@tractorscorch/clank 1.5.6 → 1.5.8

package/CHANGELOG.md

@@ -6,6 +6,25 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/).
 
 ---
 
+## [1.5.8] — 2026-03-23
+
+### Added
+- **Collapsible thinking blocks** — model thinking/reasoning is now displayed in a separate clickable block above the response instead of being streamed into the message text. Click the "Thought" toggle to expand/collapse. Shows "Thinking..." while streaming, "Thought" when complete
+
+### Fixed
+- **Thinking events were disconnected** — the full thinking pipeline (provider → agent → gateway → frontend) was broken at 3 points: provider yielded thinking as text for local models, agent only emitted a one-shot start event without content, and gateway didn't forward thinking events to clients. All 3 fixed
+- **Empty responses from thinking-only models** — when a model puts all output in `reasoning_content` with empty `content` (Qwen3.5), the thinking text is now used as the response instead of showing a blank message
+
+---
+
+## [1.5.7] — 2026-03-23
+
+### Fixed
+- **Tools completely broken for llama.cpp / LM Studio / vLLM models** — the PromptFallbackProvider (which injects tools into the system prompt as text) was only applied to Ollama models. All other local providers (llama.cpp, LM Studio, vLLM) sent tools via the API's `tools` parameter, which most local models can't handle — so they just ignored tools entirely. Now ALL local models that aren't in the known tool-capable list automatically get prompt-based tool injection
+- **Tool-capable model detection shared across providers** — moved the `TOOL_CAPABLE_PATTERNS` list from the Ollama provider to a shared `supportsNativeTools()` function in types.ts, used by the agent engine for any local provider
+
+---
+
 ## [1.5.6] — 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-1.5.6-blue.svg" alt="Version" /></a>
+  <a href="https://github.com/ItsTrag1c/Clank/releases/latest"><img src="https://img.shields.io/badge/version-1.5.8-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_1.5.6_macos](https://github.com/ItsTrag1c/Clank/releases/latest/download/Clank_1.5.6_macos) |
+| **macOS** (Apple Silicon) | [Clank_1.5.8_macos](https://github.com/ItsTrag1c/Clank/releases/latest/download/Clank_1.5.8_macos) |
 
 ## Features
 

package/dist/index.js

@@ -395,11 +395,27 @@ var init_auto_persist = __esm({
 });
 
 // src/providers/types.ts
-var BaseProvider;
+function supportsNativeTools(model) {
+  const name = model.includes("/") ? model.split("/").pop() : model;
+  const baseName = name.split(":")[0];
+  return TOOL_CAPABLE_PATTERNS.some((p) => p.test(baseName));
+}
+var TOOL_CAPABLE_PATTERNS, BaseProvider;
 var init_types = __esm({
   "src/providers/types.ts"() {
     "use strict";
     init_esm_shims();
+    TOOL_CAPABLE_PATTERNS = [
+      /^llama3\.[1-9]/i,
+      /^llama-3\.[1-9]/i,
+      /^qwen[23]/i,
+      /^mistral-nemo/i,
+      /^mistral-large/i,
+      /^command-r/i,
+      /^firefunction/i,
+      /^hermes-[23]/i,
+      /^nemotron/i
+    ];
     BaseProvider = class {
       /** Rough token estimate (~4 chars per token) */
       estimateTokens(messages) {
@@ -411,23 +427,12 @@ var init_types = __esm({
 });
 
 // src/providers/ollama.ts
-var TOOL_CAPABLE_PATTERNS, contextWindowCache, OllamaProvider;
+var contextWindowCache, OllamaProvider;
 var init_ollama = __esm({
   "src/providers/ollama.ts"() {
     "use strict";
     init_esm_shims();
     init_types();
-    TOOL_CAPABLE_PATTERNS = [
-      /^llama3\.[1-9]/i,
-      /^llama-3\.[1-9]/i,
-      /^qwen[23]/i,
-      /^mistral-nemo/i,
-      /^mistral-large/i,
-      /^command-r/i,
-      /^firefunction/i,
-      /^hermes-[23]/i,
-      /^nemotron/i
-    ];
     contextWindowCache = /* @__PURE__ */ new Map();
     OllamaProvider = class _OllamaProvider extends BaseProvider {
       name = "ollama";
@@ -497,8 +502,7 @@ var init_ollama = __esm({
       }
       /** Check if a model supports native tool calling */
       static supportsTools(model) {
-        const baseName = model.split(":")[0];
-        return TOOL_CAPABLE_PATTERNS.some((p) => p.test(baseName));
+        return supportsNativeTools(model);
       }
       contextWindow() {
         return contextWindowCache.get(this.model) ?? 32768;
@@ -821,6 +825,7 @@ var init_agent = __esm({
     init_esm_shims();
     init_context_engine();
     init_auto_persist();
+    init_types();
     init_ollama();
     init_prompt_fallback();
     MAX_ITERATIONS = 50;
@@ -902,11 +907,8 @@ var init_agent = __esm({
         const provider = this.resolvedProvider.provider;
         const isLocal = this.resolvedProvider.isLocal;
         let activeProvider = provider;
-        if (isLocal && provider instanceof OllamaProvider) {
-          const modelName = this.identity.model.primary.split("/").pop() || "";
-          if (!OllamaProvider.supportsTools(modelName)) {
-            activeProvider = new PromptFallbackProvider(provider);
-          }
+        if (isLocal && !supportsNativeTools(this.identity.model.primary)) {
+          activeProvider = new PromptFallbackProvider(provider);
         }
         let fullResponse = "";
         let iterationCount = 0;
@@ -932,6 +934,7 @@ var init_agent = __esm({
               denylist: this.identity.tools?.deny
             });
             let iterationText = "";
+            let thinkingText = "";
             const toolCalls = [];
             let promptTokens = 0;
             let outputTokens = 0;
@@ -940,6 +943,7 @@ var init_agent = __esm({
             for (let attempt = 0; attempt < 2; attempt++) {
               if (attempt > 0) {
                 iterationText = "";
+                thinkingText = "";
                 toolCalls.length = 0;
                 promptTokens = 0;
                 outputTokens = 0;
@@ -958,7 +962,8 @@ var init_agent = __esm({
                       this.emit("token", { content: event.content });
                       break;
                     case "thinking":
-                      this.emit("thinking-start");
+                      thinkingText += event.content;
+                      this.emit("thinking", { content: event.content });
                       break;
                     case "tool_call":
                       toolCalls.push({
@@ -1003,6 +1008,10 @@ var init_agent = __esm({
               contextPercent: Math.round(this.contextEngine.utilizationPercent())
             });
             if (toolCalls.length === 0) {
+              if (!iterationText && thinkingText) {
+                iterationText = thinkingText;
+                this.emit("token", { content: iterationText });
+              }
               fullResponse = iterationText;
               this.contextEngine.ingest({ role: "assistant", content: iterationText });
               this.emit("response-end", { text: iterationText });
@@ -3110,11 +3119,7 @@ var init_openai = __esm({
                 }
                 if (choice?.delta?.reasoning_content) {
                   hasContent = true;
-                  if (this.isLocal) {
-                    yield { type: "text", content: choice.delta.reasoning_content };
-                  } else {
-                    yield { type: "thinking", content: choice.delta.reasoning_content };
-                  }
+                  yield { type: "thinking", content: choice.delta.reasoning_content };
                 }
                 if (choice?.delta?.content) {
                   hasContent = true;
@@ -6240,7 +6245,7 @@ var init_server = __esm({
           res.writeHead(200, { "Content-Type": "application/json" });
           res.end(JSON.stringify({
             status: "ok",
-            version: "1.5.6",
+            version: "1.5.8",
             uptime: process.uptime(),
             clients: this.clients.size,
             agents: this.engines.size
@@ -6352,7 +6357,7 @@ var init_server = __esm({
         const hello = {
           type: "hello",
           protocol: PROTOCOL_VERSION,
-          version: "1.5.6",
+          version: "1.5.8",
           agents: this.config.agents.list.map((a) => ({
             id: a.id,
             name: a.name || a.id,
@@ -6617,6 +6622,7 @@ var init_server = __esm({
       wireEngineEvents(engine, client) {
         const eventMap = {
           "token": "token",
+          "thinking": "thinking",
           "response-start": "response-start",
           "response-end": "response-end",
           "tool-start": "tool-start",
@@ -7747,7 +7753,7 @@ async function runTui(opts) {
   ws.on("open", () => {
     ws.send(JSON.stringify({
       type: "connect",
-      params: { auth: { token }, mode: "tui", version: "1.5.6" }
+      params: { auth: { token }, mode: "tui", version: "1.5.8" }
     }));
   });
   ws.on("message", (data) => {
@@ -8176,7 +8182,7 @@ 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 = "1.5.6";
+var version = "1.5.8";
 try {
   const pkg = JSON.parse(readFileSync(join19(__dirname3, "..", "package.json"), "utf-8"));
   version = pkg.version;