@flutchai/flutch-sdk 0.7.7 → 0.7.9

package/dist/index.cjs

@@ -2131,7 +2131,8 @@ exports.GraphController = class _GraphController {
   async getSupportedGraphTypes() {
     return this.graphService.getSupportedGraphTypes();
   }
-  async generateAnswer(payload) {
+  async generateAnswer(payload, res) {
+    this.cancelOnDisconnect(res, payload.requestId);
     try {
       return await this.graphService.generateAnswer(payload);
     } catch (error) {
@@ -2144,6 +2145,7 @@ exports.GraphController = class _GraphController {
     res.setHeader("Transfer-Encoding", "chunked");
     res.setHeader("Cache-Control", "no-cache");
     res.setHeader("Connection", "keep-alive");
+    this.cancelOnDisconnect(res, payload.requestId);
     try {
       this.logger.debug(`[CONTROLLER] Starting streamAnswer for requestId: ${payload.requestId}`);
       const result = await this.graphService.streamAnswer(payload, (chunk) => {
@@ -2181,6 +2183,19 @@ exports.GraphController = class _GraphController {
       res.end();
     }
   }
+  /**
+  * A caller that hangs up before the answer is done — its own timeout, a
+  * restart, a cancelled parent run — no longer waits for it, so the run is
+  * cancelled: left going, it keeps calling models and tools nobody reads.
+  */
+  cancelOnDisconnect(res, requestId) {
+    if (!requestId) return;
+    res.on("close", () => {
+      if (res.writableFinished) return;
+      this.logger.warn(`[CONTROLLER] Caller disconnected before the answer: cancelling ${requestId}`);
+      this.graphService.cancelGeneration(requestId).catch((error) => this.logger.warn(`[CONTROLLER] Cancel after disconnect failed for ${requestId}: ${error.message}`));
+    });
+  }
   async cancelGeneration(requestId) {
     try {
       await this.graphService.cancelGeneration(requestId);
@@ -2254,9 +2269,13 @@ _ts_decorate8([
     description: "Answer generated"
   }),
   _ts_param4(0, common.Body()),
+  _ts_param4(1, common.Res({
+    passthrough: true
+  })),
   _ts_metadata6("design:type", Function),
   _ts_metadata6("design:paramtypes", [
-    typeof IGraphRequestPayload === "undefined" ? Object : IGraphRequestPayload
+    typeof IGraphRequestPayload === "undefined" ? Object : IGraphRequestPayload,
+    typeof express.Response === "undefined" ? Object : express.Response
   ]),
   _ts_metadata6("design:returntype", Promise)
 ], exports.GraphController.prototype, "generateAnswer", null);
@@ -7037,7 +7056,7 @@ exports.McpRuntimeHttpClient = class _McpRuntimeHttpClient {
   /**
   * Execute a tool by name with given arguments
   */
-  async executeTool(name, args, context, mcpServers) {
+  async executeTool(name, args, context, mcpServers, signal) {
     try {
       this.logger.debug(`Executing tool: ${name} with args:`, args);
       const payload = {
@@ -7050,7 +7069,9 @@ exports.McpRuntimeHttpClient = class _McpRuntimeHttpClient {
       if (mcpServers && mcpServers.length > 0) {
         payload.mcpServers = mcpServers;
       }
-      const response = await this.httpClient.post("/tools/execute", payload);
+      const response = signal ? await this.httpClient.post("/tools/execute", payload, {
+        signal
+      }) : await this.httpClient.post("/tools/execute", payload);
       this.logger.log(`Tool ${name} executed successfully`);
       return response.data;
     } catch (error) {
@@ -7118,7 +7139,7 @@ exports.McpRuntimeHttpClient = class _McpRuntimeHttpClient {
           toolName
         ]
       }, JSON.stringify(enrichedArgs), parsedConfig.runId, void 0, parsedConfig.tags, parsedConfig.metadata, toolName);
-      const result = await this.executeTool(toolName, enrichedArgs, executionContext, mcpServers);
+      const result = await this.executeTool(toolName, enrichedArgs, executionContext, mcpServers, config?.signal);
       const content = result.success ? JSON.stringify(result) : result.error || JSON.stringify(result);
       await runManager?.handleToolEnd(content);
       return {
@@ -7588,6 +7609,22 @@ function buildEffortRequestFields(effort) {
   } : void 0;
 }
 __name(buildEffortRequestFields, "buildEffortRequestFields");
+function buildReasoningRequestFields(effort, thinking) {
+  if (!effort && !thinking) return void 0;
+  return {
+    ...effort && {
+      output_config: {
+        effort
+      }
+    },
+    ...thinking && {
+      thinking: {
+        type: thinking
+      }
+    }
+  };
+}
+__name(buildReasoningRequestFields, "buildReasoningRequestFields");
 function resolveSamplingParams(modelIdentifiers, requested, supportsSamplingParams) {
   const accepted = supportsSamplingParams ?? modelIdentifiers.every((id) => modelAcceptsSamplingParams(id));
   if (accepted) {
@@ -7935,7 +7972,8 @@ var ModelInitializer = class _ModelInitializer {
     const withSampling = config.supportsSamplingParams === void 0 ? withMcp : `${withMcp}:sp:${config.supportsSamplingParams}`;
     const withCache = config.promptCache === void 0 && config.promptCacheTtl === void 0 ? withSampling : `${withSampling}:pc:${config.promptCache ?? "auto"}:${config.promptCacheTtl ?? "default"}`;
     const withEffort = config.effort === void 0 ? withCache : `${withCache}:eff:${config.effort}`;
-    return config.toolSearch?.enabled ? `${withEffort}:ts:${JSON.stringify(config.toolSearch)}` : withEffort;
+    const withThinking = config.thinking === void 0 ? withEffort : `${withEffort}:th:${config.thinking}`;
+    return config.toolSearch?.enabled ? `${withThinking}:ts:${JSON.stringify(config.toolSearch)}` : withThinking;
   }
   // Chat model creators
   chatModelCreators = {
@@ -7950,7 +7988,7 @@ var ModelInitializer = class _ModelInitializer {
       }
       return new openai.ChatOpenAI(config);
     },
-    [ModelProvider.ANTHROPIC]: ({ modelName, defaultTemperature, defaultMaxTokens, apiToken, baseURL }, promptCacheControl) => {
+    [ModelProvider.ANTHROPIC]: ({ modelName, defaultTemperature, defaultMaxTokens, apiToken, baseURL, defaultThinking }, promptCacheControl) => {
       const routerURL = resolveRouterURL(baseURL);
       return new anthropic.ChatAnthropic({
         modelName,
@@ -7967,6 +8005,12 @@ var ModelInitializer = class _ModelInitializer {
         ...promptCacheControl && {
           cache_control: promptCacheControl
         },
+        // The thinking mode as the API's own parameter; absent when unset.
+        ...defaultThinking && {
+          thinking: {
+            type: defaultThinking
+          }
+        },
         anthropicApiKey: apiToken || this.resolveApiKey(ModelProvider.ANTHROPIC),
         ...routerURL && {
           anthropicApiUrl: routerURL
@@ -8173,7 +8217,8 @@ var ModelInitializer = class _ModelInitializer {
       defaultTemperature: sampling.temperature,
       defaultMaxTokens: toOptionalNumber(config.maxTokens) ?? toOptionalNumber(modelConfig.defaultMaxTokens),
       baseURL: config.baseURL ?? modelConfig.baseURL,
-      defaultEffort: config.effort ?? modelConfig.defaultEffort
+      defaultEffort: config.effort ?? modelConfig.defaultEffort,
+      defaultThinking: config.thinking ?? modelConfig.defaultThinking
     };
     const promptCacheControl = resolvePromptCacheControl([
       finalConfig.bedrockModelId,
@@ -8185,7 +8230,7 @@ var ModelInitializer = class _ModelInitializer {
     this.logger.debug(`Creating new chat model instance: ${cacheKey}`);
     let model;
     if (finalConfig.useBedrock && finalConfig.bedrockModelId) {
-      const effortFields = buildEffortRequestFields(config.effort ?? modelConfig.defaultEffort);
+      const effortFields = buildReasoningRequestFields(config.effort ?? modelConfig.defaultEffort, config.thinking ?? modelConfig.defaultThinking);
       model = new FlutchChatBedrockConverse({
         model: finalConfig.bedrockModelId,
         region: this.resolveBedrockRegion(),
@@ -9190,6 +9235,7 @@ exports.bootstrap = bootstrap;
 exports.buildEffortRequestFields = buildEffortRequestFields;
 exports.buildOAuthAuthorizationUrl = buildOAuthAuthorizationUrl;
 exports.buildOpenAIModelConfig = buildOpenAIModelConfig;
+exports.buildReasoningRequestFields = buildReasoningRequestFields;
 exports.clearAttachmentDataStore = clearAttachmentDataStore;
 exports.createEndpointDescriptors = createEndpointDescriptors;
 exports.createGraphAttachment = createGraphAttachment;