open-agents-ai 0.187.370 → 0.187.371

package/dist/index.js

@@ -515664,6 +515664,26 @@ ${todoItems}
 </system-reminder>`;
   return { shouldInject: true, content, reason: "injected" };
 }
+function stripThinkBlocks(s2) {
+  if (!s2)
+    return s2;
+  return s2.replace(/<think>[\s\S]*?<\/think>/g, "").trim();
+}
+function computeEffectiveThink(params) {
+  if (process.env["OA_FORCE_NO_THINK"] === "1")
+    return false;
+  if (params.hasTools)
+    return false;
+  if (typeof params.requestThink === "boolean")
+    return params.requestThink;
+  if (process.env["OA_THINK_AUTO"] === "1" && Array.isArray(params.messages)) {
+    const blob = params.messages.filter((m2) => m2.role === "user" || m2.role === "system").map((m2) => typeof m2.content === "string" ? m2.content : "").join("\n").toLowerCase();
+    if (/\b(plan|decompose|analyze(?:\s+complex)?|step\s*by\s*step|reason through|think through)\b/.test(blob)) {
+      return true;
+    }
+  }
+  return params.defaultThink;
+}
 var SYSTEM_PROMPT, SYSTEM_PROMPT_MEDIUM, SYSTEM_PROMPT_SMALL, VISUAL_TOOLS, AUDIO_TOOLS, SOCIAL_TOOLS, SPATIAL_TOOLS, CODE_TOOLS, AgenticRunner, OllamaAgenticBackend;
 var init_agenticRunner = __esm({
   "packages/orchestrator/dist/agenticRunner.js"() {
@@ -521116,7 +521136,7 @@ ${description}`
         this.baseUrl = normalizeBaseUrl(baseUrl);
         this.model = model;
         this.apiKey = apiKey ?? "";
-        this.thinking = thinking ?? true;
+        this.thinking = thinking ?? false;
         this._isAnthropic = /api\.anthropic\.com/i.test(baseUrl);
       }
       /** Set multiple API keys for round-robin rotation per request */
@@ -521151,13 +521171,24 @@ ${description}`
         if (this._isAnthropic) {
           return this._anthropicChatCompletion(request);
         }
+        const cleanedMessages = request.messages.map((m2) => m2.role === "assistant" && typeof m2.content === "string" ? { ...m2, content: stripThinkBlocks(m2.content) } : m2);
+        const effectiveThink = computeEffectiveThink({
+          requestThink: request.think,
+          defaultThink: this.thinking,
+          hasTools: Array.isArray(request.tools) && request.tools.length > 0,
+          messages: cleanedMessages
+        });
+        let effectiveMaxTokens = request.maxTokens;
+        if (effectiveThink === true && (effectiveMaxTokens ?? 0) < 4096) {
+          effectiveMaxTokens = 4096;
+        }
         const body = {
           model: this.model,
-          messages: request.messages,
+          messages: cleanedMessages,
           tools: request.tools,
           temperature: request.temperature,
-          max_tokens: request.maxTokens,
-          think: this.thinking
+          max_tokens: effectiveMaxTokens,
+          think: effectiveThink
         };
         const fetchOpts = {
           method: "POST",
@@ -521314,15 +521345,26 @@ ${description}`
        * The existing chatCompletion() method is completely unmodified.
        */
       async *chatCompletionStream(request) {
+        const cleanedMessages = request.messages.map((m2) => m2.role === "assistant" && typeof m2.content === "string" ? { ...m2, content: stripThinkBlocks(m2.content) } : m2);
+        const effectiveThink = computeEffectiveThink({
+          requestThink: request.think,
+          defaultThink: this.thinking,
+          hasTools: Array.isArray(request.tools) && request.tools.length > 0,
+          messages: cleanedMessages
+        });
+        let effectiveMaxTokens = request.maxTokens;
+        if (effectiveThink === true && (effectiveMaxTokens ?? 0) < 4096) {
+          effectiveMaxTokens = 4096;
+        }
         const body = {
           model: this.model,
-          messages: request.messages,
+          messages: cleanedMessages,
           tools: request.tools,
           temperature: request.temperature,
-          max_tokens: request.maxTokens,
+          max_tokens: effectiveMaxTokens,
           stream: true,
           stream_options: { include_usage: true },
-          think: this.thinking
+          think: effectiveThink
         };
         const streamFetchOpts = {
           method: "POST",
@@ -546123,12 +546165,36 @@ Clone a new voice: /voice clone <wav-file> [name]`);
       return "handled";
     }
     case "think": {
-      const isOn = ctx3.thinkToggle();
+      const token = (arg || "").trim().toLowerCase();
+      const desc = (s2) => s2 ? "🧠 models that support reasoning (Qwen3, DeepSeek-R1, etc.) will show their thinking chain; tool calls still run direct" : "⚡ direct-answer mode (reasoning suppressed); recommended for tool-heavy workflows";
+      if (token === "status" || token === "?") {
+        const cur = ctx3.config.thinking ?? false;
+        renderInfo2(`Thinking mode: ${cur ? "on" : "off"} — ${desc(cur)}`);
+        if (process.env["OA_THINK_AUTO"] === "1") renderInfo2("Auto-heuristic active (OA_THINK_AUTO=1)");
+        if (process.env["OA_FORCE_NO_THINK"] === "1") renderWarning2("OA_FORCE_NO_THINK=1 forces off regardless of /think setting");
+        return "handled";
+      }
+      if (token === "auto") {
+        process.env["OA_THINK_AUTO"] = "1";
+        renderInfo2("Thinking auto-heuristic enabled: /think flips on when user message contains plan/decompose/analyze/step-by-step/reason-through. Tool calls still force off. Unset with OA_THINK_AUTO=0.");
+        return "handled";
+      }
+      let isOn;
+      if (token === "on" || token === "true" || token === "yes" || token === "1") {
+        isOn = true;
+        ctx3.config.thinking = true;
+      } else if (token === "off" || token === "false" || token === "no" || token === "0") {
+        isOn = false;
+        ctx3.config.thinking = false;
+      } else {
+        isOn = ctx3.thinkToggle();
+      }
       const save2 = hasLocal ? ctx3.saveLocalSettings.bind(ctx3) : ctx3.saveSettings.bind(ctx3);
       save2({ thinking: isOn });
-      renderInfo2(
-        `Thinking mode: ${isOn ? "on" : "off"}${hasLocal ? " (project-local)" : ""}` + (isOn ? " — models that support reasoning (Qwen3, DeepSeek-R1, etc.) will show their thinking chain" : " — reasoning chain suppressed, model responds directly")
-      );
+      renderInfo2(`Thinking mode: ${isOn ? "on" : "off"}${hasLocal ? " (project-local)" : ""} — ${desc(isOn)}`);
+      if (isOn) {
+        renderInfo2("Note: max_tokens will auto-raise to ≥4096 per request to prevent <think> truncation.");
+      }
       return "handled";
     }
     case "tools": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.187.370",
+  "version": "0.187.371",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",