feat: support reasoning in responses chat adapter

aigateway/component/openai.go

@@ -813,6 +813,7 @@ func dbUpstreamsToConfigs(dbUpstreams []database.Upstream) []commontypes.Upstrea
 			HealthCheckEnabled:    u.HealthCheckEnabled,
 			CircuitBreakerEnabled: u.CircuitBreakerEnabled,
 			Tags:                  u.Tags,
+			Metadata:              u.Metadata,
 			LimitPolicy:           u.LimitPolicy,
 		}
 		// Carry health/circuit state from DB so the proxy path can use

aigateway/component/openai_test.go

@@ -1484,3 +1484,23 @@ func TestOpenAIComponentImpl_RecordUsage_ExternalModel(t *testing.T) {
 		})
 	}
 }
+
+func TestDbUpstreamsToConfigsMetadataPassthrough(t *testing.T) {
+	metadata := map[string]any{
+		"responses": map[string]any{
+			"chat_adapter": map[string]any{
+				"reasoning_request": map[string]any{
+					"enabled": true,
+				},
+			},
+		},
+	}
+	result := dbUpstreamsToConfigs([]database.Upstream{{
+		ID:       1,
+		URL:      "http://upstream.example.com/v1/chat/completions",
+		Enabled:  true,
+		Metadata: metadata,
+	}})
+	require.Len(t, result, 1)
+	require.Equal(t, metadata, result[0].Metadata)
+}

aigateway/handler/openai_responses_adapter.go

@@ -16,7 +16,7 @@ func (h *OpenAIHandlerImpl) executeAdapterResponses(c *gin.Context, req *types.R
 		writeResponsesError(c, http.StatusBadRequest, adapterErrorCode(err), "invalid_request_error", err.Error())
 		return
 	}
-	chatReq, err := responsesToChatRequest(c.Request.Context(), req, modelTarget.ModelName)
+	chatReq, err := responsesToChatRequest(c.Request.Context(), req, modelTarget.ModelName, modelTarget.Upstream.Metadata)
 	if err != nil {
 		writeResponsesError(c, http.StatusBadRequest, adapterErrorCode(err), "invalid_request_error", err.Error())
 		return