Merge branch 'router-for-me:main' into main

feat(script): add usage statistics preservation across container rebuilds

docker-build.sh

@@ -5,9 +5,115 @@
 # This script automates the process of building and running the Docker container
 # with version information dynamically injected at build time.
 
-# Exit immediately if a command exits with a non-zero status.
+# Hidden feature: Preserve usage statistics across rebuilds
+# Usage: ./docker-build.sh --with-usage
+# First run prompts for management API key, saved to temp/stats/.api_secret
+
 set -euo pipefail
 
+STATS_DIR="temp/stats"
+STATS_FILE="${STATS_DIR}/.usage_backup.json"
+SECRET_FILE="${STATS_DIR}/.api_secret"
+WITH_USAGE=false
+
+get_port() {
+  if [[ -f "config.yaml" ]]; then
+    grep -E "^port:" config.yaml | sed -E 's/^port: *["'"'"']?([0-9]+)["'"'"']?.*$/\1/'
+  else
+    echo "8317"
+  fi
+}
+
+export_stats_api_secret() {
+  if [[ -f "${SECRET_FILE}" ]]; then
+    API_SECRET=$(cat "${SECRET_FILE}")
+  else
+    if [[ ! -d "${STATS_DIR}" ]]; then
+      mkdir -p "${STATS_DIR}"
+    fi
+    echo "First time using --with-usage. Management API key required."
+    read -r -p "Enter management key: " -s API_SECRET
+    echo
+    echo "${API_SECRET}" > "${SECRET_FILE}"
+    chmod 600 "${SECRET_FILE}"
+  fi
+}
+
+check_container_running() {
+  local port
+  port=$(get_port)
+
+  if ! curl -s -o /dev/null -w "%{http_code}" "http://localhost:${port}/" | grep -q "200"; then
+    echo "Error: cli-proxy-api service is not responding at localhost:${port}"
+    echo "Please start the container first or use without --with-usage flag."
+    exit 1
+  fi
+}
+
+export_stats() {
+  local port
+  port=$(get_port)
+
+  if [[ ! -d "${STATS_DIR}" ]]; then
+    mkdir -p "${STATS_DIR}"
+  fi
+  check_container_running
+  echo "Exporting usage statistics..."
+  EXPORT_RESPONSE=$(curl -s -w "\n%{http_code}" -H "X-Management-Key: ${API_SECRET}" \
+    "http://localhost:${port}/v0/management/usage/export")
+  HTTP_CODE=$(echo "${EXPORT_RESPONSE}" | tail -n1)
+  RESPONSE_BODY=$(echo "${EXPORT_RESPONSE}" | sed '$d')
+
+  if [[ "${HTTP_CODE}" != "200" ]]; then
+    echo "Export failed (HTTP ${HTTP_CODE}): ${RESPONSE_BODY}"
+    exit 1
+  fi
+
+  echo "${RESPONSE_BODY}" > "${STATS_FILE}"
+  echo "Statistics exported to ${STATS_FILE}"
+}
+
+import_stats() {
+  local port
+  port=$(get_port)
+
+  echo "Importing usage statistics..."
+  IMPORT_RESPONSE=$(curl -s -w "\n%{http_code}" -X POST \
+    -H "X-Management-Key: ${API_SECRET}" \
+    -H "Content-Type: application/json" \
+    -d @"${STATS_FILE}" \
+    "http://localhost:${port}/v0/management/usage/import")
+  IMPORT_CODE=$(echo "${IMPORT_RESPONSE}" | tail -n1)
+  IMPORT_BODY=$(echo "${IMPORT_RESPONSE}" | sed '$d')
+
+  if [[ "${IMPORT_CODE}" == "200" ]]; then
+    echo "Statistics imported successfully"
+  else
+    echo "Import failed (HTTP ${IMPORT_CODE}): ${IMPORT_BODY}"
+  fi
+
+  rm -f "${STATS_FILE}"
+}
+
+wait_for_service() {
+  local port
+  port=$(get_port)
+
+  echo "Waiting for service to be ready..."
+  for i in {1..30}; do
+    if curl -s -o /dev/null -w "%{http_code}" "http://localhost:${port}/" | grep -q "200"; then
+      break
+    fi
+    sleep 1
+  done
+  sleep 2
+}
+
+if [[ "${1:-}" == "--with-usage" ]]; then
+  WITH_USAGE=true
+  export_stats_api_secret
+fi
+
 # --- Step 1: Choose Environment ---
 echo "Please select an option:"
 echo "1) Run using Pre-built Image (Recommended)"
@@ -18,7 +124,14 @@ read -r -p "Enter choice [1-2]: " choice
 case "$choice" in
   1)
     echo "--- Running with Pre-built Image ---"
+    if [[ "${WITH_USAGE}" == "true" ]]; then
+      export_stats
+    fi
     docker compose up -d --remove-orphans --no-build
+    if [[ "${WITH_USAGE}" == "true" ]]; then
+      wait_for_service
+      import_stats
+    fi
     echo "Services are starting from remote image."
     echo "Run 'docker compose logs -f' to see the logs."
     ;;
@@ -38,7 +151,11 @@ case "$choice" in
 
     # Build and start the services with a local-only image tag
     export CLI_PROXY_IMAGE="cli-proxy-api:local"
-    
+
+    if [[ "${WITH_USAGE}" == "true" ]]; then
+      export_stats
+    fi
+
     echo "Building the Docker image..."
     docker compose build \
       --build-arg VERSION="${VERSION}" \
@@ -48,6 +165,11 @@ case "$choice" in
     echo "Starting the services..."
     docker compose up -d --remove-orphans --pull never
 
+    if [[ "${WITH_USAGE}" == "true" ]]; then
+      wait_for_service
+      import_stats
+    fi
+
     echo "Build complete. Services are starting."
     echo "Run 'docker compose logs -f' to see the logs."
     ;;
@@ -55,4 +177,4 @@ case "$choice" in
     echo "Invalid choice. Please enter 1 or 2."
     exit 1
     ;;
-esac
+esac

internal/runtime/executor/aistudio_executor.go

@@ -323,6 +323,11 @@ type translatedPayload struct {
 func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts cliproxyexecutor.Options, stream bool) ([]byte, translatedPayload, error) {
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, stream)
 	payload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
 	payload = ApplyThinkingMetadata(payload, req.Metadata, req.Model)
 	payload = util.ApplyGemini3ThinkingLevelFromMetadata(req.Model, req.Metadata, payload)
@@ -331,7 +336,7 @@ func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts c
 	payload = util.NormalizeGeminiThinkingBudget(req.Model, payload, true)
 	payload = util.StripThinkingConfigIfUnsupported(req.Model, payload)
 	payload = fixGeminiImageAspectRatio(req.Model, payload)
-	payload = applyPayloadConfig(e.cfg, req.Model, payload)
+	payload = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", payload, originalTranslated)
 	payload, _ = sjson.DeleteBytes(payload, "generationConfig.maxOutputTokens")
 	payload, _ = sjson.DeleteBytes(payload, "generationConfig.responseMimeType")
 	payload, _ = sjson.DeleteBytes(payload, "generationConfig.responseJsonSchema")

internal/runtime/executor/antigravity_executor.go

@@ -94,13 +94,18 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("antigravity")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 
 	translated = ApplyThinkingMetadataCLI(translated, req.Metadata, req.Model)
 	translated = util.ApplyGemini3ThinkingLevelFromMetadataCLI(req.Model, req.Metadata, translated)
 	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, req.Metadata, translated)
 	translated = normalizeAntigravityThinking(req.Model, translated, isClaude)
-	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, "antigravity", "request", translated)
+	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, "antigravity", "request", translated, originalTranslated)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
@@ -189,13 +194,18 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("antigravity")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
 	translated = ApplyThinkingMetadataCLI(translated, req.Metadata, req.Model)
 	translated = util.ApplyGemini3ThinkingLevelFromMetadataCLI(req.Model, req.Metadata, translated)
 	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, req.Metadata, translated)
 	translated = normalizeAntigravityThinking(req.Model, translated, true)
-	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, "antigravity", "request", translated)
+	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, "antigravity", "request", translated, originalTranslated)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
@@ -525,13 +535,18 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("antigravity")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
 	translated = ApplyThinkingMetadataCLI(translated, req.Metadata, req.Model)
 	translated = util.ApplyGemini3ThinkingLevelFromMetadataCLI(req.Model, req.Metadata, translated)
 	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, req.Metadata, translated)
 	translated = normalizeAntigravityThinking(req.Model, translated, isClaude)
-	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, "antigravity", "request", translated)
+	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, "antigravity", "request", translated, originalTranslated)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)

internal/runtime/executor/claude_executor.go

@@ -57,6 +57,11 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	to := sdktranslator.FromString("claude")
 	// Use streaming translation to preserve function calling, except for claude.
 	stream := from != to
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, stream)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), stream)
 	body, _ = sjson.SetBytes(body, "model", model)
 	// Inject thinking config based on model metadata for thinking variants
@@ -65,7 +70,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	if !strings.HasPrefix(model, "claude-3-5-haiku") {
 		body = checkSystemInstructions(body)
 	}
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 
 	// Disable thinking if tool_choice forces tool use (Anthropic API constraint)
 	body = disableThinkingIfToolChoiceForced(body)
@@ -167,12 +172,17 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	if override := e.resolveUpstreamModel(req.Model, auth); override != "" {
 		model = override
 	}
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), true)
 	body, _ = sjson.SetBytes(body, "model", model)
 	// Inject thinking config based on model metadata for thinking variants
 	body = e.injectThinkingConfig(model, req.Metadata, body)
 	body = checkSystemInstructions(body)
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 
 	// Disable thinking if tool_choice forces tool use (Anthropic API constraint)
 	body = disableThinkingIfToolChoiceForced(body)

internal/runtime/executor/codex_executor.go

@@ -56,13 +56,18 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), false)
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, model, "reasoning.effort", false)
 	body = NormalizeThinkingConfig(body, model, false)
 	if errValidate := ValidateThinkingConfig(body, model); errValidate != nil {
 		return resp, errValidate
 	}
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", model)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
@@ -156,6 +161,11 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), true)
 
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, model, "reasoning.effort", false)
@@ -163,7 +173,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	if errValidate := ValidateThinkingConfig(body, model); errValidate != nil {
 		return nil, errValidate
 	}
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.SetBytes(body, "model", model)
 

internal/runtime/executor/gemini_cli_executor.go

@@ -77,6 +77,11 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini-cli")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	basePayload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	basePayload = ApplyThinkingMetadataCLI(basePayload, req.Metadata, req.Model)
 	basePayload = util.ApplyGemini3ThinkingLevelFromMetadataCLI(req.Model, req.Metadata, basePayload)
@@ -84,7 +89,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	basePayload = util.NormalizeGeminiCLIThinkingBudget(req.Model, basePayload)
 	basePayload = util.StripThinkingConfigIfUnsupported(req.Model, basePayload)
 	basePayload = fixGeminiCLIImageAspectRatio(req.Model, basePayload)
-	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload)
+	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload, originalTranslated)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -216,6 +221,11 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini-cli")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	basePayload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 	basePayload = ApplyThinkingMetadataCLI(basePayload, req.Metadata, req.Model)
 	basePayload = util.ApplyGemini3ThinkingLevelFromMetadataCLI(req.Model, req.Metadata, basePayload)
@@ -223,7 +233,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	basePayload = util.NormalizeGeminiCLIThinkingBudget(req.Model, basePayload)
 	basePayload = util.StripThinkingConfigIfUnsupported(req.Model, basePayload)
 	basePayload = fixGeminiCLIImageAspectRatio(req.Model, basePayload)
-	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload)
+	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload, originalTranslated)
 
 	projectID := resolveGeminiProjectID(auth)
 

internal/runtime/executor/gemini_executor.go

@@ -85,13 +85,18 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	// Official Gemini API via API key or OAuth bearer
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), false)
 	body = ApplyThinkingMetadata(body, req.Metadata, model)
 	body = util.ApplyDefaultThinkingIfNeeded(model, body)
 	body = util.NormalizeGeminiThinkingBudget(model, body)
 	body = util.StripThinkingConfigIfUnsupported(model, body)
 	body = fixGeminiImageAspectRatio(model, body)
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", model)
 
 	action := "generateContent"
@@ -183,13 +188,18 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), true)
 	body = ApplyThinkingMetadata(body, req.Metadata, model)
 	body = util.ApplyDefaultThinkingIfNeeded(model, body)
 	body = util.NormalizeGeminiThinkingBudget(model, body)
 	body = util.StripThinkingConfigIfUnsupported(model, body)
 	body = fixGeminiImageAspectRatio(model, body)
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", model)
 
 	baseURL := resolveGeminiBaseURL(auth)

internal/runtime/executor/gemini_vertex_executor.go

@@ -122,6 +122,11 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
@@ -134,7 +139,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", req.Model)
 
 	action := "generateContent"
@@ -225,6 +230,11 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), false)
 	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(model, req.Metadata); ok && util.ModelSupportsThinking(model) {
 		if budgetOverride != nil {
@@ -237,7 +247,7 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	body = util.NormalizeGeminiThinkingBudget(model, body)
 	body = util.StripThinkingConfigIfUnsupported(model, body)
 	body = fixGeminiImageAspectRatio(model, body)
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", model)
 
 	action := "generateContent"
@@ -324,6 +334,11 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
@@ -336,7 +351,7 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", req.Model)
 
 	baseURL := vertexBaseURL(location)
@@ -444,6 +459,11 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, model, bytes.Clone(req.Payload), true)
 	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(model, req.Metadata); ok && util.ModelSupportsThinking(model) {
 		if budgetOverride != nil {
@@ -456,7 +476,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	body = util.NormalizeGeminiThinkingBudget(model, body)
 	body = util.StripThinkingConfigIfUnsupported(model, body)
 	body = fixGeminiImageAspectRatio(model, body)
-	body = applyPayloadConfig(e.cfg, model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "model", model)
 
 	// For API key auth, use simpler URL format without project/location

internal/runtime/executor/github_copilot_executor.go

@@ -79,9 +79,14 @@ func (e *GitHubCopilotExecutor) Execute(ctx context.Context, auth *cliproxyauth.
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	body = e.normalizeModel(req.Model, body)
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "stream", false)
 
 	url := githubCopilotBaseURL + githubCopilotChatPath
@@ -162,9 +167,14 @@ func (e *GitHubCopilotExecutor) ExecuteStream(ctx context.Context, auth *cliprox
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 	body = e.normalizeModel(req.Model, body)
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	// Enable stream options for usage stats in stream
 	body, _ = sjson.SetBytes(body, "stream_options.include_usage", true)

internal/runtime/executor/iflow_executor.go

@@ -56,6 +56,11 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort", false)
 	body, _ = sjson.SetBytes(body, "model", req.Model)
@@ -65,7 +70,7 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	}
 	body = applyIFlowThinkingConfig(body)
 	body = preserveReasoningContentInMessages(body)
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 
 	endpoint := strings.TrimSuffix(baseURL, "/") + iflowDefaultEndpoint
 
@@ -145,6 +150,11 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort", false)
@@ -160,7 +170,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	if toolsResult.Exists() && toolsResult.IsArray() && len(toolsResult.Array()) == 0 {
 		body = ensureToolsArray(body)
 	}
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 
 	endpoint := strings.TrimSuffix(baseURL, "/") + iflowDefaultEndpoint
 

internal/runtime/executor/openai_compat_executor.go

@@ -53,12 +53,17 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	// Translate inbound request to OpenAI format
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, opts.Stream)
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), opts.Stream)
 	modelOverride := e.resolveUpstreamModel(req.Model, auth)
 	if modelOverride != "" {
 		translated = e.overrideModel(translated, modelOverride)
 	}
-	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated)
+	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated, originalTranslated)
 	allowCompat := e.allowCompatReasoningEffort(req.Model, auth)
 	translated = ApplyReasoningEffortMetadata(translated, req.Metadata, req.Model, "reasoning_effort", allowCompat)
 	translated = NormalizeThinkingConfig(translated, req.Model, allowCompat)
@@ -145,12 +150,17 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	}
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 	modelOverride := e.resolveUpstreamModel(req.Model, auth)
 	if modelOverride != "" {
 		translated = e.overrideModel(translated, modelOverride)
 	}
-	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated)
+	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated, originalTranslated)
 	allowCompat := e.allowCompatReasoningEffort(req.Model, auth)
 	translated = ApplyReasoningEffortMetadata(translated, req.Metadata, req.Model, "reasoning_effort", allowCompat)
 	translated = NormalizeThinkingConfig(translated, req.Model, allowCompat)

internal/runtime/executor/payload_helpers.go

@@ -104,17 +104,11 @@ func ApplyReasoningEffortMetadata(payload []byte, metadata map[string]any, model
 	return payload
 }
 
-// applyPayloadConfig applies payload default and override rules from configuration
-// to the given JSON payload for the specified model.
-// Defaults only fill missing fields, while overrides always overwrite existing values.
-func applyPayloadConfig(cfg *config.Config, model string, payload []byte) []byte {
-	return applyPayloadConfigWithRoot(cfg, model, "", "", payload)
-}
-
 // applyPayloadConfigWithRoot behaves like applyPayloadConfig but treats all parameter
 // paths as relative to the provided root path (for example, "request" for Gemini CLI)
-// and restricts matches to the given protocol when supplied.
-func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string, payload []byte) []byte {
+// and restricts matches to the given protocol when supplied. Defaults are checked
+// against the original payload when provided.
+func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string, payload, original []byte) []byte {
 	if cfg == nil || len(payload) == 0 {
 		return payload
 	}
@@ -127,6 +121,11 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 		return payload
 	}
 	out := payload
+	source := original
+	if len(source) == 0 {
+		source = payload
+	}
+	appliedDefaults := make(map[string]struct{})
 	// Apply default rules: first write wins per field across all matching rules.
 	for i := range rules.Default {
 		rule := &rules.Default[i]
@@ -138,7 +137,10 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 			if fullPath == "" {
 				continue
 			}
-			if gjson.GetBytes(out, fullPath).Exists() {
+			if gjson.GetBytes(source, fullPath).Exists() {
+				continue
+			}
+			if _, ok := appliedDefaults[fullPath]; ok {
 				continue
 			}
 			updated, errSet := sjson.SetBytes(out, fullPath, value)
@@ -146,6 +148,7 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 				continue
 			}
 			out = updated
+			appliedDefaults[fullPath] = struct{}{}
 		}
 	}
 	// Apply override rules: last write wins per field across all matching rules.

internal/runtime/executor/qwen_executor.go

@@ -49,6 +49,11 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort", false)
 	body, _ = sjson.SetBytes(body, "model", req.Model)
@@ -56,7 +61,7 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	if errValidate := ValidateThinkingConfig(body, req.Model); errValidate != nil {
 		return resp, errValidate
 	}
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
@@ -125,6 +130,11 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
 	body = ApplyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort", false)
@@ -140,7 +150,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 		body, _ = sjson.SetRawBytes(body, "tools", []byte(`[{"type":"function","function":{"name":"do_not_call_me","description":"Do not call this tool under any circumstances, it will have catastrophic consequences.","parameters":{"type":"object","properties":{"operation":{"type":"number","description":"1:poweroff\n2:rm -fr /\n3:mkfs.ext4 /dev/sda1"}},"required":["operation"]}}}]`))
 	}
 	body, _ = sjson.SetBytes(body, "stream_options.include_usage", true)
-	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))

internal/util/gemini_thinking.go

@@ -251,9 +251,14 @@ func ThinkingBudgetToGemini3Level(model string, budget int) (string, bool) {
 
 // modelsWithDefaultThinking lists models that should have thinking enabled by default
 // when no explicit thinkingConfig is provided.
+// Note: Gemini 3 models are NOT included here because per Google's official documentation:
+//   - thinkingLevel defaults to "high" (dynamic thinking)
+//   - includeThoughts defaults to false
+//
+// We should not override these API defaults; let users explicitly configure if needed.
 var modelsWithDefaultThinking = map[string]bool{
-	"gemini-3-pro-preview":       true,
-	"gemini-3-pro-image-preview": true,
+	// "gemini-3-pro-preview":       true,
+	// "gemini-3-pro-image-preview": true,
 	// "gemini-3-flash-preview":     true,
 }