@joshuaswarren/openclaw-engram 9.0.13 → 9.0.14

package/dist/index.js

@@ -985,6 +985,7 @@ var LocalLlmClient = class _LocalLlmClient {
   consecutive400s = 0;
   cooldownUntilMs = 0;
   modelRegistry;
+  _disableThinking = false;
   static HEALTH_CHECK_INTERVAL_MS = 6e4;
   // 1 minute
   static LMS_CACHE_INTERVAL_MS = 3e4;
@@ -993,6 +994,14 @@ var LocalLlmClient = class _LocalLlmClient {
     this.config = config;
     this.modelRegistry = modelRegistry;
   }
+  /**
+   * Disable thinking/reasoning mode for models that support it (e.g. Qwen 3.5).
+   * When enabled, adds chat_template_kwargs to suppress chain-of-thought,
+   * reducing latency for fast-tier operations.
+   */
+  set disableThinking(value) {
+    this._disableThinking = value;
+  }
   resolveHomeDir() {
     return this.config.localLlmHomeDir || process.env.HOME || os.homedir();
   }
@@ -1432,6 +1441,9 @@ var LocalLlmClient = class _LocalLlmClient {
       if (options.responseFormat?.type === "json_schema") {
         requestBody.response_format = options.responseFormat;
       }
+      if (this._disableThinking) {
+        requestBody.chat_template_kwargs = { enable_thinking: false };
+      }
       const baseUrl = this.config.localLlmUrl.replace("localhost", "127.0.0.1").replace(/\/+$/, "");
       const chatUrl = `${baseUrl}/chat/completions`;
       const requestBodyJson = JSON.stringify(requestBody);
@@ -16495,10 +16507,14 @@ var Orchestrator = class _Orchestrator {
     this.policyRuntime = new PolicyRuntimeManager(config.memoryDir, config);
     this.summarizer = new HourlySummarizer(config, config.gatewayConfig, this.modelRegistry, this.transcript);
     this.localLlm = new LocalLlmClient(config, this.modelRegistry);
-    this.fastLlm = config.localLlmFastEnabled ? new LocalLlmClient(
-      { ...config, localLlmModel: config.localLlmFastModel || config.localLlmModel, localLlmUrl: config.localLlmFastUrl, localLlmTimeoutMs: config.localLlmFastTimeoutMs },
-      this.modelRegistry
-    ) : this.localLlm;
+    this.fastLlm = config.localLlmFastEnabled ? (() => {
+      const client = new LocalLlmClient(
+        { ...config, localLlmModel: config.localLlmFastModel || config.localLlmModel, localLlmUrl: config.localLlmFastUrl, localLlmTimeoutMs: config.localLlmFastTimeoutMs },
+        this.modelRegistry
+      );
+      client.disableThinking = true;
+      return client;
+    })() : this.localLlm;
     this.extraction = new ExtractionEngine(config, this.localLlm, config.gatewayConfig, this.modelRegistry);
     this.threading = new ThreadingManager(
       path30.join(config.memoryDir, "threads"),