@tractorscorch/clank 1.5.0 → 1.5.2

package/CHANGELOG.md

@@ -6,6 +6,24 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/).
 
 ---
 
+## [1.5.2] — 2026-03-23
+
+### Fixed
+- **Thinking models (Qwen3.5) exhaust tokens on reasoning** — the model generates `<think>` reasoning tokens that eat the entire context window, leaving nothing for actual content. Added default `max_tokens: 4096` for local models and `reasoning_effort: "low"` to reduce thinking overhead
+- **Telegram shows nothing during model thinking** — added periodic "typing" indicator every 4 seconds so the bot doesn't appear dead while the model processes internally
+- **Root cause found via direct API testing** — Qwen3.5-35B returns empty `content` with all output in `reasoning_content`; without a max_tokens cap, the model spends all its budget on thinking
+
+---
+
+## [1.5.1] — 2026-03-23
+
+### Fixed
+- **Local models timing out on tool calls** — removed per-chunk read timeout that was killing legitimate slow processing; a 35B quantized model can take minutes for prefill on large contexts, that's normal not a hang
+- **Local model timeout increased to 5 minutes** — was 120s (too short for large quantized models doing prefill on big contexts with tool results)
+- **Memory budget reduced for local models** — memory injection now uses 1.5K chars (was 4K) to avoid eating the limited context window of local models (8K-32K vs 128K+ for cloud)
+
+---
+
 ## [1.5.0] — 2026-03-23
 
 ### Fixed

package/README.md

@@ -9,7 +9,7 @@
 </p>
 
 <p align="center">
-  <a href="https://github.com/ItsTrag1c/Clank/releases/latest"><img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version" /></a>
+  <a href="https://github.com/ItsTrag1c/Clank/releases/latest"><img src="https://img.shields.io/badge/version-1.5.2-blue.svg" alt="Version" /></a>
   <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License" /></a>
   <a href="https://www.npmjs.com/package/@tractorscorch/clank"><img src="https://img.shields.io/npm/v/@tractorscorch/clank.svg" alt="npm" /></a>
   <a href="https://github.com/ItsTrag1c/Clank/stargazers"><img src="https://img.shields.io/github/stars/ItsTrag1c/Clank.svg" alt="Stars" /></a>
@@ -75,7 +75,7 @@ That's it. Setup auto-detects your local models, configures the gateway, and get
 | Platform | Download |
 |----------|----------|
 | **npm** (all platforms) | `npm install -g @tractorscorch/clank` |
-| **macOS** (Apple Silicon) | [Clank_1.5.0_macos](https://github.com/ItsTrag1c/Clank/releases/latest/download/Clank_1.5.0_macos) |
+| **macOS** (Apple Silicon) | [Clank_1.5.2_macos](https://github.com/ItsTrag1c/Clank/releases/latest/download/Clank_1.5.2_macos) |
 
 ## Features
 

package/dist/index.js

@@ -561,7 +561,7 @@ var init_ollama = __esm({
         if (this.maxResponseTokens) {
           body.max_tokens = this.maxResponseTokens;
         }
-        const timeoutSignal = AbortSignal.timeout(12e4);
+        const timeoutSignal = AbortSignal.timeout(3e5);
         const effectiveSignal = signal ? AbortSignal.any([signal, timeoutSignal]) : timeoutSignal;
         const res = await fetch(`${this.baseUrl}/v1/chat/completions`, {
           method: "POST",
@@ -580,14 +580,9 @@ var init_ollama = __esm({
         const decoder = new TextDecoder();
         let buffer = "";
         const toolCalls = /* @__PURE__ */ new Map();
-        const CHUNK_TIMEOUT = 6e4;
         try {
           while (true) {
-            const readPromise = reader.read();
-            const timeoutPromise = new Promise(
-              (_, reject) => setTimeout(() => reject(new Error("Model stopped responding (no data for 60s)")), CHUNK_TIMEOUT)
-            );
-            const { done, value } = await Promise.race([readPromise, timeoutPromise]);
+            const { done, value } = await reader.read();
             if (done) break;
             buffer += decoder.decode(value, { stream: true });
             const lines = buffer.split("\n");
@@ -2992,11 +2987,16 @@ var init_openai = __esm({
           stream: true,
           stream_options: { include_usage: true }
         };
+        if (this.isLocal) {
+          body.reasoning_effort = "low";
+        }
         if (tools.length > 0) {
           body.tools = this.formatTools(tools);
         }
         if (this.maxResponseTokens) {
           body.max_tokens = this.maxResponseTokens;
+        } else if (this.isLocal) {
+          body.max_tokens = 4096;
         }
         const headers = {
           "Content-Type": "application/json"
@@ -3004,7 +3004,7 @@ var init_openai = __esm({
         if (this.apiKey) {
           headers["Authorization"] = `Bearer ${this.apiKey}`;
         }
-        const timeoutMs = this.isLocal ? 12e4 : 9e4;
+        const timeoutMs = this.isLocal ? 3e5 : 9e4;
         const timeoutSignal = AbortSignal.timeout(timeoutMs);
         const effectiveSignal = signal ? AbortSignal.any([signal, timeoutSignal]) : timeoutSignal;
         const res = await fetch(`${this.baseUrl}/v1/chat/completions`, {
@@ -3022,14 +3022,9 @@ var init_openai = __esm({
         const decoder = new TextDecoder();
         let buffer = "";
         const toolCalls = /* @__PURE__ */ new Map();
-        const CHUNK_TIMEOUT = 6e4;
         try {
           while (true) {
-            const readPromise = reader.read();
-            const timeoutPromise = new Promise(
-              (_, reject) => setTimeout(() => reject(new Error("Model stopped responding (no data for 60s)")), CHUNK_TIMEOUT)
-            );
-            const { done, value } = await Promise.race([readPromise, timeoutPromise]);
+            const { done, value } = await reader.read();
             if (done) break;
             buffer += decoder.decode(value, { stream: true });
             const lines = buffer.split("\n");
@@ -5308,6 +5303,10 @@ var init_telegram = __esm({
               try {
                 console.log(`  Telegram: processing message from ${userId} in ${chatId}`);
                 await ctx.api.sendChatAction(chatId, "typing");
+                const typingInterval2 = setInterval(() => {
+                  bot.api.sendChatAction(chatId, "typing").catch(() => {
+                  });
+                }, 4e3);
                 let streamMsgId = null;
                 let sendingInitial = false;
                 let accumulated = "";
@@ -5376,8 +5375,10 @@ var init_telegram = __esm({
                     await ctx.api.sendMessage(chatId, chunk);
                   }
                 }
+                clearInterval(typingInterval2);
                 console.log(`  Telegram: response complete (${response?.length || 0} chars)`);
               } catch (err) {
+                clearInterval(typingInterval);
                 const errMsg = err instanceof Error ? err.message : String(err);
                 console.error(`  Telegram: message handler error \u2014 ${errMsg}`);
                 await ctx.api.sendMessage(chatId, `Error: ${errMsg.slice(0, 200)}`).catch(() => {
@@ -6165,7 +6166,7 @@ var init_server = __esm({
           res.writeHead(200, { "Content-Type": "application/json" });
           res.end(JSON.stringify({
             status: "ok",
-            version: "1.5.0",
+            version: "1.5.2",
             uptime: process.uptime(),
             clients: this.clients.size,
             agents: this.engines.size
@@ -6277,7 +6278,7 @@ var init_server = __esm({
         const hello = {
           type: "hello",
           protocol: PROTOCOL_VERSION,
-          version: "1.5.0",
+          version: "1.5.2",
           agents: this.config.agents.list.map((a) => ({
             id: a.id,
             name: a.name || a.id,
@@ -6519,7 +6520,8 @@ var init_server = __esm({
           compact,
           thinking
         });
-        const memoryBlock = await this.memoryManager.buildMemoryBlock("", identity.workspace);
+        const memoryBudget = resolved.isLocal ? 1500 : 4e3;
+        const memoryBlock = await this.memoryManager.buildMemoryBlock("", identity.workspace, memoryBudget);
         const fullPrompt = memoryBlock ? systemPrompt + "\n\n---\n\n" + memoryBlock : systemPrompt;
         engine = new AgentEngine({
           identity,
@@ -7671,7 +7673,7 @@ async function runTui(opts) {
   ws.on("open", () => {
     ws.send(JSON.stringify({
       type: "connect",
-      params: { auth: { token }, mode: "tui", version: "1.5.0" }
+      params: { auth: { token }, mode: "tui", version: "1.5.2" }
     }));
   });
   ws.on("message", (data) => {
@@ -8100,7 +8102,7 @@ import { fileURLToPath as fileURLToPath5 } from "url";
 import { dirname as dirname5, join as join19 } from "path";
 var __filename3 = fileURLToPath5(import.meta.url);
 var __dirname3 = dirname5(__filename3);
-var version = "1.5.0";
+var version = "1.5.2";
 try {
   const pkg = JSON.parse(readFileSync(join19(__dirname3, "..", "package.json"), "utf-8"));
   version = pkg.version;