fix(xai): route grok build compactions via large context model

Author: GGBond6

internal/runtime/executor/xai_executor.go

@@ -480,6 +480,7 @@ type xaiPreparedRequest struct {
 
 func (e *XAIExecutor) prepareResponsesRequest(ctx context.Context, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, stream bool) (*xaiPreparedRequest, error) {
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
+	baseModel = xaiCompactionModel(baseModel, opts)
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	originalPayloadSource := req.Payload
@@ -601,6 +602,43 @@ func xaiExecutionSessionID(req cliproxyexecutor.Request, opts cliproxyexecutor.O
 	return ""
 }
 
+func xaiCompactionModel(model string, opts cliproxyexecutor.Options) string {
+	name := strings.ToLower(strings.TrimSpace(thinking.ParseSuffix(model).ModelName))
+	if idx := strings.LastIndex(name, "/"); idx >= 0 {
+		name = name[idx+1:]
+	}
+	if name != "grok-build-0.1" || !xaiIsCompactionRequest(opts) {
+		return model
+	}
+	return "grok-4.3"
+}
+
+func xaiIsCompactionRequest(opts cliproxyexecutor.Options) bool {
+	if opts.Alt == "responses/compact" {
+		return true
+	}
+	if opts.Headers != nil {
+		if xaiTurnMetadataIsCompaction(opts.Headers.Get("X-Codex-Turn-Metadata")) {
+			return true
+		}
+	}
+	if raw, ok := opts.Metadata["X-Codex-Turn-Metadata"]; ok {
+		return xaiTurnMetadataIsCompaction(fmt.Sprint(raw))
+	}
+	if raw, ok := opts.Metadata["x-codex-turn-metadata"]; ok {
+		return xaiTurnMetadataIsCompaction(fmt.Sprint(raw))
+	}
+	return false
+}
+
+func xaiTurnMetadataIsCompaction(raw string) bool {
+	raw = strings.TrimSpace(raw)
+	if raw == "" || !gjson.Valid(raw) {
+		return false
+	}
+	return strings.TrimSpace(gjson.Get(raw, "request_kind").String()) == "compaction"
+}
+
 func xaiImageEndpointPath(opts cliproxyexecutor.Options) string {
 	if opts.SourceFormat.String() != xaiImageHandlerType {
 		return ""

internal/runtime/executor/xai_executor_test.go

@@ -196,6 +196,57 @@ func TestXAIExecutorOmitsUnsupportedReasoningEffort(t *testing.T) {
 	}
 }
 
+func TestXAIExecutorUsesLargeContextModelForGrokBuildCompaction(t *testing.T) {
+	exec := NewXAIExecutor(&config.Config{})
+	req := cliproxyexecutor.Request{
+		Model:   "grok-build-0.1",
+		Payload: []byte(`{"model":"grok-build-0.1","input":"hello"}`),
+	}
+
+	for _, opts := range []cliproxyexecutor.Options{
+		{
+			SourceFormat: sdktranslator.FormatOpenAIResponse,
+			Alt:          "responses/compact",
+		},
+		{
+			SourceFormat: sdktranslator.FormatOpenAIResponse,
+			Headers: http.Header{
+				"X-Codex-Turn-Metadata": []string{`{"request_kind":"compaction","compaction":{"trigger":"auto","reason":"context_limit"}}`},
+			},
+		},
+	} {
+		prepared, err := exec.prepareResponsesRequest(context.Background(), req, opts, true)
+		if err != nil {
+			t.Fatalf("prepareResponsesRequest() error = %v", err)
+		}
+		if prepared.baseModel != "grok-4.3" {
+			t.Fatalf("baseModel = %q, want grok-4.3", prepared.baseModel)
+		}
+		if got := gjson.GetBytes(prepared.body, "model").String(); got != "grok-4.3" {
+			t.Fatalf("body model = %q, want grok-4.3; body=%s", got, string(prepared.body))
+		}
+	}
+}
+
+func TestXAIExecutorKeepsGrokBuildModelForNormalRequests(t *testing.T) {
+	exec := NewXAIExecutor(&config.Config{})
+	prepared, err := exec.prepareResponsesRequest(context.Background(), cliproxyexecutor.Request{
+		Model:   "grok-build-0.1",
+		Payload: []byte(`{"model":"grok-build-0.1","input":"hello"}`),
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FormatOpenAIResponse,
+	}, true)
+	if err != nil {
+		t.Fatalf("prepareResponsesRequest() error = %v", err)
+	}
+	if prepared.baseModel != "grok-build-0.1" {
+		t.Fatalf("baseModel = %q, want grok-build-0.1", prepared.baseModel)
+	}
+	if got := gjson.GetBytes(prepared.body, "model").String(); got != "grok-build-0.1" {
+		t.Fatalf("body model = %q, want grok-build-0.1; body=%s", got, string(prepared.body))
+	}
+}
+
 func TestXAIExecutorAppliesThinkingSuffix(t *testing.T) {
 	var gotBody []byte
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {