCreateDecisionInput {
-    auth_context: crate::control::GatewayControlAuthContext,
+    auth_context: GatewayControlAuthContext,
     requested_model: String,
-    auth_snapshot: crate::data::auth::GatewayAuthApiKeySnapshot,
+    auth_snapshot: GatewayAuthApiKeySnapshot,
 }
 
 #[derive(Debug, Clone)]
 struct LocalVideoCreateCandidateAttempt {
-    candidate: GatewayMinimalCandidateSelectionCandidate,
+    candidate: SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: String,
 }
@@ -141,24 +137,6 @@ pub(crate) async fn maybe_build_sync_local_video_decision_payload(
     Ok(None)
 }
 
-fn resolve_sync_spec(plan_kind: &str) -> Option<LocalVideoCreateSpec> {
-    match plan_kind {
-        OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND => Some(LocalVideoCreateSpec {
-            api_format: "openai:video",
-            decision_kind: OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
-            report_kind: "openai_video_create_sync_finalize",
-            family: LocalVideoCreateFamily::OpenAi,
-        }),
-        GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND => Some(LocalVideoCreateSpec {
-            api_format: "gemini:video",
-            decision_kind: GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND,
-            report_kind: "gemini_video_create_sync_finalize",
-            family: LocalVideoCreateFamily::Gemini,
-        }),
-        _ => None,
-    }
-}
-
 async fn build_local_sync_plan_and_reports(
     state: &AppState,
     parts: &http::request::Parts,
@@ -257,8 +235,8 @@ async fn resolve_local_video_create_decision_input(
         LocalVideoCreateFamily::Gemini => extract_gemini_video_model_from_path(parts.uri.path())?,
     };
 
-    let auth_snapshot = match state
+    let auth_snapshot = match read_auth_api_key_snapshot(
-        .read_auth_api_key_snapshot(
+        state,
         &auth_context.user_id,
         &auth_context.api_key_id,
         current_unix_secs(),
@@ -299,8 +277,8 @@ async fn maybe_build_local_video_create_decision_payload_for_candidate(
         candidate_index,
         candidate_id,
     } = attempt;
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -461,16 +439,8 @@ async fn maybe_build_local_video_create_decision_payload_for_candidate(
     Some(GatewayControlSyncDecisionResponse {
         action: EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string(),
         decision_kind: Some(spec.decision_kind.to_string()),
-        execution_strategy: Some(
+        execution_strategy: Some(ExecutionStrategy::LocalSameFormat.as_str().to_string()),
-            crate::execution_runtime::ExecutionStrategy::LocalSameFormat
+        conversion_mode: Some(ConversionMode::None.as_str().to_string()),
-                .as_str()
-                .to_string(),
-        ),
-        conversion_mode: Some(
-            crate::execution_runtime::ConversionMode::None
-                .as_str()
-                .to_string(),
-        ),
         request_id: Some(trace_id.to_string()),
         candidate_id: Some(candidate_id.clone()),
         provider_name: Some(transport.provider.name.clone()),
@@ -552,7 +522,7 @@ fn build_provider_request_body(
 
 fn build_video_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     mapped_model: &str,
     family: LocalVideoCreateFamily,
 ) -> Option<String> {
@@ -595,7 +565,7 @@ async fn materialize_local_video_create_candidate_attempts(
     state: &AppState,
     trace_id: &str,
     input: &LocalVideoCreateDecisionInput,
-    candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
+    candidates: Vec<SchedulerMinimalCandidateSelectionCandidate>,
     api_format: &str,
 ) -> Vec<LocalVideoCreateCandidateAttempt> {
     let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
@@ -616,47 +586,19 @@ async fn materialize_local_video_create_candidate_attempts(
             "key_name": candidate.key_name.clone(),
         });
 
-        let candidate_id = match state
+        let candidate_id = persist_available_local_candidate(
-            .upsert_request_candidate(UpsertRequestCandidateRecord {
+            state,
-                id: generated_candidate_id.clone(),
+            trace_id,
-                request_id: trace_id.to_string(),
+            &input.auth_context.user_id,
-                user_id: Some(input.auth_context.user_id.clone()),
+            &input.auth_context.api_key_id,
-                api_key_id: Some(input.auth_context.api_key_id.clone()),
+            &candidate,
-                username: None,
+            candidate_index as u32,
-                api_key_name: None,
+            &generated_candidate_id,
-                candidate_index: candidate_index as u32,
+            Some(extra_data),
-                retry_index: 0,
+            created_at_unix_secs,
-                provider_id: Some(candidate.provider_id.clone()),
+            "gateway local video decision request candidate upsert failed",
-                endpoint_id: Some(candidate.endpoint_id.clone()),
+        )
-                key_id: Some(candidate.key_id.clone()),
+        .await;
-                status: RequestCandidateStatus::Available,
-                skip_reason: None,
-                is_cached: Some(false),
-                status_code: None,
-                error_type: None,
-                error_message: None,
-                latency_ms: None,
-                concurrent_requests: None,
-                extra_data: Some(extra_data),
-                required_capabilities: candidate.key_capabilities.clone(),
-                created_at_unix_secs: Some(created_at_unix_secs),
-                started_at_unix_secs: None,
-                finished_at_unix_secs: None,
-            })
-            .await
-        {
-            Ok(Some(stored)) => stored.id,
-            Ok(None) => generated_candidate_id.clone(),
-            Err(err) => {
-                warn!(
-                    trace_id = %trace_id,
-                    decision_api_format = api_format,
-                    error = ?err,
-                    "gateway local video decision request candidate upsert failed"
-                );
-                generated_candidate_id.clone()
-            }
-        };
 
         attempts.push(LocalVideoCreateCandidateAttempt {
             candidate,
@@ -672,71 +614,38 @@ async fn mark_skipped_local_video_candidate(
     state: &AppState,
     input: &LocalVideoCreateDecisionInput,
     trace_id: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     skip_reason: &'static str,
 ) {
-    let terminal_unix_secs = current_unix_secs();
+    persist_skipped_local_candidate(
-    if let Err(err) = state
+        state,
-        .upsert_request_candidate(UpsertRequestCandidateRecord {
+        trace_id,
-            id: candidate_id.to_string(),
+        &input.auth_context.user_id,
-            request_id: trace_id.to_string(),
+        &input.auth_context.api_key_id,
-            user_id: Some(input.auth_context.user_id.clone()),
+        candidate,
-            api_key_id: Some(input.auth_context.api_key_id.clone()),
-            username: None,
-            api_key_name: None,
         candidate_index,
-            retry_index: 0,
+        candidate_id,
-            provider_id: Some(candidate.provider_id.clone()),
-            endpoint_id: Some(candidate.endpoint_id.clone()),
-            key_id: Some(candidate.key_id.clone()),
-            status: RequestCandidateStatus::Skipped,
-            skip_reason: Some(skip_reason.to_string()),
-            is_cached: Some(false),
-            status_code: None,
-            error_type: None,
-            error_message: None,
-            latency_ms: None,
-            concurrent_requests: None,
-            extra_data: None,
-            required_capabilities: candidate.key_capabilities.clone(),
-            created_at_unix_secs: None,
-            started_at_unix_secs: None,
-            finished_at_unix_secs: Some(terminal_unix_secs),
-        })
-        .await
-    {
-        warn!(
-            trace_id = %trace_id,
-            candidate_id = %candidate_id,
         skip_reason,
-            error = ?err,
+        current_unix_secs(),
-            "gateway local video decision failed to persist skipped candidate"
+        "gateway local video decision failed to persist skipped candidate",
-        );
+    )
-    }
+    .await;
 }
 
 async fn mark_unused_local_video_candidates(
     state: &AppState,
     remaining: Vec<LocalSyncPlanAndReport>,
 ) {
-    for plan_and_report in remaining {
+    mark_unused_local_candidate_items(
-        record_local_request_candidate_status(
         state,
-            &plan_and_report.plan,
+        remaining,
-            plan_and_report.report_context.as_ref(),
+        |item| &item.plan,
-            RequestCandidateStatus::Unused,
+        |item| item.report_context.as_ref(),
-            None,
-            None,
-            None,
-            None,
-            None,
-            None,
     )
     .await;
 }
-}
 
 fn extract_gemini_video_model_from_path(path: &str) -> Option<String> {
     let suffix = path.strip_prefix("/v1beta/models/")?;

apps/aether-gateway/src/ai_pipeline/planner/standard/claude/chat.rs

@@ -1,33 +0,0 @@
-use crate::ai_pipeline::planner::common::{
-    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND,
-};
-
-use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
-
-pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        CLAUDE_CHAT_SYNC_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "claude:chat",
-            decision_kind: CLAUDE_CHAT_SYNC_PLAN_KIND,
-            report_kind: "claude_chat_sync_finalize",
-            family: LocalStandardSourceFamily::Standard,
-            mode: LocalStandardSourceMode::Chat,
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        CLAUDE_CHAT_STREAM_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "claude:chat",
-            decision_kind: CLAUDE_CHAT_STREAM_PLAN_KIND,
-            report_kind: "claude_chat_stream_success",
-            family: LocalStandardSourceFamily::Standard,
-            mode: LocalStandardSourceMode::Chat,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}

apps/aether-gateway/src/ai_pipeline/planner/standard/claude/cli.rs

@@ -1,33 +0,0 @@
-use crate::ai_pipeline::planner::common::{
-    CLAUDE_CLI_STREAM_PLAN_KIND, CLAUDE_CLI_SYNC_PLAN_KIND,
-};
-
-use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
-
-pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        CLAUDE_CLI_SYNC_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "claude:cli",
-            decision_kind: CLAUDE_CLI_SYNC_PLAN_KIND,
-            report_kind: "claude_cli_sync_finalize",
-            family: LocalStandardSourceFamily::Standard,
-            mode: LocalStandardSourceMode::Cli,
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        CLAUDE_CLI_STREAM_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "claude:cli",
-            decision_kind: CLAUDE_CLI_STREAM_PLAN_KIND,
-            report_kind: "claude_cli_stream_success",
-            family: LocalStandardSourceFamily::Standard,
-            mode: LocalStandardSourceMode::Cli,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}

apps/aether-gateway/src/ai_pipeline/planner/standard/claude/mod.rs

@@ -1,13 +1,23 @@
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
+use aether_ai_pipeline::planner::standard::claude::{
+    resolve_stream_spec as resolve_pipeline_stream_spec,
+    resolve_sync_spec as resolve_pipeline_sync_spec,
+};
+
 use super::family::{
     maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
 };
 pub(crate) use crate::ai_pipeline::conversion::request::normalize_claude_request_to_openai_chat_request;
 
-pub(crate) mod chat;
+pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<super::family::LocalStandardSpec> {
-pub(crate) mod cli;
+    resolve_pipeline_sync_spec(plan_kind)
+}
+
+pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<super::family::LocalStandardSpec> {
+    resolve_pipeline_stream_spec(plan_kind)
+}
 
 pub(crate) async fn maybe_build_sync_local_claude_decision_payload(
     state: &AppState,
@@ -24,9 +34,7 @@ pub(crate) async fn maybe_build_sync_local_claude_decision_payload(
         decision,
         body_json,
         plan_kind,
-        |plan_kind| {
+        resolve_sync_spec,
-            chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
-        },
     )
     .await
 }
@@ -46,9 +54,7 @@ pub(crate) async fn maybe_build_stream_local_claude_decision_payload(
         decision,
         body_json,
         plan_kind,
-        |plan_kind| {
+        resolve_stream_spec,
-            chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
-        },
     )
     .await
 }

apps/aether-gateway/src/ai_pipeline/planner/standard/codex.rs

@@ -0,0 +1,433 @@
+use std::collections::BTreeMap;
+use std::fmt::Write;
+
+use crate::ai_pipeline::provider_transport_facade::body_rules_handle_path;
+use serde_json::{json, Value};
+use sha1::{Digest as Sha1Digest, Sha1};
+use sha2::{Digest as Sha2Digest, Sha256};
+use uuid::Uuid;
+
+const CODEX_PROMPT_CACHE_NAMESPACE_VERSION: &str = "v3";
+const UUID_NAMESPACE_OID_BYTES: [u8; 16] = [
+    0x6b, 0xa7, 0xb8, 0x12, 0x9d, 0xad, 0x11, 0xd1, 0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30, 0xc8,
+];
+
+fn is_codex_openai_cli_request(provider_type: &str, provider_api_format: &str) -> bool {
+    provider_type.trim().eq_ignore_ascii_case("codex")
+        && matches!(
+            provider_api_format.trim().to_ascii_lowercase().as_str(),
+            "openai:cli" | "openai:compact"
+        )
+}
+
+fn build_stable_codex_prompt_cache_key(user_api_key_id: &str) -> Option<String> {
+    let normalized = user_api_key_id.trim();
+    if normalized.is_empty() {
+        return None;
+    }
+
+    let namespace = format!(
+        "aether:codex:prompt-cache:{CODEX_PROMPT_CACHE_NAMESPACE_VERSION}:user:{normalized}"
+    );
+    let mut hasher = Sha1::new();
+    hasher.update(UUID_NAMESPACE_OID_BYTES);
+    hasher.update(namespace.as_bytes());
+
+    let digest = hasher.finalize();
+    let mut bytes = [0u8; 16];
+    bytes.copy_from_slice(&digest[..16]);
+    bytes[6] = (bytes[6] & 0x0f) | 0x50;
+    bytes[8] = (bytes[8] & 0x3f) | 0x80;
+    Some(Uuid::from_bytes(bytes).to_string())
+}
+
+fn maybe_inject_codex_prompt_cache_key(
+    provider_request_body: &mut Value,
+    provider_type: &str,
+    provider_api_format: &str,
+    user_api_key_id: Option<&str>,
+) {
+    if !is_codex_openai_cli_request(provider_type, provider_api_format) {
+        return;
+    }
+
+    let Some(body_object) = provider_request_body.as_object_mut() else {
+        return;
+    };
+
+    let existing = body_object
+        .get("prompt_cache_key")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .unwrap_or_default();
+    if !existing.is_empty() {
+        return;
+    }
+
+    let Some(prompt_cache_key) = user_api_key_id.and_then(build_stable_codex_prompt_cache_key)
+    else {
+        return;
+    };
+
+    body_object.insert(
+        "prompt_cache_key".to_string(),
+        Value::String(prompt_cache_key),
+    );
+}
+
+fn build_short_codex_header_id(seed: &str) -> Option<String> {
+    let normalized = seed.trim();
+    if normalized.is_empty() {
+        return None;
+    }
+
+    let digest = Sha256::digest(normalized.as_bytes());
+    let mut short_id = String::with_capacity(16);
+    for byte in digest.iter().take(8) {
+        let _ = write!(&mut short_id, "{byte:02x}");
+    }
+    Some(short_id)
+}
+
+fn header_map_has_non_empty_value(headers: &http::HeaderMap, header_name: &str) -> bool {
+    let target = header_name.trim().to_ascii_lowercase();
+    if target.is_empty() {
+        return false;
+    }
+
+    headers.iter().any(|(name, value)| {
+        if name.as_str().trim().to_ascii_lowercase() != target {
+            return false;
+        }
+        value
+            .to_str()
+            .ok()
+            .map(str::trim)
+            .map(|value| !value.is_empty())
+            .unwrap_or(false)
+    })
+}
+
+fn extract_codex_account_id(decrypted_auth_config_raw: Option<&str>) -> Option<String> {
+    let raw = decrypted_auth_config_raw?.trim();
+    if raw.is_empty() {
+        return None;
+    }
+
+    serde_json::from_str::<Value>(raw).ok().and_then(|value| {
+        value
+            .get("account_id")
+            .and_then(Value::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+            .map(ToOwned::to_owned)
+    })
+}
+
+pub(crate) fn apply_codex_openai_cli_special_headers(
+    provider_request_headers: &mut BTreeMap<String, String>,
+    provider_request_body: &Value,
+    original_headers: &http::HeaderMap,
+    provider_type: &str,
+    provider_api_format: &str,
+    request_id: Option<&str>,
+    decrypted_auth_config_raw: Option<&str>,
+) {
+    if !is_codex_openai_cli_request(provider_type, provider_api_format) {
+        return;
+    }
+
+    if let Some(account_id) = extract_codex_account_id(decrypted_auth_config_raw) {
+        provider_request_headers.insert("chatgpt-account-id".to_string(), account_id);
+    }
+    if !provider_request_headers
+        .get("x-client-request-id")
+        .map(|value| !value.trim().is_empty())
+        .unwrap_or(false)
+    {
+        if let Some(request_id) = request_id.map(str::trim).filter(|value| !value.is_empty()) {
+            provider_request_headers
+                .insert("x-client-request-id".to_string(), request_id.to_string());
+        }
+    }
+
+    let prompt_cache_key = provider_request_body
+        .get("prompt_cache_key")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty());
+    let Some(short_id) = prompt_cache_key.and_then(build_short_codex_header_id) else {
+        return;
+    };
+
+    if !header_map_has_non_empty_value(original_headers, "session_id") {
+        provider_request_headers.insert("session_id".to_string(), short_id.clone());
+    }
+
+    if provider_api_format.trim().to_ascii_lowercase() != "openai:compact"
+        && !header_map_has_non_empty_value(original_headers, "conversation_id")
+    {
+        provider_request_headers.insert("conversation_id".to_string(), short_id);
+    }
+}
+
+pub(crate) fn apply_codex_openai_cli_special_body_edits(
+    provider_request_body: &mut Value,
+    provider_type: &str,
+    provider_api_format: &str,
+    body_rules: Option<&Value>,
+    user_api_key_id: Option<&str>,
+) {
+    if !is_codex_openai_cli_request(provider_type, provider_api_format) {
+        return;
+    }
+
+    let Some(body_object) = provider_request_body.as_object_mut() else {
+        return;
+    };
+
+    if !body_rules_handle_path(body_rules, "max_output_tokens") {
+        body_object.remove("max_output_tokens");
+    }
+    if !body_rules_handle_path(body_rules, "temperature") {
+        body_object.remove("temperature");
+    }
+    if !body_rules_handle_path(body_rules, "top_p") {
+        body_object.remove("top_p");
+    }
+    if !body_rules_handle_path(body_rules, "metadata") {
+        body_object.remove("metadata");
+    }
+    if !body_rules_handle_path(body_rules, "store") {
+        body_object.insert("store".to_string(), json!(false));
+    }
+    if !body_rules_handle_path(body_rules, "instructions")
+        && !body_object.contains_key("instructions")
+    {
+        body_object.insert("instructions".to_string(), json!("You are GPT-5."));
+    }
+
+    maybe_inject_codex_prompt_cache_key(
+        provider_request_body,
+        provider_type,
+        provider_api_format,
+        user_api_key_id,
+    );
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::BTreeMap;
+
+    use super::{
+        apply_codex_openai_cli_special_body_edits, apply_codex_openai_cli_special_headers,
+    };
+    use http::{HeaderMap, HeaderValue};
+    use serde_json::json;
+
+    #[test]
+    fn applies_codex_defaults_when_body_rules_do_not_handle_fields() {
+        let mut body = json!({
+            "model": "gpt-5",
+            "max_output_tokens": 128,
+            "temperature": 0.3,
+            "top_p": 0.9,
+            "metadata": {"client": "desktop"},
+            "store": true
+        });
+
+        apply_codex_openai_cli_special_body_edits(&mut body, "codex", "openai:cli", None, None);
+
+        assert!(body.get("max_output_tokens").is_none());
+        assert!(body.get("temperature").is_none());
+        assert!(body.get("top_p").is_none());
+        assert!(body.get("metadata").is_none());
+        assert_eq!(body["store"], false);
+        assert_eq!(body["instructions"], "You are GPT-5.");
+    }
+
+    #[test]
+    fn defers_to_user_body_rules_for_handled_fields() {
+        let body_rules = json!([
+            {"action":"set","path":"store","value":true},
+            {"action":"set","path":"instructions","value":"Keep custom"},
+            {"action":"set","path":"metadata","value":{"client":"desktop","mode":"custom"}},
+            {"action":"set","path":"top_p","value":0.5}
+        ]);
+        let mut body = json!({
+            "model": "gpt-5",
+            "max_output_tokens": 128,
+            "metadata": {"client": "desktop", "mode": "custom"},
+            "store": true,
+            "instructions": "Keep custom",
+            "top_p": 0.5
+        });
+
+        apply_codex_openai_cli_special_body_edits(
+            &mut body,
+            "codex",
+            "openai:compact",
+            Some(&body_rules),
+            None,
+        );
+
+        assert!(body.get("max_output_tokens").is_none());
+        assert_eq!(body["store"], true);
+        assert_eq!(body["instructions"], "Keep custom");
+        assert_eq!(body["metadata"]["mode"], "custom");
+        assert_eq!(body["top_p"], 0.5);
+    }
+
+    #[test]
+    fn injects_stable_prompt_cache_key_for_codex_requests() {
+        let mut body = json!({
+            "model": "gpt-5",
+            "input": "hello",
+        });
+
+        apply_codex_openai_cli_special_body_edits(
+            &mut body,
+            "codex",
+            "openai:cli",
+            None,
+            Some("key-123"),
+        );
+
+        assert_eq!(
+            body["prompt_cache_key"],
+            "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        );
+    }
+
+    #[test]
+    fn keeps_existing_prompt_cache_key_for_codex_requests() {
+        let mut body = json!({
+            "model": "gpt-5",
+            "input": "hello",
+            "prompt_cache_key": "existing-key",
+        });
+
+        apply_codex_openai_cli_special_body_edits(
+            &mut body,
+            "codex",
+            "openai:cli",
+            None,
+            Some("key-123"),
+        );
+
+        assert_eq!(body["prompt_cache_key"], "existing-key");
+    }
+
+    #[test]
+    fn injects_chatgpt_account_id_and_session_headers_for_codex_requests() {
+        let mut headers = BTreeMap::new();
+        let body = json!({
+            "model": "gpt-5",
+            "prompt_cache_key": "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3",
+        });
+
+        apply_codex_openai_cli_special_headers(
+            &mut headers,
+            &body,
+            &HeaderMap::new(),
+            "codex",
+            "openai:cli",
+            Some("trace-codex-123"),
+            Some(r#"{"account_id":"acc-123"}"#),
+        );
+
+        assert_eq!(
+            headers.get("chatgpt-account-id"),
+            Some(&"acc-123".to_string())
+        );
+        assert_eq!(
+            headers.get("x-client-request-id"),
+            Some(&"trace-codex-123".to_string())
+        );
+        assert_eq!(
+            headers.get("session_id"),
+            Some(&"ab5ecce4f0d110fe".to_string())
+        );
+        assert_eq!(
+            headers.get("conversation_id"),
+            Some(&"ab5ecce4f0d110fe".to_string())
+        );
+    }
+
+    #[test]
+    fn respects_existing_codex_request_and_session_headers() {
+        let mut headers = BTreeMap::new();
+        headers.insert(
+            "x-client-request-id".to_string(),
+            "kept-by-rule-request".to_string(),
+        );
+        headers.insert("session_id".to_string(), "kept-by-rule".to_string());
+        let body = json!({
+            "model": "gpt-5",
+            "prompt_cache_key": "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3",
+        });
+        let mut original_headers = HeaderMap::new();
+        original_headers.insert(
+            "x-client-request-id",
+            HeaderValue::from_static("user-specified-request"),
+        );
+        original_headers.insert(
+            "session_id",
+            HeaderValue::from_static("user-specified-session"),
+        );
+        original_headers.insert(
+            "conversation_id",
+            HeaderValue::from_static("user-specified-conversation"),
+        );
+
+        apply_codex_openai_cli_special_headers(
+            &mut headers,
+            &body,
+            &original_headers,
+            "codex",
+            "openai:cli",
+            Some("trace-codex-123"),
+            Some(r#"{"account_id":"acc-123"}"#),
+        );
+
+        assert_eq!(
+            headers.get("x-client-request-id"),
+            Some(&"kept-by-rule-request".to_string())
+        );
+        assert_eq!(headers.get("session_id"), Some(&"kept-by-rule".to_string()));
+        assert!(headers.get("conversation_id").is_none());
+    }
+
+    #[test]
+    fn skips_conversation_id_for_compact_codex_requests() {
+        let mut headers = BTreeMap::new();
+        let body = json!({
+            "model": "gpt-5",
+            "prompt_cache_key": "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3",
+        });
+
+        apply_codex_openai_cli_special_headers(
+            &mut headers,
+            &body,
+            &HeaderMap::new(),
+            "codex",
+            "openai:compact",
+            Some("trace-codex-compact-123"),
+            Some(r#"{"account_id":"acc-123"}"#),
+        );
+
+        assert_eq!(
+            headers.get("chatgpt-account-id"),
+            Some(&"acc-123".to_string())
+        );
+        assert_eq!(
+            headers.get("x-client-request-id"),
+            Some(&"trace-codex-compact-123".to_string())
+        );
+        assert_eq!(
+            headers.get("session_id"),
+            Some(&"ab5ecce4f0d110fe".to_string())
+        );
+        assert!(headers.get("conversation_id").is_none());
+    }
+}

apps/aether-gateway/src/ai_pipeline/planner/standard/family/build.rs

@@ -1,11 +1,11 @@
 use tracing::warn;
 
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::ai_pipeline::planner::plan_builders::{
     build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
     build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlDecision;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
 use super::candidates::{

apps/aether-gateway/src/ai_pipeline/planner/standard/family/candidates.rs

@@ -1,16 +1,19 @@
 use std::collections::BTreeSet;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 use serde_json::json;
 use tracing::warn;
 use uuid::Uuid;
 
-use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::auth_snapshot_facade::{
-use crate::scheduler::{
+    read_auth_api_key_snapshot, GatewayAuthApiKeySnapshot,
-    current_unix_secs, list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
 };
+use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::planner::candidate_runtime_facade::persist_available_local_candidate;
+use crate::ai_pipeline::planner::scheduler_facade::list_selectable_candidates;
+use crate::clock::current_unix_secs;
 use crate::{append_execution_contract_fields_to_value, AppState, GatewayError};
 
 use super::types::{
@@ -42,8 +45,8 @@ pub(super) async fn resolve_local_standard_decision_input(
         LocalStandardSourceFamily::Gemini => extract_gemini_model_from_path(parts.uri.path())?,
     };
 
-    let auth_snapshot = match state
+    let auth_snapshot = match read_auth_api_key_snapshot(
-        .read_auth_api_key_snapshot(
+        state,
         &auth_context.user_id,
         &auth_context.api_key_id,
         current_unix_secs(),
@@ -157,47 +160,19 @@ pub(super) async fn materialize_local_standard_candidate_attempts(
             candidate.endpoint_api_format.as_str(),
         );
 
-        let stored_candidate_id = match state
+        let stored_candidate_id = persist_available_local_candidate(
-            .upsert_request_candidate(UpsertRequestCandidateRecord {
+            state,
-                id: candidate_id.clone(),
+            trace_id,
-                request_id: trace_id.to_string(),
+            &input.auth_context.user_id,
-                user_id: Some(input.auth_context.user_id.clone()),
+            &input.auth_context.api_key_id,
-                api_key_id: Some(input.auth_context.api_key_id.clone()),
+            &candidate,
-                username: None,
+            candidate_index as u32,
-                api_key_name: None,
+            &candidate_id,
-                candidate_index: candidate_index as u32,
+            Some(extra_data),
-                retry_index: 0,
+            created_at_unix_secs,
-                provider_id: Some(candidate.provider_id.clone()),
+            "gateway local standard decision request candidate upsert failed",
-                endpoint_id: Some(candidate.endpoint_id.clone()),
+        )
-                key_id: Some(candidate.key_id.clone()),
+        .await;
-                status: RequestCandidateStatus::Available,
-                skip_reason: None,
-                is_cached: Some(false),
-                status_code: None,
-                error_type: None,
-                error_message: None,
-                latency_ms: None,
-                concurrent_requests: None,
-                extra_data: Some(extra_data),
-                required_capabilities: candidate.key_capabilities.clone(),
-                created_at_unix_secs: Some(created_at_unix_secs),
-                started_at_unix_secs: None,
-                finished_at_unix_secs: None,
-            })
-            .await
-        {
-            Ok(Some(stored)) => stored.id,
-            Ok(None) => candidate_id.clone(),
-            Err(err) => {
-                warn!(
-                    trace_id = %trace_id,
-                    api_format = spec.api_format,
-                    error = ?err,
-                    "gateway local standard decision request candidate upsert failed"
-                );
-                candidate_id.clone()
-            }
-        };
 
         attempts.push(LocalStandardCandidateAttempt {
             candidate,
@@ -210,9 +185,9 @@ pub(super) async fn materialize_local_standard_candidate_attempts(
 }
 
 fn auth_snapshot_allows_cross_format_candidate(
-    auth_snapshot: &crate::data::auth::GatewayAuthApiKeySnapshot,
+    auth_snapshot: &GatewayAuthApiKeySnapshot,
     requested_model: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
 ) -> bool {
     if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
         let provider_allowed = allowed_providers.iter().any(|value| {

apps/aether-gateway/src/ai_pipeline/planner/standard/family/payload.rs

@@ -1,20 +1,25 @@
 use std::collections::BTreeMap;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
 use serde_json::json;
 use tracing::warn;
 
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::control_facade::collect_control_headers;
-use crate::headers::collect_control_headers;
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::planner::candidate_runtime_facade::persist_skipped_local_candidate;
-    build_openai_passthrough_headers, ensure_upstream_auth_header,
+use crate::ai_pipeline::planner::common::force_upstream_streaming_for_provider;
-};
+use crate::ai_pipeline::planner::standard::apply_codex_openai_cli_special_headers;
-use crate::provider_transport::{
+use crate::ai_pipeline::planner::transport_facade::{
-    apply_local_header_rules, resolve_transport_execution_timeouts,
+    read_provider_transport_snapshot, resolve_local_oauth_request_auth,
-    resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
     LocalResolvedOAuthRequestAuth,
 };
-use crate::scheduler::current_unix_secs;
+use crate::ai_pipeline::provider_transport_facade::auth::{
+    build_openai_passthrough_headers, ensure_upstream_auth_header,
+};
+use crate::ai_pipeline::provider_transport_facade::{
+    apply_local_header_rules, resolve_transport_execution_timeouts,
+    resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
+};
+use crate::clock::current_unix_secs;
 use crate::{
     append_execution_contract_fields_to_value, AppState, GatewayControlSyncDecisionResponse,
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
@@ -47,8 +52,8 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
         return None;
     };
 
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -107,12 +112,10 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
         return None;
     }
 
-    let resolved_auth = crate::ai_pipeline::conversion::request_conversion_direct_auth(
+    let resolved_auth =
-        &transport,
+        crate::ai_pipeline::conversion::request_conversion_direct_auth(&transport, conversion_kind);
-        conversion_kind,
-    );
     let oauth_auth = if resolved_auth.is_none() {
-        match state.resolve_local_oauth_request_auth(&transport).await {
+        match resolve_local_oauth_request_auth(state, &transport).await {
             Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
             Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
             Ok(None) => None,
@@ -160,15 +163,22 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
         return None;
     }
 
+    let upstream_is_stream = spec.require_streaming
+        || force_upstream_streaming_for_provider(
+            transport.provider.provider_type.as_str(),
+            provider_api_format.as_str(),
+        );
     let provider_request_body =
         match crate::ai_pipeline::planner::standard::build_standard_request_body(
             body_json,
             spec.api_format,
             &mapped_model,
+            transport.provider.provider_type.as_str(),
             provider_api_format.as_str(),
             parts.uri.path(),
-            spec.require_streaming,
+            upstream_is_stream,
             transport.endpoint.body_rules.as_ref(),
+            Some(input.auth_context.api_key_id.as_str()),
         ) {
             Some(body) => body,
             None => {
@@ -186,13 +196,12 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
             }
         };
 
-    let upstream_url =
+    let upstream_url = match crate::ai_pipeline::planner::standard::build_standard_upstream_url(
-        match crate::ai_pipeline::planner::standard::build_standard_upstream_url(
         parts,
         &transport,
         &mapped_model,
         provider_api_format.as_str(),
-            spec.require_streaming,
+        upstream_is_stream,
     ) {
         Some(url) => url,
         None => {
@@ -236,8 +245,17 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
         .await;
         return None;
     }
+    apply_codex_openai_cli_special_headers(
+        &mut provider_request_headers,
+        &provider_request_body,
+        &parts.headers,
+        transport.provider.provider_type.as_str(),
+        provider_api_format.as_str(),
+        Some(trace_id),
+        transport.key.decrypted_auth_config.as_deref(),
+    );
     ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
-    if spec.require_streaming {
+    if upstream_is_stream {
         provider_request_headers
             .entry("accept".to_string())
             .or_insert_with(|| "text/event-stream".to_string());
@@ -278,7 +296,7 @@ pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
         proxy: resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await,
         tls_profile: resolve_transport_tls_profile(&transport),
         timeouts: resolve_transport_execution_timeouts(&transport),
-        upstream_is_stream: spec.require_streaming,
+        upstream_is_stream,
         report_kind: Some(spec.report_kind.to_string()),
         report_context: Some(append_execution_contract_fields_to_value(
             json!({
@@ -319,46 +337,22 @@ pub(super) async fn mark_skipped_local_standard_candidate(
     state: &AppState,
     input: &LocalStandardDecisionInput,
     trace_id: &str,
-    candidate: &crate::scheduler::GatewayMinimalCandidateSelectionCandidate,
+    candidate: &aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     skip_reason: &'static str,
 ) {
-    if let Err(err) = state
+    persist_skipped_local_candidate(
-        .upsert_request_candidate(UpsertRequestCandidateRecord {
+        state,
-            id: candidate_id.to_string(),
+        trace_id,
-            request_id: trace_id.to_string(),
+        &input.auth_context.user_id,
-            user_id: Some(input.auth_context.user_id.clone()),
+        &input.auth_context.api_key_id,
-            api_key_id: Some(input.auth_context.api_key_id.clone()),
+        candidate,
-            username: None,
-            api_key_name: None,
         candidate_index,
-            retry_index: 0,
+        candidate_id,
-            provider_id: Some(candidate.provider_id.clone()),
-            endpoint_id: Some(candidate.endpoint_id.clone()),
-            key_id: Some(candidate.key_id.clone()),
-            status: RequestCandidateStatus::Skipped,
-            skip_reason: Some(skip_reason.to_string()),
-            is_cached: Some(false),
-            status_code: None,
-            error_type: None,
-            error_message: None,
-            latency_ms: None,
-            concurrent_requests: None,
-            extra_data: None,
-            required_capabilities: candidate.key_capabilities.clone(),
-            created_at_unix_secs: None,
-            started_at_unix_secs: None,
-            finished_at_unix_secs: Some(current_unix_secs()),
-        })
-        .await
-    {
-        warn!(
-            trace_id = %trace_id,
-            candidate_id = %candidate_id,
         skip_reason,
-            error = ?err,
+        current_unix_secs(),
-            "gateway local standard decision failed to persist skipped candidate"
+        "gateway local standard decision failed to persist skipped candidate",
-        );
+    )
-    }
+    .await;
 }

apps/aether-gateway/src/ai_pipeline/planner/standard/family/types.rs

@@ -1,35 +1,20 @@
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+use crate::ai_pipeline::control_facade::GatewayControlAuthContext;
-pub(crate) enum LocalStandardSourceFamily {
+use crate::ai_pipeline::planner::auth_snapshot_facade::GatewayAuthApiKeySnapshot;
-    Standard,
-    Gemini,
-}
 
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) use aether_ai_pipeline::planner::standard::family::{
-pub(crate) enum LocalStandardSourceMode {
+    LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec,
-    Chat,
+};
-    Cli,
-}
-
-#[derive(Debug, Clone, Copy)]
-pub(crate) struct LocalStandardSpec {
-    pub(crate) api_format: &'static str,
-    pub(crate) decision_kind: &'static str,
-    pub(crate) report_kind: &'static str,
-    pub(crate) family: LocalStandardSourceFamily,
-    pub(crate) mode: LocalStandardSourceMode,
-    pub(crate) require_streaming: bool,
-}
 
 #[derive(Debug, Clone)]
 pub(super) struct LocalStandardDecisionInput {
-    pub(super) auth_context: crate::control::GatewayControlAuthContext,
+    pub(super) auth_context: GatewayControlAuthContext,
     pub(super) requested_model: String,
-    pub(super) auth_snapshot: crate::data::auth::GatewayAuthApiKeySnapshot,
+    pub(super) auth_snapshot: GatewayAuthApiKeySnapshot,
 }
 
 #[derive(Debug, Clone)]
 pub(super) struct LocalStandardCandidateAttempt {
-    pub(super) candidate: crate::scheduler::GatewayMinimalCandidateSelectionCandidate,
+    pub(super) candidate: aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate,
     pub(super) candidate_index: u32,
     pub(super) candidate_id: String,
 }

apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/chat.rs

@@ -1,33 +0,0 @@
-use crate::ai_pipeline::planner::common::{
-    GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
-};
-
-use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
-
-pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        GEMINI_CHAT_SYNC_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "gemini:chat",
-            decision_kind: GEMINI_CHAT_SYNC_PLAN_KIND,
-            report_kind: "gemini_chat_sync_finalize",
-            family: LocalStandardSourceFamily::Gemini,
-            mode: LocalStandardSourceMode::Chat,
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        GEMINI_CHAT_STREAM_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "gemini:chat",
-            decision_kind: GEMINI_CHAT_STREAM_PLAN_KIND,
-            report_kind: "gemini_chat_stream_success",
-            family: LocalStandardSourceFamily::Gemini,
-            mode: LocalStandardSourceMode::Chat,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}

apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/cli.rs

@@ -1,33 +0,0 @@
-use crate::ai_pipeline::planner::common::{
-    GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
-};
-
-use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
-
-pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        GEMINI_CLI_SYNC_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "gemini:cli",
-            decision_kind: GEMINI_CLI_SYNC_PLAN_KIND,
-            report_kind: "gemini_cli_sync_finalize",
-            family: LocalStandardSourceFamily::Gemini,
-            mode: LocalStandardSourceMode::Cli,
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
-    match plan_kind {
-        GEMINI_CLI_STREAM_PLAN_KIND => Some(LocalStandardSpec {
-            api_format: "gemini:cli",
-            decision_kind: GEMINI_CLI_STREAM_PLAN_KIND,
-            report_kind: "gemini_cli_stream_success",
-            family: LocalStandardSourceFamily::Gemini,
-            mode: LocalStandardSourceMode::Cli,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}

apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/mod.rs

@@ -1,13 +1,23 @@
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
+use aether_ai_pipeline::planner::standard::gemini::{
+    resolve_stream_spec as resolve_pipeline_stream_spec,
+    resolve_sync_spec as resolve_pipeline_sync_spec,
+};
+
 use super::family::{
     maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
 };
 pub(crate) use crate::ai_pipeline::conversion::request::normalize_gemini_request_to_openai_chat_request;
 
-pub(crate) mod chat;
+pub(crate) fn resolve_sync_spec(plan_kind: &str) -> Option<super::family::LocalStandardSpec> {
-pub(crate) mod cli;
+    resolve_pipeline_sync_spec(plan_kind)
+}
+
+pub(crate) fn resolve_stream_spec(plan_kind: &str) -> Option<super::family::LocalStandardSpec> {
+    resolve_pipeline_stream_spec(plan_kind)
+}
 
 pub(crate) async fn maybe_build_sync_local_gemini_decision_payload(
     state: &AppState,
@@ -24,9 +34,7 @@ pub(crate) async fn maybe_build_sync_local_gemini_decision_payload(
         decision,
         body_json,
         plan_kind,
-        |plan_kind| {
+        resolve_sync_spec,
-            chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
-        },
     )
     .await
 }
@@ -46,9 +54,7 @@ pub(crate) async fn maybe_build_stream_local_gemini_decision_payload(
         decision,
         body_json,
         plan_kind,
-        |plan_kind| {
+        resolve_stream_spec,
-            chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
-        },
     )
     .await
 }

apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/plan_builders.rs

@@ -4,7 +4,7 @@ use super::{
     augment_sync_report_context, generic_decision_missing_exact_provider_request,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::provider_transport::ensure_upstream_auth_header;
+use crate::ai_pipeline::provider_transport_facade::ensure_upstream_auth_header;
 use crate::{GatewayControlSyncDecisionResponse, GatewayError};
 
 pub(crate) fn build_gemini_sync_plan_from_decision(

apps/aether-gateway/src/ai_pipeline/planner/standard/matrix.rs

@@ -1,79 +1,59 @@
-use std::collections::BTreeMap;
+use super::codex::apply_codex_openai_cli_special_body_edits;
-
+use crate::ai_pipeline::planner::transport_facade::GatewayProviderTransportSnapshot;
-use serde_json::{json, Map, Value};
+use crate::ai_pipeline::provider_transport_facade::apply_local_body_rules;
-
+use crate::ai_pipeline::provider_transport_facade::url::{
-use super::{
-    claude::normalize_claude_request_to_openai_chat_request,
-    gemini::normalize_gemini_request_to_openai_chat_request,
-};
-use crate::ai_pipeline::conversion::request::{
-    convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
-    convert_openai_chat_request_to_openai_cli_request,
-    normalize_openai_cli_request_to_openai_chat_request,
-};
-use crate::ai_pipeline::conversion::{request_conversion_kind, RequestConversionKind};
-use crate::provider_transport::apply_local_body_rules;
-use crate::provider_transport::url::{
     build_claude_messages_url, build_gemini_content_url, build_openai_chat_url,
     build_openai_cli_url, build_passthrough_path_url,
 };
+use aether_ai_pipeline::planner::matrix::build_standard_request_body_from_canonical;
+pub(crate) use aether_ai_pipeline::planner::standard::normalize_standard_request_to_openai_chat_request;
+use serde_json::{json, Value};
 
 pub(crate) fn build_standard_request_body(
     body_json: &Value,
     client_api_format: &str,
     mapped_model: &str,
+    provider_type: &str,
     provider_api_format: &str,
     request_path: &str,
     upstream_is_stream: bool,
     body_rules: Option<&Value>,
+    user_api_key_id: Option<&str>,
 ) -> Option<Value> {
     let canonical_request = normalize_standard_request_to_openai_chat_request(
         body_json,
         client_api_format,
         request_path,
     )?;
-    let conversion_kind = request_conversion_kind(client_api_format, provider_api_format)?;
+    if cfg!(test) {
-    let mut provider_request_body = match conversion_kind {
+        println!("canonical_request: {canonical_request:#?}");
-        RequestConversionKind::ToOpenAIChat => {
-            build_openai_chat_request_body(&canonical_request, mapped_model, upstream_is_stream)?
     }
-        RequestConversionKind::ToOpenAIFamilyCli => {
+    let mut provider_request_body = build_standard_request_body_from_canonical(
-            convert_openai_chat_request_to_openai_cli_request(
         &canonical_request,
         mapped_model,
+        provider_api_format,
         upstream_is_stream,
-                false,
+    )?;
-            )?
+    if cfg!(test) {
+        println!("provider_request_body before rules: {provider_request_body:#?}");
     }
-        RequestConversionKind::ToOpenAICompact => {
-            convert_openai_chat_request_to_openai_cli_request(
-                &canonical_request,
-                mapped_model,
-                false,
-                true,
-            )?
-        }
-        RequestConversionKind::ToClaudeStandard => convert_openai_chat_request_to_claude_request(
-            &canonical_request,
-            mapped_model,
-            upstream_is_stream,
-        )?,
-        RequestConversionKind::ToGeminiStandard => convert_openai_chat_request_to_gemini_request(
-            &canonical_request,
-            mapped_model,
-            upstream_is_stream,
-        )?,
-    };
 
     if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
         return None;
     }
+    apply_codex_openai_cli_special_body_edits(
+        &mut provider_request_body,
+        provider_type,
+        provider_api_format,
+        body_rules,
+        user_api_key_id,
+    );
     Some(provider_request_body)
 }
 
 pub(crate) fn build_standard_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     mapped_model: &str,
     provider_api_format: &str,
     upstream_is_stream: bool,
@@ -119,42 +99,6 @@ pub(crate) fn build_standard_upstream_url(
     }
 }
 
-pub(crate) fn normalize_standard_request_to_openai_chat_request(
-    body_json: &Value,
-    client_api_format: &str,
-    request_path: &str,
-) -> Option<Value> {
-    match client_api_format.trim().to_ascii_lowercase().as_str() {
-        "openai:chat" => Some(body_json.clone()),
-        "openai:cli" | "openai:compact" => {
-            normalize_openai_cli_request_to_openai_chat_request(body_json)
-        }
-        "claude:chat" | "claude:cli" => normalize_claude_request_to_openai_chat_request(body_json),
-        "gemini:chat" | "gemini:cli" => {
-            normalize_gemini_request_to_openai_chat_request(body_json, request_path)
-        }
-        _ => None,
-    }
-}
-
-fn build_openai_chat_request_body(
-    body_json: &Value,
-    mapped_model: &str,
-    upstream_is_stream: bool,
-) -> Option<Value> {
-    let request_body_object = body_json.as_object()?;
-    let mut provider_request_body = serde_json::Map::from_iter(
-        request_body_object
-            .iter()
-            .map(|(key, value)| (key.clone(), value.clone())),
-    );
-    provider_request_body.insert("model".to_string(), Value::String(mapped_model.to_string()));
-    if upstream_is_stream {
-        provider_request_body.insert("stream".to_string(), Value::Bool(true));
-    }
-    Some(Value::Object(provider_request_body))
-}
-
 #[cfg(test)]
 mod tests {
     use super::build_standard_request_body;
@@ -178,10 +122,12 @@ mod tests {
             &request,
             "claude:chat",
             "gpt-5",
+            "openai",
             "openai:chat",
             "/v1/messages",
             false,
             None,
+            None,
         )
         .expect("claude chat should convert to openai chat");
 
@@ -210,10 +156,12 @@ mod tests {
             &request,
             "gemini:chat",
             "claude-sonnet-4-5",
+            "anthropic",
             "claude:chat",
             "/v1beta/models/gemini-2.5-pro:generateContent",
             false,
             None,
+            None,
         )
         .expect("gemini chat should convert to claude chat");
 
@@ -244,10 +192,12 @@ mod tests {
             &request,
             "claude:cli",
             "gemini-2.5-pro",
+            "google",
             "gemini:cli",
             "/v1/messages",
             false,
             None,
+            None,
         )
         .expect("claude cli should convert to gemini cli");
 
@@ -257,4 +207,133 @@ mod tests {
             "Need CLI output"
         );
     }
+
+    #[test]
+    fn builds_openai_cli_request_from_claude_cli_source_with_forced_stream() {
+        let request = json!({
+            "model": "claude-sonnet-4-5",
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [{"type": "text", "text": "Need OpenAI CLI output"}]
+                }
+            ],
+            "max_tokens": 64
+        });
+
+        let converted = build_standard_request_body(
+            &request,
+            "claude:cli",
+            "gpt-5",
+            "openai",
+            "openai:cli",
+            "/v1/messages",
+            true,
+            None,
+            None,
+        )
+        .expect("claude cli should convert to openai cli");
+
+        assert_eq!(converted["model"], "gpt-5");
+        assert_eq!(converted["input"][0]["role"], "user");
+        assert_eq!(converted["input"][0]["content"][0]["type"], "input_text");
+        assert_eq!(
+            converted["input"][0]["content"][0]["text"],
+            "Need OpenAI CLI output"
+        );
+        assert_eq!(converted["stream"], true);
+    }
+
+    #[test]
+    fn strips_metadata_for_codex_openai_cli_requests() {
+        let request = json!({
+            "model": "claude-sonnet-4-5",
+            "metadata": {"trace_id": "abc"},
+            "messages": [{
+                "role": "user",
+                "content": [{"type": "text", "text": "Need OpenAI CLI output"}]
+            }],
+            "max_tokens": 64
+        });
+
+        let converted = build_standard_request_body(
+            &request,
+            "claude:cli",
+            "gpt-5",
+            "codex",
+            "openai:cli",
+            "/v1/messages",
+            true,
+            None,
+            None,
+        )
+        .expect("claude cli should convert to codex cli");
+
+        assert!(converted.get("metadata").is_none());
+    }
+
+    #[test]
+    fn applies_codex_defaults_unless_body_rules_handle_the_field() {
+        let request = json!({
+            "model": "claude-sonnet-4-5",
+            "metadata": {"trace_id": "abc"},
+            "messages": [{
+                "role": "user",
+                "content": [{"type": "text", "text": "Need OpenAI CLI output"}]
+            }],
+            "max_tokens": 64
+        });
+        let body_rules = json!([
+            {"action":"set","path":"store","value":true},
+            {"action":"set","path":"instructions","value":"Custom instructions"},
+            {"action":"set","path":"metadata","value":{"trace_id":"keep-me"}}
+        ]);
+
+        let converted = build_standard_request_body(
+            &request,
+            "claude:cli",
+            "gpt-5",
+            "codex",
+            "openai:cli",
+            "/v1/messages",
+            true,
+            Some(&body_rules),
+            None,
+        )
+        .expect("claude cli should convert to codex cli");
+
+        assert_eq!(converted["store"], true);
+        assert_eq!(converted["instructions"], "Custom instructions");
+        assert_eq!(converted["metadata"]["trace_id"], "keep-me");
+    }
+
+    #[test]
+    fn injects_codex_prompt_cache_key_for_standard_requests() {
+        let request = json!({
+            "model": "claude-sonnet-4-5",
+            "messages": [{
+                "role": "user",
+                "content": [{"type": "text", "text": "Need OpenAI CLI output"}]
+            }],
+            "max_tokens": 64
+        });
+
+        let converted = build_standard_request_body(
+            &request,
+            "claude:cli",
+            "gpt-5",
+            "codex",
+            "openai:cli",
+            "/v1/messages",
+            true,
+            None,
+            Some("key-123"),
+        )
+        .expect("claude cli should convert to codex cli");
+
+        assert_eq!(
+            converted["prompt_cache_key"],
+            "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        );
+    }
 }

apps/aether-gateway/src/ai_pipeline/planner/standard/mod.rs

@@ -3,16 +3,18 @@
 //! This groups the standard planning surface in one place:
 //! request-side conversion, matrix registry, and decision payload builders.
 
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
 pub(crate) mod claude;
+mod codex;
 pub(crate) mod family;
 pub(crate) mod gemini;
 mod matrix;
 mod normalize;
 pub(crate) mod openai;
 
+pub(crate) use self::codex::apply_codex_openai_cli_special_headers;
 pub(crate) use self::matrix::{
     build_standard_request_body, build_standard_upstream_url,
     normalize_standard_request_to_openai_chat_request,

apps/aether-gateway/src/ai_pipeline/planner/standard/normalize.rs

@@ -3,20 +3,23 @@ use std::collections::BTreeMap;
 use serde_json::{json, Map, Value};
 use url::form_urlencoded;
 
-use crate::ai_pipeline::conversion::request::{
+use super::codex::apply_codex_openai_cli_special_body_edits;
-    convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
-    convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
-    normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
-};
 use crate::ai_pipeline::conversion::{request_conversion_kind, RequestConversionKind};
-use crate::provider_transport::antigravity::{
+use crate::ai_pipeline::planner::transport_facade::GatewayProviderTransportSnapshot;
+use crate::ai_pipeline::provider_transport_facade::antigravity::{
     build_antigravity_v1internal_url, AntigravityRequestUrlAction,
 };
-use crate::provider_transport::apply_local_body_rules;
+use crate::ai_pipeline::provider_transport_facade::apply_local_body_rules;
-use crate::provider_transport::url::{
+use crate::ai_pipeline::provider_transport_facade::url::{
     build_claude_messages_url, build_gemini_content_url, build_openai_chat_url,
     build_openai_cli_url, build_passthrough_path_url,
 };
+use aether_ai_pipeline::planner::standard::normalize::{
+    build_cross_format_openai_chat_request_body as pipeline_build_cross_format_openai_chat_request_body,
+    build_cross_format_openai_cli_request_body as pipeline_build_cross_format_openai_cli_request_body,
+    build_local_openai_chat_request_body as pipeline_build_local_openai_chat_request_body,
+    build_local_openai_cli_request_body as pipeline_build_local_openai_cli_request_body,
+};
 
 pub(crate) fn build_local_openai_chat_request_body(
     body_json: &Value,
@@ -24,17 +27,8 @@ pub(crate) fn build_local_openai_chat_request_body(
     upstream_is_stream: bool,
     body_rules: Option<&Value>,
 ) -> Option<Value> {
-    let request_body_object = body_json.as_object()?;
+    let mut provider_request_body =
-    let mut provider_request_body = serde_json::Map::from_iter(
+        pipeline_build_local_openai_chat_request_body(body_json, mapped_model, upstream_is_stream)?;
-        request_body_object
-            .iter()
-            .map(|(key, value)| (key.clone(), value.clone())),
-    );
-    provider_request_body.insert("model".to_string(), Value::String(mapped_model.to_string()));
-    if upstream_is_stream {
-        provider_request_body.insert("stream".to_string(), Value::Bool(true));
-    }
-    let mut provider_request_body = Value::Object(provider_request_body);
     if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
         return None;
     }
@@ -43,7 +37,7 @@ pub(crate) fn build_local_openai_chat_request_body(
 
 pub(crate) fn build_local_openai_chat_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
 ) -> Option<String> {
     let custom_path = transport
         .endpoint
@@ -66,45 +60,34 @@ pub(crate) fn build_local_openai_chat_upstream_url(
 pub(crate) fn build_cross_format_openai_chat_request_body(
     body_json: &Value,
     mapped_model: &str,
+    provider_type: &str,
     provider_api_format: &str,
     upstream_is_stream: bool,
     body_rules: Option<&Value>,
+    user_api_key_id: Option<&str>,
 ) -> Option<Value> {
-    let conversion_kind = request_conversion_kind("openai:chat", provider_api_format)?;
+    let mut provider_request_body = pipeline_build_cross_format_openai_chat_request_body(
-    let mut provider_request_body = match conversion_kind {
-        RequestConversionKind::ToClaudeStandard => convert_openai_chat_request_to_claude_request(
         body_json,
         mapped_model,
+        provider_api_format,
         upstream_is_stream,
-        )?,
+    )?;
-        RequestConversionKind::ToGeminiStandard => convert_openai_chat_request_to_gemini_request(
-            body_json,
-            mapped_model,
-            upstream_is_stream,
-        )?,
-        RequestConversionKind::ToOpenAIFamilyCli => {
-            convert_openai_chat_request_to_openai_cli_request(
-                body_json,
-                mapped_model,
-                upstream_is_stream,
-                false,
-            )?
-        }
-        RequestConversionKind::ToOpenAICompact => {
-            convert_openai_chat_request_to_openai_cli_request(body_json, mapped_model, false, true)?
-        }
-        _ => return None,
-    };
-
     if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
         return None;
     }
+    apply_codex_openai_cli_special_body_edits(
+        &mut provider_request_body,
+        provider_type,
+        provider_api_format,
+        body_rules,
+        user_api_key_id,
+    );
     Some(provider_request_body)
 }
 
 pub(crate) fn build_cross_format_openai_chat_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     mapped_model: &str,
     provider_api_format: &str,
     upstream_is_stream: bool,
@@ -151,22 +134,23 @@ pub(crate) fn build_local_openai_cli_request_body(
     body_json: &Value,
     mapped_model: &str,
     require_streaming: bool,
+    provider_type: &str,
+    provider_api_format: &str,
     body_rules: Option<&Value>,
+    user_api_key_id: Option<&str>,
 ) -> Option<Value> {
-    let request_body_object = body_json.as_object()?;
+    let mut provider_request_body =
-    let mut provider_request_body = serde_json::Map::from_iter(
+        pipeline_build_local_openai_cli_request_body(body_json, mapped_model, require_streaming)?;
-        request_body_object
-            .iter()
-            .map(|(key, value)| (key.clone(), value.clone())),
-    );
-    provider_request_body.insert("model".to_string(), Value::String(mapped_model.to_string()));
-    if require_streaming {
-        provider_request_body.insert("stream".to_string(), Value::Bool(true));
-    }
-    let mut provider_request_body = Value::Object(provider_request_body);
     if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
         return None;
     }
+    apply_codex_openai_cli_special_body_edits(
+        &mut provider_request_body,
+        provider_type,
+        provider_api_format,
+        body_rules,
+        user_api_key_id,
+    );
     Some(provider_request_body)
 }
 
@@ -176,48 +160,33 @@ pub(crate) fn build_cross_format_openai_cli_request_body(
     client_api_format: &str,
     provider_api_format: &str,
     upstream_is_stream: bool,
+    provider_type: &str,
     body_rules: Option<&Value>,
+    user_api_key_id: Option<&str>,
 ) -> Option<Value> {
-    let chat_like_request = normalize_openai_cli_request_to_openai_chat_request(body_json)?;
+    let mut provider_request_body = pipeline_build_cross_format_openai_cli_request_body(
-    let conversion_kind = request_conversion_kind(client_api_format, provider_api_format)?;
+        body_json,
-    let mut provider_request_body = match conversion_kind {
-        RequestConversionKind::ToOpenAIFamilyCli => {
-            convert_openai_chat_request_to_openai_cli_request(
-                &chat_like_request,
         mapped_model,
+        client_api_format,
+        provider_api_format,
         upstream_is_stream,
-                false,
+    )?;
-            )?
-        }
-        RequestConversionKind::ToOpenAICompact => {
-            convert_openai_chat_request_to_openai_cli_request(
-                &chat_like_request,
-                mapped_model,
-                false,
-                true,
-            )?
-        }
-        RequestConversionKind::ToClaudeStandard => convert_openai_chat_request_to_claude_request(
-            &chat_like_request,
-            mapped_model,
-            upstream_is_stream,
-        )?,
-        RequestConversionKind::ToGeminiStandard => convert_openai_chat_request_to_gemini_request(
-            &chat_like_request,
-            mapped_model,
-            upstream_is_stream,
-        )?,
-        _ => return None,
-    };
     if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
         return None;
     }
+    apply_codex_openai_cli_special_body_edits(
+        &mut provider_request_body,
+        provider_type,
+        provider_api_format,
+        body_rules,
+        user_api_key_id,
+    );
     Some(provider_request_body)
 }
 
 pub(crate) fn build_local_openai_cli_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     compact: bool,
 ) -> Option<String> {
     let custom_path = transport
@@ -241,7 +210,7 @@ pub(crate) fn build_local_openai_cli_upstream_url(
 
 pub(crate) fn build_cross_format_openai_cli_upstream_url(
     parts: &http::request::Parts,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     mapped_model: &str,
     client_api_format: &str,
     provider_api_format: &str,
@@ -325,6 +294,8 @@ mod tests {
             "openai:compact",
             "openai:cli",
             false,
+            "openai",
+            None,
             None,
         )
         .expect("compact to openai cli body should build");
@@ -333,4 +304,119 @@ mod tests {
         assert_eq!(provider_request_body["input"][0]["type"], "message");
         assert_eq!(provider_request_body["input"][0]["role"], "user");
     }
+
+    #[test]
+    fn strips_metadata_for_codex_openai_cli_requests() {
+        let body_json = json!({
+            "model": "claude-sonnet-4-5",
+            "metadata": {"trace_id": "abc"},
+            "messages": [{
+                "role": "user",
+                "content": [{"type": "text", "text": "hello"}]
+            }],
+        });
+
+        let provider_request_body = build_cross_format_openai_cli_request_body(
+            &body_json,
+            "gpt-5-upstream",
+            "claude:cli",
+            "openai:cli",
+            true,
+            "codex",
+            None,
+            None,
+        )
+        .expect("claude cli to codex request should build");
+
+        assert!(provider_request_body.get("metadata").is_none());
+    }
+
+    #[test]
+    fn applies_codex_defaults_unless_body_rules_handle_the_field() {
+        let body_json = json!({
+            "model": "claude-sonnet-4-5",
+            "messages": [{
+                "role": "user",
+                "content": [{"type": "text", "text": "hello"}]
+            }],
+            "metadata": {"trace_id": "abc"},
+            "store": true
+        });
+        let body_rules = json!([
+            {"action":"set","path":"store","value":true},
+            {"action":"set","path":"instructions","value":"Custom instructions"},
+            {"action":"set","path":"metadata","value":{"trace_id":"keep-me"}}
+        ]);
+
+        let provider_request_body = build_cross_format_openai_cli_request_body(
+            &body_json,
+            "gpt-5-upstream",
+            "claude:cli",
+            "openai:cli",
+            true,
+            "codex",
+            Some(&body_rules),
+            None,
+        )
+        .expect("claude cli to codex request should build");
+
+        assert_eq!(provider_request_body["store"], true);
+        assert_eq!(provider_request_body["instructions"], "Custom instructions");
+        assert_eq!(provider_request_body["metadata"]["trace_id"], "keep-me");
+    }
+
+    #[test]
+    fn injects_codex_prompt_cache_key_for_openai_cli_cross_format_requests() {
+        let body_json = json!({
+            "model": "claude-sonnet-4-5",
+            "messages": [{
+                "role": "user",
+                "content": [{"type": "text", "text": "hello"}]
+            }],
+        });
+
+        let provider_request_body = build_cross_format_openai_cli_request_body(
+            &body_json,
+            "gpt-5-upstream",
+            "claude:cli",
+            "openai:cli",
+            true,
+            "codex",
+            None,
+            Some("key-123"),
+        )
+        .expect("claude cli to codex request should build");
+
+        assert_eq!(
+            provider_request_body["prompt_cache_key"],
+            "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        );
+    }
+
+    #[test]
+    fn injects_codex_prompt_cache_key_for_openai_chat_cross_format_requests() {
+        let body_json = json!({
+            "model": "gpt-5",
+            "messages": [{
+                "role": "user",
+                "content": "hello"
+            }],
+        });
+
+        let provider_request_body = super::build_cross_format_openai_chat_request_body(
+            &body_json,
+            "gpt-5-upstream",
+            "codex",
+            "openai:cli",
+            false,
+            None,
+            Some("key-123"),
+        )
+        .expect("openai chat to codex request should build");
+
+        assert_eq!(
+            provider_request_body["prompt_cache_key"],
+            "172c39e6-c0a0-5a70-8b63-e0f8e0d185a3"
+        );
+    }
 }

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

@@ -1,55 +1,61 @@
 use std::collections::BTreeMap;
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
 use serde_json::json;
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::ai_pipeline::control_facade::{collect_control_headers, GatewayControlAuthContext};
 use crate::ai_pipeline::conversion::{
     request_conversion_direct_auth, request_conversion_kind, request_conversion_transport_supported,
 };
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::auth_snapshot_facade::GatewayAuthApiKeySnapshot;
 use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::planner::candidate_runtime_facade::{
+    persist_available_local_candidate, persist_skipped_local_candidate,
+};
 use crate::ai_pipeline::planner::common::{
     EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
     OPENAI_CHAT_STREAM_PLAN_KIND,
 };
+use crate::ai_pipeline::planner::executor_facade::mark_unused_local_candidate_items;
 use crate::ai_pipeline::planner::plan_builders::{
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+use crate::ai_pipeline::planner::transport_facade::{
-use crate::headers::collect_control_headers;
+    read_provider_transport_snapshot, resolve_local_oauth_request_auth,
-use crate::provider_transport::auth::{
+    GatewayProviderTransportSnapshot, LocalResolvedOAuthRequestAuth,
+};
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_openai_passthrough_headers, ensure_upstream_auth_header, resolve_local_openai_chat_auth,
 };
-use crate::provider_transport::policy::supports_local_openai_chat_transport;
+use crate::ai_pipeline::provider_transport_facade::policy::supports_local_openai_chat_transport;
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     apply_local_header_rules, resolve_transport_execution_timeouts,
     resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
-    LocalResolvedOAuthRequestAuth,
-};
-use crate::scheduler::{
-    record_local_request_candidate_status, GatewayMinimalCandidateSelectionCandidate,
 };
 use crate::{
     append_execution_contract_fields_to_value, AppState, GatewayControlSyncDecisionResponse,
 };
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 
 use super::plans::current_unix_secs;
 use crate::ai_pipeline::planner::standard::{
-    build_cross_format_openai_chat_request_body, build_cross_format_openai_chat_upstream_url,
+    apply_codex_openai_cli_special_headers, build_cross_format_openai_chat_request_body,
-    build_local_openai_chat_request_body, build_local_openai_chat_upstream_url,
+    build_cross_format_openai_chat_upstream_url, build_local_openai_chat_request_body,
+    build_local_openai_chat_upstream_url,
 };
 
 #[derive(Debug, Clone)]
 pub(super) struct LocalOpenAiChatDecisionInput {
-    pub(super) auth_context: crate::control::GatewayControlAuthContext,
+    pub(super) auth_context: GatewayControlAuthContext,
     pub(super) requested_model: String,
-    pub(super) auth_snapshot: crate::data::auth::GatewayAuthApiKeySnapshot,
+    pub(super) auth_snapshot: GatewayAuthApiKeySnapshot,
 }
 
 #[derive(Debug, Clone)]
 pub(super) struct LocalOpenAiChatCandidateAttempt {
-    pub(super) candidate: GatewayMinimalCandidateSelectionCandidate,
+    pub(super) candidate: SchedulerMinimalCandidateSelectionCandidate,
     pub(super) candidate_index: u32,
     pub(super) candidate_id: String,
 }
@@ -70,8 +76,8 @@ pub(super) async fn maybe_build_local_openai_chat_decision_payload_for_candidate
         candidate_index,
         candidate_id,
     } = attempt;
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -171,13 +177,13 @@ async fn build_same_format_local_openai_chat_decision_payload_for_candidate(
     trace_id: &str,
     body_json: &serde_json::Value,
     input: &LocalOpenAiChatDecisionInput,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     decision_kind: &str,
     report_kind: &str,
     upstream_is_stream: bool,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
 ) -> Option<GatewayControlSyncDecisionResponse> {
     if !supports_local_openai_chat_transport(transport) {
         mark_skipped_local_openai_chat_candidate(
@@ -194,7 +200,7 @@ async fn build_same_format_local_openai_chat_decision_payload_for_candidate(
     }
 
     let oauth_auth = if resolve_local_openai_chat_auth(transport).is_none() {
-        match state.resolve_local_oauth_request_auth(transport).await {
+        match resolve_local_oauth_request_auth(state, transport).await {
             Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
             Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
             Ok(None) => None,
@@ -300,6 +306,15 @@ async fn build_same_format_local_openai_chat_decision_payload_for_candidate(
         .await;
         return None;
     }
+    apply_codex_openai_cli_special_headers(
+        &mut provider_request_headers,
+        &provider_request_body,
+        &parts.headers,
+        transport.provider.provider_type.as_str(),
+        transport.endpoint.api_format.as_str(),
+        Some(trace_id),
+        transport.key.decrypted_auth_config.as_deref(),
+    );
     ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
     if upstream_is_stream {
         provider_request_headers
@@ -394,12 +409,12 @@ async fn build_cross_format_local_openai_chat_decision_payload_for_candidate(
     trace_id: &str,
     body_json: &serde_json::Value,
     input: &LocalOpenAiChatDecisionInput,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     decision_kind: &str,
     upstream_is_stream: bool,
-    transport: &crate::provider_transport::GatewayProviderTransportSnapshot,
+    transport: &GatewayProviderTransportSnapshot,
     provider_api_format: &str,
 ) -> Option<GatewayControlSyncDecisionResponse> {
     let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
@@ -425,7 +440,7 @@ async fn build_cross_format_local_openai_chat_decision_payload_for_candidate(
 
     let resolve_auth = request_conversion_direct_auth(transport, conversion_kind);
     let oauth_auth = if resolve_auth.is_none() {
-        match state.resolve_local_oauth_request_auth(transport).await {
+        match resolve_local_oauth_request_auth(state, transport).await {
             Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
             Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
             Ok(None) => None,
@@ -476,9 +491,11 @@ async fn build_cross_format_local_openai_chat_decision_payload_for_candidate(
     let Some(provider_request_body) = build_cross_format_openai_chat_request_body(
         body_json,
         &mapped_model,
+        transport.provider.provider_type.as_str(),
         provider_api_format.as_str(),
         upstream_is_stream,
         transport.endpoint.body_rules.as_ref(),
+        Some(input.auth_context.api_key_id.as_str()),
     ) else {
         mark_skipped_local_openai_chat_candidate(
             state,
@@ -539,6 +556,15 @@ async fn build_cross_format_local_openai_chat_decision_payload_for_candidate(
         .await;
         return None;
     }
+    apply_codex_openai_cli_special_headers(
+        &mut provider_request_headers,
+        &provider_request_body,
+        &parts.headers,
+        transport.provider.provider_type.as_str(),
+        provider_api_format.as_str(),
+        Some(trace_id),
+        transport.key.decrypted_auth_config.as_deref(),
+    );
     ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
     if upstream_is_stream {
         provider_request_headers
@@ -636,7 +662,7 @@ pub(super) async fn mark_skipped_local_openai_chat_candidate(
     state: &AppState,
     input: &LocalOpenAiChatDecisionInput,
     trace_id: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     skip_reason: &'static str,
@@ -649,50 +675,26 @@ pub(super) async fn mark_skipped_local_openai_chat_candidate(
         *diagnostic.skipped_candidate_count.get_or_insert(0) += 1;
     });
     let terminal_unix_secs = current_unix_secs();
-    if let Err(err) = state
+    persist_skipped_local_candidate(
-        .upsert_request_candidate(UpsertRequestCandidateRecord {
+        state,
-            id: candidate_id.to_string(),
+        trace_id,
-            request_id: trace_id.to_string(),
+        &input.auth_context.user_id,
-            user_id: Some(input.auth_context.user_id.clone()),
+        &input.auth_context.api_key_id,
-            api_key_id: Some(input.auth_context.api_key_id.clone()),
+        candidate,
-            username: None,
-            api_key_name: None,
         candidate_index,
-            retry_index: 0,
+        candidate_id,
-            provider_id: Some(candidate.provider_id.clone()),
-            endpoint_id: Some(candidate.endpoint_id.clone()),
-            key_id: Some(candidate.key_id.clone()),
-            status: RequestCandidateStatus::Skipped,
-            skip_reason: Some(skip_reason.to_string()),
-            is_cached: Some(false),
-            status_code: None,
-            error_type: None,
-            error_message: None,
-            latency_ms: None,
-            concurrent_requests: None,
-            extra_data: None,
-            required_capabilities: candidate.key_capabilities.clone(),
-            created_at_unix_secs: None,
-            started_at_unix_secs: None,
-            finished_at_unix_secs: Some(terminal_unix_secs),
-        })
-        .await
-    {
-        warn!(
-            trace_id = %trace_id,
-            candidate_id = %candidate_id,
         skip_reason,
-            error = ?err,
+        terminal_unix_secs,
-            "gateway local openai chat decision failed to persist skipped candidate"
+        "gateway local openai chat decision failed to persist skipped candidate",
-        );
+    )
-    }
+    .await;
 }
 
 pub(super) async fn materialize_local_openai_chat_candidate_attempts(
     state: &AppState,
     trace_id: &str,
     input: &LocalOpenAiChatDecisionInput,
-    candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
+    candidates: Vec<SchedulerMinimalCandidateSelectionCandidate>,
 ) -> Vec<LocalOpenAiChatCandidateAttempt> {
     let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
     let created_at_unix_secs = current_unix_secs();
@@ -727,46 +729,19 @@ pub(super) async fn materialize_local_openai_chat_candidate_attempts(
             candidate.endpoint_api_format.trim(),
         );
 
-        let candidate_id = match state
+        let candidate_id = persist_available_local_candidate(
-            .upsert_request_candidate(UpsertRequestCandidateRecord {
+            state,
-                id: generated_candidate_id.clone(),
+            trace_id,
-                request_id: trace_id.to_string(),
+            &input.auth_context.user_id,
-                user_id: Some(input.auth_context.user_id.clone()),
+            &input.auth_context.api_key_id,
-                api_key_id: Some(input.auth_context.api_key_id.clone()),
+            &candidate,
-                username: None,
+            candidate_index as u32,
-                api_key_name: None,
+            &generated_candidate_id,
-                candidate_index: candidate_index as u32,
+            Some(extra_data),
-                retry_index: 0,
+            created_at_unix_secs,
-                provider_id: Some(candidate.provider_id.clone()),
+            "gateway local openai chat decision request candidate upsert failed",
-                endpoint_id: Some(candidate.endpoint_id.clone()),
+        )
-                key_id: Some(candidate.key_id.clone()),
+        .await;
-                status: RequestCandidateStatus::Available,
-                skip_reason: None,
-                is_cached: Some(false),
-                status_code: None,
-                error_type: None,
-                error_message: None,
-                latency_ms: None,
-                concurrent_requests: None,
-                extra_data: Some(extra_data),
-                required_capabilities: candidate.key_capabilities.clone(),
-                created_at_unix_secs: Some(created_at_unix_secs),
-                started_at_unix_secs: None,
-                finished_at_unix_secs: None,
-            })
-            .await
-        {
-            Ok(Some(stored)) => stored.id,
-            Ok(None) => generated_candidate_id.clone(),
-            Err(err) => {
-                warn!(
-                    trace_id = %trace_id,
-                    error = ?err,
-                    "gateway local openai chat decision request candidate upsert failed"
-                );
-                generated_candidate_id.clone()
-            }
-        };
 
         attempts.push(LocalOpenAiChatCandidateAttempt {
             candidate,
@@ -782,22 +757,14 @@ pub(super) async fn mark_unused_local_openai_chat_candidates<T>(state: &AppState
 where
     T: LocalOpenAiChatPlanAndReport,
 {
-    for plan_and_report in remaining {
+    mark_unused_local_candidate_items(
-        record_local_request_candidate_status(
         state,
-            plan_and_report.plan(),
+        remaining,
-            plan_and_report.report_context(),
+        |item| item.plan(),
-            RequestCandidateStatus::Unused,
+        |item| item.report_context(),
-            None,
-            None,
-            None,
-            None,
-            None,
-            None,
     )
     .await;
 }
-}
 
 pub(super) trait LocalOpenAiChatPlanAndReport {
     fn plan(&self) -> &aether_contracts::ExecutionPlan;

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

@@ -1,10 +1,10 @@
-use serde_json::{Map, Value};
+use serde_json::Value;
 use tracing::warn;
 
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::ai_pipeline::planner::common::{
     OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
 };
-use crate::control::GatewayControlDecision;
 use crate::{
     AppState, GatewayControlSyncDecisionResponse, GatewayError, LocalExecutionRuntimeMissDiagnostic,
 };
@@ -30,10 +30,7 @@ pub(crate) async fn build_local_openai_chat_sync_plan_and_reports_for_kind(
     decision: &GatewayControlDecision,
     body_json: &serde_json::Value,
     plan_kind: &str,
-) -> Result<
+) -> Result<Vec<crate::ai_pipeline::planner::plan_builders::LocalSyncPlanAndReport>, GatewayError> {
-    Vec<crate::ai_pipeline::planner::plan_builders::LocalSyncPlanAndReport>,
-    GatewayError,
-> {
     build_local_openai_chat_sync_plan_and_reports(
         state, parts, trace_id, decision, body_json, plan_kind,
     )
@@ -47,10 +44,8 @@ pub(crate) async fn build_local_openai_chat_stream_plan_and_reports_for_kind(
     decision: &GatewayControlDecision,
     body_json: &serde_json::Value,
     plan_kind: &str,
-) -> Result<
+) -> Result<Vec<crate::ai_pipeline::planner::plan_builders::LocalStreamPlanAndReport>, GatewayError>
-    Vec<crate::ai_pipeline::planner::plan_builders::LocalStreamPlanAndReport>,
+{
-    GatewayError,
-> {
     build_local_openai_chat_stream_plan_and_reports(
         state, parts, trace_id, decision, body_json, plan_kind,
     )
@@ -205,76 +200,3 @@ pub(crate) async fn maybe_build_stream_local_decision_payload(
 
     Ok(None)
 }
-
-pub(crate) fn parse_openai_stop_sequences(stop: Option<&Value>) -> Option<Vec<Value>> {
-    match stop {
-        Some(Value::String(value)) if !value.trim().is_empty() => {
-            Some(vec![Value::String(value.clone())])
-        }
-        Some(Value::Array(values)) => Some(
-            values
-                .iter()
-                .filter_map(|value| value.as_str())
-                .map(str::trim)
-                .filter(|value| !value.is_empty())
-                .map(|value| Value::String(value.to_string()))
-                .collect::<Vec<_>>(),
-        )
-        .filter(|values| !values.is_empty()),
-        _ => None,
-    }
-}
-
-pub(crate) fn resolve_openai_chat_max_tokens(request: &Map<String, Value>) -> u64 {
-    request
-        .get("max_completion_tokens")
-        .and_then(value_as_u64)
-        .or_else(|| request.get("max_tokens").and_then(value_as_u64))
-        .unwrap_or(4096)
-}
-
-pub(crate) fn value_as_u64(value: &Value) -> Option<u64> {
-    value
-        .as_u64()
-        .or_else(|| value.as_i64().and_then(|value| u64::try_from(value).ok()))
-}
-
-pub(crate) fn copy_request_number_field(
-    request: &Map<String, Value>,
-    target: &mut Map<String, Value>,
-    key: &str,
-) {
-    copy_request_number_field_as(request, target, key, key);
-}
-
-pub(crate) fn copy_request_number_field_as(
-    request: &Map<String, Value>,
-    target: &mut Map<String, Value>,
-    source_key: &str,
-    target_key: &str,
-) {
-    if let Some(value) = request.get(source_key).cloned() {
-        if value.is_number() {
-            target.insert(target_key.to_string(), value);
-        }
-    }
-}
-
-pub(crate) fn map_openai_reasoning_effort_to_claude_output(value: &str) -> Option<&'static str> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "low" => Some("low"),
-        "medium" => Some("medium"),
-        "high" | "xhigh" => Some("high"),
-        _ => None,
-    }
-}
-
-pub(crate) fn map_openai_reasoning_effort_to_gemini_budget(value: &str) -> Option<u64> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "low" => Some(1024),
-        "medium" => Some(4096),
-        "high" => Some(8192),
-        "xhigh" => Some(16_384),
-        _ => None,
-    }
-}

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

@@ -1,6 +1,7 @@
 use std::collections::{BTreeMap, BTreeSet};
 use std::time::{SystemTime, UNIX_EPOCH};
 
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 use tracing::warn;
 
 use super::{
@@ -8,6 +9,9 @@ use super::{
     maybe_build_local_openai_chat_decision_payload_for_candidate, AppState, GatewayControlDecision,
     GatewayError, LocalExecutionRuntimeMissDiagnostic, LocalOpenAiChatDecisionInput,
 };
+use crate::ai_pipeline::planner::auth_snapshot_facade::{
+    read_auth_api_key_snapshot, GatewayAuthApiKeySnapshot,
+};
 use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
 use crate::ai_pipeline::planner::common::{
     OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
@@ -16,9 +20,7 @@ use crate::ai_pipeline::planner::plan_builders::{
     build_openai_chat_stream_plan_from_decision, build_openai_chat_sync_plan_from_decision,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::scheduler::{
+use crate::ai_pipeline::planner::scheduler_facade::list_selectable_candidates;
-    list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
-};
 
 pub(super) fn build_local_openai_chat_miss_diagnostic(
     decision: &GatewayControlDecision,
@@ -292,7 +294,7 @@ pub(super) async fn list_local_openai_chat_candidates(
     state: &AppState,
     input: &LocalOpenAiChatDecisionInput,
     require_streaming: bool,
-) -> Result<Vec<GatewayMinimalCandidateSelectionCandidate>, GatewayError> {
+) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, GatewayError> {
     let now_unix_secs = current_unix_secs();
     let mut combined = Vec::new();
     let mut seen = BTreeSet::new();
@@ -352,9 +354,9 @@ pub(super) async fn list_local_openai_chat_candidates(
 }
 
 fn auth_snapshot_allows_cross_format_openai_chat_candidate(
-    auth_snapshot: &crate::data::auth::GatewayAuthApiKeySnapshot,
+    auth_snapshot: &GatewayAuthApiKeySnapshot,
     requested_model: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
 ) -> bool {
     if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
         let provider_allowed = allowed_providers.iter().any(|value| {
@@ -438,8 +440,8 @@ pub(super) async fn resolve_local_openai_chat_decision_input(
     };
 
     let now_unix_secs = current_unix_secs();
-    let auth_snapshot = match state
+    let auth_snapshot = match read_auth_api_key_snapshot(
-        .read_auth_api_key_snapshot(
+        state,
         &auth_context.user_id,
         &auth_context.api_key_id,
         now_unix_secs,

apps/aether-gateway/src/ai_pipeline/planner/standard/openai/cli/decision.rs

@@ -1,46 +1,57 @@
 use std::collections::{BTreeMap, BTreeSet};
 
-use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 use serde_json::{json, Value};
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::ai_pipeline::control_facade::{
+    collect_control_headers, GatewayControlAuthContext, GatewayControlDecision,
+};
 use crate::ai_pipeline::conversion::{
     request_conversion_direct_auth, request_conversion_kind, request_conversion_transport_supported,
 };
-use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::execution_facade::{ConversionMode, ExecutionStrategy};
-use crate::ai_pipeline::planner::common::{
+use crate::ai_pipeline::planner::auth_snapshot_facade::{
-    EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+    read_auth_api_key_snapshot, GatewayAuthApiKeySnapshot,
 };
+use crate::ai_pipeline::planner::candidate_affinity::prefer_local_tunnel_owner_candidates;
+use crate::ai_pipeline::planner::candidate_runtime_facade::{
+    persist_available_local_candidate, persist_skipped_local_candidate,
+};
+use crate::ai_pipeline::planner::common::{
+    force_upstream_streaming_for_provider, EXECUTION_RUNTIME_STREAM_DECISION_ACTION,
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+};
+use crate::ai_pipeline::planner::executor_facade::mark_unused_local_candidate_items;
 use crate::ai_pipeline::planner::plan_builders::{
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
+use crate::ai_pipeline::planner::scheduler_facade::list_selectable_candidates;
 use crate::ai_pipeline::planner::standard::{
-    build_cross_format_openai_cli_request_body, build_cross_format_openai_cli_upstream_url,
+    apply_codex_openai_cli_special_headers, build_cross_format_openai_cli_request_body,
-    build_local_openai_cli_request_body, build_local_openai_cli_upstream_url,
+    build_cross_format_openai_cli_upstream_url, build_local_openai_cli_request_body,
+    build_local_openai_cli_upstream_url,
 };
-use crate::control::GatewayControlDecision;
+use crate::ai_pipeline::planner::transport_facade::{
-use crate::execution_runtime::{ConversionMode, ExecutionStrategy};
+    read_provider_transport_snapshot, resolve_local_oauth_request_auth,
-use crate::headers::collect_control_headers;
+    LocalResolvedOAuthRequestAuth,
-use crate::provider_transport::antigravity::{
+};
+use crate::ai_pipeline::provider_transport_facade::antigravity::{
     build_antigravity_safe_v1internal_request, build_antigravity_static_identity_headers,
     classify_local_antigravity_request_support, AntigravityEnvelopeRequestType,
     AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport,
 };
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_openai_passthrough_headers, ensure_upstream_auth_header, resolve_local_gemini_auth,
     resolve_local_standard_auth,
 };
-use crate::provider_transport::policy::supports_local_standard_transport_with_network;
+use crate::ai_pipeline::provider_transport_facade::policy::supports_local_standard_transport_with_network;
-use crate::provider_transport::{
+use crate::ai_pipeline::provider_transport_facade::{
     apply_local_body_rules, apply_local_header_rules, resolve_transport_execution_timeouts,
     resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
-    LocalResolvedOAuthRequestAuth,
-};
-use crate::scheduler::{
-    current_unix_secs, list_selectable_candidates, record_local_request_candidate_status,
-    GatewayMinimalCandidateSelectionCandidate,
 };
+use crate::clock::current_unix_secs;
 use crate::{
     append_execution_contract_fields_to_value, AppState, GatewayControlSyncDecisionResponse,
     GatewayError,
@@ -48,25 +59,18 @@ use crate::{
 
 const ANTIGRAVITY_ENVELOPE_NAME: &str = "antigravity:v1internal";
 
-#[derive(Debug, Clone, Copy)]
+pub(super) use aether_ai_pipeline::planner::standard::openai_cli::LocalOpenAiCliSpec;
-pub(super) struct LocalOpenAiCliSpec {
-    pub(super) api_format: &'static str,
-    pub(super) decision_kind: &'static str,
-    pub(super) report_kind: &'static str,
-    pub(super) compact: bool,
-    pub(super) require_streaming: bool,
-}
 
 #[derive(Debug, Clone)]
 pub(super) struct LocalOpenAiCliDecisionInput {
-    pub(super) auth_context: crate::control::GatewayControlAuthContext,
+    pub(super) auth_context: GatewayControlAuthContext,
     pub(super) requested_model: String,
-    pub(super) auth_snapshot: crate::data::auth::GatewayAuthApiKeySnapshot,
+    pub(super) auth_snapshot: GatewayAuthApiKeySnapshot,
 }
 
 #[derive(Debug, Clone)]
 pub(super) struct LocalOpenAiCliCandidateAttempt {
-    pub(super) candidate: GatewayMinimalCandidateSelectionCandidate,
+    pub(super) candidate: SchedulerMinimalCandidateSelectionCandidate,
     pub(super) candidate_index: u32,
     pub(super) candidate_id: String,
 }
@@ -90,8 +94,8 @@ pub(super) async fn resolve_local_openai_cli_decision_input(
         .filter(|value| !value.is_empty())
         .map(ToOwned::to_owned)?;
 
-    let auth_snapshot = match state
+    let auth_snapshot = match read_auth_api_key_snapshot(
-        .read_auth_api_key_snapshot(
+        state,
         &auth_context.user_id,
         &auth_context.api_key_id,
         current_unix_secs(),
@@ -201,47 +205,19 @@ pub(super) async fn materialize_local_openai_cli_candidate_attempts(
             candidate.endpoint_api_format.as_str(),
         );
 
-        let candidate_id = match state
+        let candidate_id = persist_available_local_candidate(
-            .upsert_request_candidate(UpsertRequestCandidateRecord {
+            state,
-                id: generated_candidate_id.clone(),
+            trace_id,
-                request_id: trace_id.to_string(),
+            &input.auth_context.user_id,
-                user_id: Some(input.auth_context.user_id.clone()),
+            &input.auth_context.api_key_id,
-                api_key_id: Some(input.auth_context.api_key_id.clone()),
+            &candidate,
-                username: None,
+            candidate_index as u32,
-                api_key_name: None,
+            &generated_candidate_id,
-                candidate_index: candidate_index as u32,
+            Some(extra_data),
-                retry_index: 0,
+            created_at_unix_secs,
-                provider_id: Some(candidate.provider_id.clone()),
+            "gateway local openai cli decision request candidate upsert failed",
-                endpoint_id: Some(candidate.endpoint_id.clone()),
+        )
-                key_id: Some(candidate.key_id.clone()),
+        .await;
-                status: RequestCandidateStatus::Available,
-                skip_reason: None,
-                is_cached: Some(false),
-                status_code: None,
-                error_type: None,
-                error_message: None,
-                latency_ms: None,
-                concurrent_requests: None,
-                extra_data: Some(extra_data),
-                required_capabilities: candidate.key_capabilities.clone(),
-                created_at_unix_secs: Some(created_at_unix_secs),
-                started_at_unix_secs: None,
-                finished_at_unix_secs: None,
-            })
-            .await
-        {
-            Ok(Some(stored)) => stored.id,
-            Ok(None) => generated_candidate_id.clone(),
-            Err(err) => {
-                warn!(
-                    trace_id = %trace_id,
-                    api_format = spec.api_format,
-                    error = ?err,
-                    "gateway local openai cli decision request candidate upsert failed"
-                );
-                generated_candidate_id.clone()
-            }
-        };
 
         attempts.push(LocalOpenAiCliCandidateAttempt {
             candidate,
@@ -254,9 +230,9 @@ pub(super) async fn materialize_local_openai_cli_candidate_attempts(
 }
 
 fn auth_snapshot_allows_cross_format_openai_cli_candidate(
-    auth_snapshot: &crate::data::auth::GatewayAuthApiKeySnapshot,
+    auth_snapshot: &GatewayAuthApiKeySnapshot,
     requested_model: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
 ) -> bool {
     if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
         let provider_allowed = allowed_providers.iter().any(|value| {
@@ -300,8 +276,8 @@ pub(super) async fn maybe_build_local_openai_cli_decision_payload_for_candidate(
     } = attempt;
     let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
 
-    let transport = match state
+    let transport = match read_provider_transport_snapshot(
-        .read_provider_transport_snapshot(
+        state,
         &candidate.provider_id,
         &candidate.endpoint_id,
         &candidate.key_id,
@@ -385,7 +361,7 @@ pub(super) async fn maybe_build_local_openai_cli_decision_payload_for_candidate(
         conversion_kind.and_then(|kind| request_conversion_direct_auth(&transport, kind))
     };
     let oauth_auth = if resolved_auth.is_none() {
-        match state.resolve_local_oauth_request_auth(&transport).await {
+        match resolve_local_oauth_request_auth(state, &transport).await {
             Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
             Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
             Ok(None) => None,
@@ -434,21 +410,32 @@ pub(super) async fn maybe_build_local_openai_cli_decision_payload_for_candidate(
     }
 
     let needs_bidirectional_conversion = !same_format && conversion_kind.is_some();
+    let upstream_is_stream = spec.require_streaming
+        || is_antigravity
+        || force_upstream_streaming_for_provider(
+            transport.provider.provider_type.as_str(),
+            provider_api_format.as_str(),
+        );
     let Some(base_provider_request_body) = (if needs_bidirectional_conversion {
         build_cross_format_openai_cli_request_body(
             body_json,
             &mapped_model,
             spec.api_format,
             provider_api_format.as_str(),
-            spec.require_streaming,
+            upstream_is_stream,
+            transport.provider.provider_type.as_str(),
             transport.endpoint.body_rules.as_ref(),
+            Some(input.auth_context.api_key_id.as_str()),
         )
     } else {
         build_local_openai_cli_request_body(
             body_json,
             &mapped_model,
-            spec.require_streaming,
+            upstream_is_stream,
+            transport.provider.provider_type.as_str(),
+            provider_api_format.as_str(),
             transport.endpoint.body_rules.as_ref(),
+            Some(input.auth_context.api_key_id.as_str()),
         )
     }) else {
         mark_skipped_local_openai_cli_candidate(
@@ -513,7 +500,6 @@ pub(super) async fn maybe_build_local_openai_cli_decision_payload_for_candidate(
     } else {
         base_provider_request_body
     };
-    let upstream_is_stream = spec.require_streaming || is_antigravity;
 
     let Some(upstream_url) = (if needs_bidirectional_conversion {
         build_cross_format_openai_cli_upstream_url(
@@ -572,6 +558,15 @@ pub(super) async fn maybe_build_local_openai_cli_decision_payload_for_candidate(
         .await;
         return None;
     }
+    apply_codex_openai_cli_special_headers(
+        &mut provider_request_headers,
+        &provider_request_body,
+        &parts.headers,
+        transport.provider.provider_type.as_str(),
+        provider_api_format.as_str(),
+        Some(trace_id),
+        transport.key.decrypted_auth_config.as_deref(),
+    );
     ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
     if upstream_is_stream {
         provider_request_headers
@@ -685,70 +680,38 @@ async fn mark_skipped_local_openai_cli_candidate(
     state: &AppState,
     input: &LocalOpenAiCliDecisionInput,
     trace_id: &str,
-    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate: &SchedulerMinimalCandidateSelectionCandidate,
     candidate_index: u32,
     candidate_id: &str,
     skip_reason: &'static str,
 ) {
-    if let Err(err) = state
+    persist_skipped_local_candidate(
-        .upsert_request_candidate(UpsertRequestCandidateRecord {
+        state,
-            id: candidate_id.to_string(),
+        trace_id,
-            request_id: trace_id.to_string(),
+        &input.auth_context.user_id,
-            user_id: Some(input.auth_context.user_id.clone()),
+        &input.auth_context.api_key_id,
-            api_key_id: Some(input.auth_context.api_key_id.clone()),
+        candidate,
-            username: None,
-            api_key_name: None,
         candidate_index,
-            retry_index: 0,
+        candidate_id,
-            provider_id: Some(candidate.provider_id.clone()),
-            endpoint_id: Some(candidate.endpoint_id.clone()),
-            key_id: Some(candidate.key_id.clone()),
-            status: RequestCandidateStatus::Skipped,
-            skip_reason: Some(skip_reason.to_string()),
-            is_cached: Some(false),
-            status_code: None,
-            error_type: None,
-            error_message: None,
-            latency_ms: None,
-            concurrent_requests: None,
-            extra_data: None,
-            required_capabilities: candidate.key_capabilities.clone(),
-            created_at_unix_secs: None,
-            started_at_unix_secs: None,
-            finished_at_unix_secs: Some(current_unix_secs()),
-        })
-        .await
-    {
-        warn!(
-            trace_id = %trace_id,
-            candidate_id = %candidate_id,
         skip_reason,
-            error = ?err,
+        current_unix_secs(),
-            "gateway local openai cli decision failed to persist skipped candidate"
+        "gateway local openai cli decision failed to persist skipped candidate",
-        );
+    )
-    }
+    .await;
 }
 
 pub(super) async fn mark_unused_local_openai_cli_candidates<T>(state: &AppState, remaining: Vec<T>)
 where
     T: LocalOpenAiCliPlanAndReport,
 {
-    for plan_and_report in remaining {
+    mark_unused_local_candidate_items(
-        record_local_request_candidate_status(
         state,
-            plan_and_report.plan(),
+        remaining,
-            plan_and_report.report_context(),
+        |item| item.plan(),
-            RequestCandidateStatus::Unused,
+        |item| item.report_context(),
-            None,
-            None,
-            None,
-            None,
-            None,
-            None,
     )
     .await;
 }
-}
 
 pub(super) trait LocalOpenAiCliPlanAndReport {
     fn plan(&self) -> &aether_contracts::ExecutionPlan;

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

@@ -1,7 +1,7 @@
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
 use crate::ai_pipeline::planner::plan_builders::{
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlDecision;
 use crate::{AppState, GatewayControlSyncDecisionResponse, GatewayError};
 
 mod decision;

apps/aether-gateway/src/ai_pipeline/planner/standard/openai/cli/plans.rs

@@ -5,56 +5,15 @@ use super::decision::{
     maybe_build_local_openai_cli_decision_payload_for_candidate,
     resolve_local_openai_cli_decision_input, LocalOpenAiCliSpec,
 };
-use crate::ai_pipeline::planner::common::{
+use crate::ai_pipeline::control_facade::GatewayControlDecision;
-    OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
-    OPENAI_COMPACT_SYNC_PLAN_KIND,
-};
 use crate::ai_pipeline::planner::plan_builders::{
     build_openai_cli_stream_plan_from_decision, build_openai_cli_sync_plan_from_decision,
     LocalStreamPlanAndReport, LocalSyncPlanAndReport,
 };
-use crate::control::GatewayControlDecision;
 use crate::{AppState, GatewayError};
-
+pub(super) use aether_ai_pipeline::planner::standard::openai_cli::{
-pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalOpenAiCliSpec> {
+    resolve_stream_spec, resolve_sync_spec,
-    match plan_kind {
+};
-        OPENAI_CLI_SYNC_PLAN_KIND => Some(LocalOpenAiCliSpec {
-            api_format: "openai:cli",
-            decision_kind: OPENAI_CLI_SYNC_PLAN_KIND,
-            report_kind: "openai_cli_sync_success",
-            compact: false,
-            require_streaming: false,
-        }),
-        OPENAI_COMPACT_SYNC_PLAN_KIND => Some(LocalOpenAiCliSpec {
-            api_format: "openai:compact",
-            decision_kind: OPENAI_COMPACT_SYNC_PLAN_KIND,
-            report_kind: "openai_cli_sync_success",
-            compact: true,
-            require_streaming: false,
-        }),
-        _ => None,
-    }
-}
-
-pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalOpenAiCliSpec> {
-    match plan_kind {
-        OPENAI_CLI_STREAM_PLAN_KIND => Some(LocalOpenAiCliSpec {
-            api_format: "openai:cli",
-            decision_kind: OPENAI_CLI_STREAM_PLAN_KIND,
-            report_kind: "openai_cli_stream_success",
-            compact: false,
-            require_streaming: true,
-        }),
-        OPENAI_COMPACT_STREAM_PLAN_KIND => Some(LocalOpenAiCliSpec {
-            api_format: "openai:compact",
-            decision_kind: OPENAI_COMPACT_STREAM_PLAN_KIND,
-            report_kind: "openai_cli_stream_success",
-            compact: true,
-            require_streaming: true,
-        }),
-        _ => None,
-    }
-}
 
 pub(super) async fn build_local_sync_plan_and_reports(
     state: &AppState,

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

@@ -1,12 +1,14 @@
 pub(crate) mod chat;
 pub(crate) mod cli;
 
-pub(crate) use chat::{
+pub(crate) use aether_ai_pipeline::planner::openai::{
     copy_request_number_field, copy_request_number_field_as,
     map_openai_reasoning_effort_to_claude_output, map_openai_reasoning_effort_to_gemini_budget,
-    maybe_build_stream_local_decision_payload, maybe_build_sync_local_decision_payload,
     parse_openai_stop_sequences, resolve_openai_chat_max_tokens, value_as_u64,
 };
+pub(crate) use chat::{
+    maybe_build_stream_local_decision_payload, maybe_build_sync_local_decision_payload,
+};
 pub(crate) use cli::{
     maybe_build_stream_local_openai_cli_decision_payload,
     maybe_build_sync_local_openai_cli_decision_payload,

apps/aether-gateway/src/ai_pipeline/planner/standard/openai/plan_builders.rs

@@ -6,10 +6,12 @@ use super::{
     LocalSyncPlanAndReport,
 };
 use crate::ai_pipeline::adaptation::surfaces::provider_adaptation_requires_eventstream_accept;
-use crate::provider_transport::auth::{
+use crate::ai_pipeline::provider_transport_facade::auth::{
     build_openai_passthrough_headers, ensure_upstream_auth_header,
 };
-use crate::provider_transport::url::{build_openai_chat_url, build_openai_cli_url};
+use crate::ai_pipeline::provider_transport_facade::url::{
+    build_openai_chat_url, build_openai_cli_url,
+};
 
 pub(crate) fn build_openai_chat_sync_plan_from_decision(
     parts: &http::request::Parts,

apps/aether-gateway/src/ai_pipeline/planner/standard/plan_builders.rs

@@ -3,7 +3,7 @@ use aether_contracts::{ExecutionPlan, RequestBody};
 use super::{augment_sync_report_context, LocalStreamPlanAndReport, LocalSyncPlanAndReport};
 use crate::ai_pipeline::adaptation::surfaces::provider_adaptation_requires_eventstream_accept;
 use crate::ai_pipeline::contracts::generic_decision_missing_exact_provider_request;
-use crate::provider_transport::ensure_upstream_auth_header;
+use crate::ai_pipeline::provider_transport_facade::ensure_upstream_auth_header;
 use crate::{GatewayControlSyncDecisionResponse, GatewayError};
 
 pub(crate) fn build_standard_sync_plan_from_decision(

apps/aether-gateway/src/ai_pipeline/planner/transport_facade.rs

@@ -0,0 +1,22 @@
+pub(crate) use crate::ai_pipeline::provider_transport_facade::{
+    GatewayProviderTransportSnapshot, LocalResolvedOAuthRequestAuth,
+};
+use crate::{AppState, GatewayError};
+
+pub(crate) async fn read_provider_transport_snapshot(
+    state: &AppState,
+    provider_id: &str,
+    endpoint_id: &str,
+    key_id: &str,
+) -> Result<Option<GatewayProviderTransportSnapshot>, GatewayError> {
+    state
+        .read_provider_transport_snapshot(provider_id, endpoint_id, key_id)
+        .await
+}
+
+pub(crate) async fn resolve_local_oauth_request_auth(
+    state: &AppState,
+    transport: &GatewayProviderTransportSnapshot,
+) -> Result<Option<LocalResolvedOAuthRequestAuth>, GatewayError> {
+    state.resolve_local_oauth_request_auth(transport).await
+}

apps/aether-gateway/src/ai_pipeline/provider_transport_facade.rs

@@ -0,0 +1,54 @@
+pub(crate) mod antigravity {
+    pub(crate) use crate::provider_transport::antigravity::*;
+}
+
+pub(crate) mod auth {
+    pub(crate) use crate::provider_transport::auth::*;
+}
+
+pub(crate) mod claude_code {
+    pub(crate) use crate::provider_transport::claude_code::*;
+}
+
+pub(crate) mod kiro {
+    pub(crate) use crate::provider_transport::kiro::*;
+}
+
+pub(crate) mod oauth_refresh {
+    pub(crate) use crate::provider_transport::oauth_refresh::*;
+}
+
+pub(crate) mod policy {
+    pub(crate) use crate::provider_transport::policy::*;
+}
+
+pub(crate) mod provider_types {
+    pub(crate) use crate::provider_transport::provider_types::*;
+}
+
+pub(crate) mod rules {
+    pub(crate) use crate::provider_transport::rules::*;
+}
+
+pub(crate) mod snapshot {
+    pub(crate) use crate::provider_transport::snapshot::*;
+}
+
+pub(crate) mod url {
+    pub(crate) use crate::provider_transport::url::*;
+}
+
+pub(crate) mod vertex {
+    pub(crate) use crate::provider_transport::vertex::*;
+}
+
+pub(crate) use crate::provider_transport::{
+    apply_local_body_rules, apply_local_header_rules, body_rules_handle_path,
+    build_passthrough_headers, ensure_upstream_auth_header, resolve_transport_execution_timeouts,
+    resolve_transport_proxy_snapshot, resolve_transport_proxy_snapshot_with_tunnel_affinity,
+    resolve_transport_tls_profile, should_skip_upstream_passthrough_header,
+    supports_local_gemini_transport_with_network,
+    supports_local_generic_oauth_request_auth_resolution,
+    supports_local_oauth_request_auth_resolution, GatewayProviderTransportSnapshot,
+    LocalResolvedOAuthRequestAuth,
+};

apps/aether-gateway/src/ai_pipeline/runtime/adapters/antigravity/auth.rs

@@ -2,7 +2,7 @@ use std::collections::BTreeMap;
 
 use serde_json::Value;
 
-use crate::provider_transport::snapshot::GatewayProviderTransportSnapshot;
+use crate::ai_pipeline::provider_transport_facade::snapshot::GatewayProviderTransportSnapshot;
 
 pub(crate) const ANTIGRAVITY_PROVIDER_TYPE: &str = "antigravity";
 pub(crate) const ANTIGRAVITY_REQUEST_USER_AGENT: &str = "antigravity";

apps/aether-gateway/src/ai_pipeline/runtime/adapters/antigravity/mod.rs

@@ -1,6 +1,6 @@
 #![allow(dead_code, unused_imports)]
 
-pub(crate) use crate::provider_transport::antigravity::{
+pub(crate) use crate::ai_pipeline::provider_transport_facade::antigravity::{
     build_antigravity_safe_v1internal_request, build_antigravity_static_identity_headers,
     build_antigravity_v1internal_url, classify_antigravity_safe_request_body,
     classify_local_antigravity_request_support, AntigravityEnvelopeRequestType,

apps/aether-gateway/src/ai_pipeline/runtime/adapters/antigravity/policy.rs

@@ -1,14 +1,14 @@
 use serde_json::Value;
 
 use super::auth::{
-    resolve_local_antigravity_request_auth, AntigravityRequestAuth, AntigravityRequestAuthSupport,
+    ANTIGRAVITY_PROVIDER_TYPE, AntigravityRequestAuth, AntigravityRequestAuthSupport,
-    AntigravityRequestAuthUnsupportedReason, ANTIGRAVITY_PROVIDER_TYPE,
+    AntigravityRequestAuthUnsupportedReason, resolve_local_antigravity_request_auth,
 };
 use super::request::{
-    classify_antigravity_safe_request_body, AntigravityEnvelopeRequestType,
+    AntigravityEnvelopeRequestType, AntigravityRequestEnvelopeUnsupportedReason,
-    AntigravityRequestEnvelopeUnsupportedReason,
+    classify_antigravity_safe_request_body,
 };
-use crate::provider_transport::snapshot::GatewayProviderTransportSnapshot;
+use crate::ai_pipeline::provider_transport_facade::snapshot::GatewayProviderTransportSnapshot;
 
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub(crate) struct AntigravityRequestSideSpec {

apps/aether-gateway/src/ai_pipeline/runtime/adapters/claude/auth.rs

@@ -1 +1 @@
-pub(crate) use crate::provider_transport::auth::resolve_local_standard_auth;
+pub(crate) use crate::ai_pipeline::provider_transport_facade::auth::resolve_local_standard_auth;

apps/aether-gateway/src/ai_pipeline/runtime/adapters/claude/policy.rs

@@ -1 +1 @@
-pub(crate) use crate::provider_transport::policy::supports_local_standard_transport_with_network;
+pub(crate) use crate::ai_pipeline::provider_transport_facade::policy::supports_local_standard_transport_with_network;