Add reasoning effort to usage events

sdk/api/handlers/handlers.go

@@ -231,6 +231,17 @@ func requestExecutionMetadata(ctx context.Context) map[string]any {
 	return meta
 }
 
+func setReasoningEffortMetadata(meta map[string]any, handlerType, model string, rawJSON []byte) {
+	if meta == nil {
+		return
+	}
+	effort := thinking.ExtractReasoningEffort(rawJSON, handlerType, model)
+	if effort == "" {
+		return
+	}
+	meta[coreexecutor.ReasoningEffortMetadataKey] = effort
+}
+
 // headersFromContext extracts the original HTTP request headers from the gin context
 // embedded in the provided context. This allows session affinity selectors to read
 // client headers like X-Amp-Thread-Id.
@@ -550,6 +561,7 @@ func (h *BaseAPIHandler) executeWithAuthManager(ctx context.Context, handlerType
 	}
 	reqMeta := requestExecutionMetadata(ctx)
 	reqMeta[coreexecutor.RequestedModelMetadataKey] = modelName
+	setReasoningEffortMetadata(reqMeta, handlerType, normalizedModel, rawJSON)
 	payload := rawJSON
 	if len(payload) == 0 {
 		payload = nil
@@ -598,6 +610,7 @@ func (h *BaseAPIHandler) ExecuteCountWithAuthManager(ctx context.Context, handle
 	}
 	reqMeta := requestExecutionMetadata(ctx)
 	reqMeta[coreexecutor.RequestedModelMetadataKey] = modelName
+	setReasoningEffortMetadata(reqMeta, handlerType, normalizedModel, rawJSON)
 	payload := rawJSON
 	if len(payload) == 0 {
 		payload = nil
@@ -659,6 +672,7 @@ func (h *BaseAPIHandler) executeStreamWithAuthManager(ctx context.Context, handl
 	}
 	reqMeta := requestExecutionMetadata(ctx)
 	reqMeta[coreexecutor.RequestedModelMetadataKey] = modelName
+	setReasoningEffortMetadata(reqMeta, handlerType, normalizedModel, rawJSON)
 	payload := rawJSON
 	if len(payload) == 0 {
 		payload = nil

sdk/api/handlers/handlers_metadata_test.go

@@ -18,3 +18,23 @@ func TestRequestExecutionMetadataIncludesExecutionSessionWithoutIdempotencyKey(t
 		t.Fatalf("unexpected idempotency key in metadata: %v", meta[idempotencyKeyMetadataKey])
 	}
 }
+
+func TestSetReasoningEffortMetadataUsesSuffixOverBody(t *testing.T) {
+	meta := make(map[string]any)
+
+	setReasoningEffortMetadata(meta, "openai", "gpt-5.4(high)", []byte(`{"reasoning_effort":"low"}`))
+
+	if got := meta[coreexecutor.ReasoningEffortMetadataKey]; got != "high" {
+		t.Fatalf("ReasoningEffortMetadataKey = %v, want %q", got, "high")
+	}
+}
+
+func TestSetReasoningEffortMetadataSupportsOpenAIResponses(t *testing.T) {
+	meta := make(map[string]any)
+
+	setReasoningEffortMetadata(meta, "openai-response", "gpt-5.4", []byte(`{"reasoning":{"effort":"medium"}}`))
+
+	if got := meta[coreexecutor.ReasoningEffortMetadataKey]; got != "medium" {
+		t.Fatalf("ReasoningEffortMetadataKey = %v, want %q", got, "medium")
+	}
+}

sdk/cliproxy/auth/conductor.go

@@ -1632,7 +1632,11 @@ func hasRequestedModelMetadata(meta map[string]any) bool {
 
 func contextWithRequestedModelAlias(ctx context.Context, opts cliproxyexecutor.Options, fallback string) context.Context {
 	alias := requestedModelAliasFromOptions(opts, fallback)
-	return coreusage.WithRequestedModelAlias(ctx, alias)
+	ctx = coreusage.WithRequestedModelAlias(ctx, alias)
+	if effort := reasoningEffortFromOptions(opts); effort != "" {
+		ctx = coreusage.WithReasoningEffort(ctx, effort)
+	}
+	return ctx
 }
 
 func requestedModelAliasFromOptions(opts cliproxyexecutor.Options, fallback string) string {
@@ -1660,6 +1664,24 @@ func requestedModelAliasFromOptions(opts cliproxyexecutor.Options, fallback stri
 	}
 }
 
+func reasoningEffortFromOptions(opts cliproxyexecutor.Options) string {
+	if len(opts.Metadata) == 0 {
+		return ""
+	}
+	raw, ok := opts.Metadata[cliproxyexecutor.ReasoningEffortMetadataKey]
+	if !ok || raw == nil {
+		return ""
+	}
+	switch value := raw.(type) {
+	case string:
+		return strings.TrimSpace(value)
+	case []byte:
+		return strings.TrimSpace(string(value))
+	default:
+		return ""
+	}
+}
+
 func pinnedAuthIDFromMetadata(meta map[string]any) string {
 	if len(meta) == 0 {
 		return ""

sdk/cliproxy/auth/conductor_usage_test.go

@@ -0,0 +1,25 @@
+package auth
+
+import (
+	"context"
+	"testing"
+
+	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/executor"
+	coreusage "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/usage"
+)
+
+func TestContextWithRequestedModelAliasIncludesReasoningEffort(t *testing.T) {
+	ctx := contextWithRequestedModelAlias(context.Background(), cliproxyexecutor.Options{
+		Metadata: map[string]any{
+			cliproxyexecutor.RequestedModelMetadataKey:  "client-model",
+			cliproxyexecutor.ReasoningEffortMetadataKey: "medium",
+		},
+	}, "fallback-model")
+
+	if got := coreusage.RequestedModelAliasFromContext(ctx); got != "client-model" {
+		t.Fatalf("requested model alias = %q, want %q", got, "client-model")
+	}
+	if got := coreusage.ReasoningEffortFromContext(ctx); got != "medium" {
+		t.Fatalf("reasoning effort = %q, want %q", got, "medium")
+	}
+}

sdk/cliproxy/executor/types.go

@@ -17,6 +17,9 @@ const RequestPathMetadataKey = "request_path"
 // DisallowFreeAuthMetadataKey instructs auth selection to skip known free-tier credentials.
 const DisallowFreeAuthMetadataKey = "disallow_free_auth"
 
+// ReasoningEffortMetadataKey stores the client-requested reasoning effort for usage logs.
+const ReasoningEffortMetadataKey = "reasoning_effort"
+
 const (
 	// PinnedAuthMetadataKey locks execution to a specific auth ID.
 	PinnedAuthMetadataKey = "pinned_auth_id"

sdk/cliproxy/usage/manager.go

@@ -12,19 +12,21 @@ import (
 
 // Record contains the usage statistics captured for a single provider request.
 type Record struct {
-	Provider    string
-	Model       string
-	Alias       string
-	APIKey      string
-	AuthID      string
-	AuthIndex   string
-	AuthType    string
-	Source      string
-	RequestedAt time.Time
-	Latency     time.Duration
-	Failed      bool
-	Fail        Failure
-	Detail      Detail
+	Provider  string
+	Model     string
+	Alias     string
+	APIKey    string
+	AuthID    string
+	AuthIndex string
+	AuthType  string
+	Source    string
+	// ReasoningEffort stores the client-requested thinking level for request event logs.
+	ReasoningEffort string
+	RequestedAt     time.Time
+	Latency         time.Duration
+	Failed          bool
+	Fail            Failure
+	Detail          Detail
 	// ResponseHeaders stores a snapshot of upstream response headers for usage sinks.
 	ResponseHeaders http.Header
 }
@@ -47,6 +49,7 @@ type Detail struct {
 }
 
 type requestedModelAliasContextKey struct{}
+type reasoningEffortContextKey struct{}
 
 // WithRequestedModelAlias stores the client-requested model name for usage sinks.
 func WithRequestedModelAlias(ctx context.Context, alias string) context.Context {
@@ -76,6 +79,34 @@ func RequestedModelAliasFromContext(ctx context.Context) string {
 	}
 }
 
+// WithReasoningEffort stores the client-requested reasoning effort for usage sinks.
+func WithReasoningEffort(ctx context.Context, effort string) context.Context {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	effort = strings.TrimSpace(effort)
+	if effort == "" {
+		return ctx
+	}
+	return context.WithValue(ctx, reasoningEffortContextKey{}, effort)
+}
+
+// ReasoningEffortFromContext returns the client-requested reasoning effort stored in ctx.
+func ReasoningEffortFromContext(ctx context.Context) string {
+	if ctx == nil {
+		return ""
+	}
+	raw := ctx.Value(reasoningEffortContextKey{})
+	switch value := raw.(type) {
+	case string:
+		return strings.TrimSpace(value)
+	case []byte:
+		return strings.TrimSpace(string(value))
+	default:
+		return ""
+	}
+}
+
 // Plugin consumes usage records emitted by the proxy runtime.
 type Plugin interface {
 	HandleUsage(ctx context.Context, record Record)