@stevederico/dotbot 0.28.0 → 0.31.0

package/CHANGELOG.md

@@ -1,3 +1,35 @@
+0.31
+
+  Document mlx_local provider
+  Fix CLI local auth
+  Remove shields.io badges
+  Generify personal path example
+
+0.30
+
+  Flush buffered plain text
+  Add agentLoop regression tests
+  Rename dottie_desktop to mlx_local
+  Rename getDottieDesktopStatus helper
+  Rename DOTTIE_DESKTOP_URL env var
+  Parameterize notification title
+  Default agentName to Assistant
+  Scrub host-specific doc references
+  Skip heartbeat without tasks
+  Fail closed on task fetch error
+  Add mlx_local provider entry
+  Add cron_handler regression tests
+
+0.29
+
+  Extract shared streamEvents
+  Remove dead databaseManager logging
+  Remove dead CDP methods
+  Remove dead compat aliases
+  Remove dead observer module
+  Consolidate cron row mapping
+  Update README sandbox docs
+
 0.28
 
   Add --sandbox mode

package/README.md

@@ -5,18 +5,6 @@
     The ultra-lean AI agent.<br>
     11k lines. 53 tools. 0 dependencies.
   </h3>
-  <p align="center">
-    <a href="https://opensource.org/licenses/mit">
-      <img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="MIT License">
-    </a>
-    <a href="https://github.com/stevederico/dotbot/stargazers">
-      <img src="https://img.shields.io/github/stars/stevederico/dotbot?style=social" alt="GitHub stars">
-    </a>
-    <a href="https://github.com/stevederico/dotbot">
-      <img src="https://img.shields.io/badge/version-0.25-green" alt="version">
-    </a>
-    <img src="https://img.shields.io/badge/LOC-11k-orange" alt="Lines of Code">
-  </p>
 </div>
 
 <br />
@@ -27,9 +15,10 @@
 
 | | dotbot | nanobot | OpenClaw |
 |---|:---:|:---:|:---:|
-| **Lines of Code** | **11k** | 22k | 1M+ |
+| **Lines of Code** | **~11k** | 22k | 1M+ |
 | **Tools** | **53** | ~10 | ~50 |
-| **Dependencies** | Minimal | Heavy | Heavy |
+| **Dependencies** | **0** | Heavy | Heavy |
+| **Sandbox Mode** | **Built-in** | No | Requires NemoClaw |
 
 Everything you need for AI agents. Nothing you don't. No bloated abstractions. No dependency hell. Just a clean, focused agent that works.
 
@@ -43,7 +32,9 @@ A **streaming AI agent** with tool execution, autonomous tasks, and scheduled jo
 ```bash
 dotbot "What's the weather in San Francisco?"
 dotbot                  # Interactive mode
+dotbot --sandbox        # Sandbox mode (restricted tools)
 dotbot serve --port 3000
+dotbot models           # List available models
 dotbot tools            # List all 53 tools
 ```
 
@@ -80,6 +71,49 @@ dotbot stats
 dotbot memory
 ```
 
+### Sandbox Mode
+
+Run dotbot with restricted tool access — deny-by-default.
+
+```bash
+# Full lockdown — safe tools only (memory, search, weather, tasks)
+dotbot --sandbox "What is 2+2?"
+
+# Allow specific domains for web_fetch and browser_navigate
+dotbot --sandbox --allow github
+dotbot --sandbox --allow github --allow slack
+
+# Allow specific tool groups
+dotbot --sandbox --allow messages
+dotbot --sandbox --allow images
+
+# Mix domains and tool groups
+dotbot --sandbox --allow github --allow messages --allow npm
+
+# Custom domain
+dotbot --sandbox --allow api.mycompany.com
+
+# Persistent config in ~/.dotbotrc
+# { "sandbox": true, "sandboxAllow": ["github", "slack", "messages"] }
+```
+
+**What's blocked by default:**
+
+| Category | Tools | How to unlock |
+|----------|-------|---------------|
+| Filesystem writes | `file_write`, `file_delete`, `file_move`, `folder_create` | Cannot unlock |
+| Arbitrary HTTP | `web_fetch` | `--allow <domain>` |
+| Browser | `browser_navigate` | `--allow <domain>` |
+| Code execution | `run_code` | Always allowed (Node.js permission model) |
+| Messaging | `message_*` | `--allow messages` |
+| Images | `image_*` | `--allow images` |
+| Notifications | `notify_user` | `--allow notifications` |
+| App generation | `app_generate`, `app_validate` | Cannot unlock |
+
+**What's always allowed:** `memory_*`, `web_search`, `grokipedia_search`, `file_read`, `file_list`, `weather_get`, `event_*`, `task_*`, `trigger_*`, `schedule_job`, `list_jobs`, `toggle_job`, `cancel_job`
+
+**Domain presets:** `github`, `slack`, `discord`, `npm`, `pypi`, `jira`, `huggingface`, `docker`, `telegram`
+
 ### Library Usage
 
 ```bash
@@ -139,9 +173,13 @@ for await (const event of agent.chat({
 - **Cerebras** — ultra-fast inference
 - **Ollama** — local models, no API cost
 
+### 🔒 **Sandbox Mode**
+- **Deny-by-default** tool access — no files, code, browser, or messaging
+- **Domain allowlists** — `--allow github`, `--allow slack`
+- **Preset-based** tool unlocking — `--allow messages`, `--allow images`
+
 ### 💾 **Pluggable Storage**
 - **SQLite** — zero dependencies with Node.js 22.5+
-- **MongoDB** — scalable with full-text search
 - **Memory** — in-memory for testing
 
 ### 📊 **Full Audit Trail**
@@ -154,7 +192,7 @@ for await (const event of agent.chat({
 ## CLI Reference
 
 ```
-dotbot v0.25 — AI agent CLI
+dotbot — AI agent CLI
 
 Usage:
   dotbot "message"            One-shot query
@@ -164,6 +202,7 @@ Usage:
   echo "msg" | dotbot         Pipe input from stdin
 
 Commands:
+  models                      List available models from provider
   doctor                      Check environment and configuration
   tools                       List all available tools
   stats                       Show database statistics
@@ -178,10 +217,12 @@ Commands:
   events [--summary]          View audit log
 
 Options:
-  --provider, -p   AI provider: xai, anthropic, openai, ollama (default: xai)
+  --provider, -p   AI provider: xai, anthropic, openai, ollama, mlx_local (default: xai)
   --model, -m      Model name (default: grok-4-1-fast-reasoning)
   --system, -s     Custom system prompt (prepended to default)
   --session        Resume a specific session by ID
+  --sandbox        Restrict tools to safe subset (deny-by-default)
+  --allow          Allow domain/preset in sandbox (github, slack, messages, etc.)
   --db             SQLite database path (default: ./dotbot.db)
   --port           Server port for 'serve' command
   --openai         Enable OpenAI-compatible API endpoints
@@ -195,9 +236,10 @@ Environment Variables:
   ANTHROPIC_API_KEY    API key for Anthropic
   OPENAI_API_KEY       API key for OpenAI
   OLLAMA_BASE_URL      Base URL for Ollama (default: http://localhost:11434)
+  MLX_LOCAL_URL        Base URL for a local MLX-style OpenAI-compatible server (default: http://127.0.0.1:1316/v1)
 
 Config File:
-  ~/.dotbotrc          JSON config for defaults (provider, model, db)
+  ~/.dotbotrc          JSON config for defaults (provider, model, db, sandbox)
 ```
 
 <br />
@@ -323,9 +365,8 @@ await agent.chat({
 | Technology | Purpose |
 |------------|---------|
 | **Node.js 22.5+** | Runtime with built-in SQLite |
-| **Playwright** | Browser automation |
+| **Chrome DevTools Protocol** | Browser automation (zero deps) |
 | **SQLite** | Default storage (zero deps) |
-| **MongoDB** | Scalable storage option |
 
 <br />
 
@@ -334,12 +375,13 @@ await agent.chat({
 ```
 dotbot/
 ├── bin/
-│   └── dotbot.js           # CLI entry point
+│   └── dotbot.js           # CLI entry point (REPL, server, sandbox mode)
 ├── core/
 │   ├── agent.js            # Streaming agent loop
 │   ├── events.js           # SSE event schemas
 │   ├── compaction.js       # Context window management
 │   ├── normalize.js        # Message format conversion
+│   ├── failover.js         # Cross-provider failover
 │   ├── cron_handler.js     # Scheduled job execution
 │   └── trigger_handler.js  # Event-driven triggers
 ├── storage/
@@ -347,9 +389,8 @@ dotbot/
 │   ├── TaskStore.js        # Task interface
 │   ├── CronStore.js        # Job scheduling interface
 │   ├── TriggerStore.js     # Trigger interface
-│   ├── SQLite*.js          # SQLite adapters
-│   └── Mongo*.js           # MongoDB adapters
-├── tools/                  # 47 built-in tools
+│   └── SQLite*.js          # SQLite adapters
+├── tools/                  # 53 built-in tools
 │   ├── memory.js
 │   ├── web.js
 │   ├── browser.js

package/bin/dotbot.js

@@ -115,7 +115,7 @@ Commands:
   events [--summary]          View audit log
 
 Options:
-  --provider, -p   AI provider: xai, anthropic, openai, ollama (default: xai)
+  --provider, -p   AI provider: xai, anthropic, openai, ollama, mlx_local (default: xai)
   --model, -m      Model name (default: grok-4-1-fast-reasoning)
   --system, -s     Custom system prompt (prepended to default)
   --session        Resume a specific session by ID
@@ -134,6 +134,7 @@ Environment Variables:
   ANTHROPIC_API_KEY    API key for Anthropic
   OPENAI_API_KEY       API key for OpenAI
   OLLAMA_BASE_URL      Base URL for Ollama (default: http://localhost:11434)
+  MLX_LOCAL_URL        Base URL for a local MLX-style OpenAI-compatible server (default: http://127.0.0.1:1316/v1)
 
 Config File:
   ~/.dotbotrc                 JSON config for defaults (provider, model, db)
@@ -295,6 +296,13 @@ async function getProviderConfig(providerId) {
     return { ...base, apiUrl: `${baseUrl}/api/chat` };
   }
 
+  // Local OpenAI-compatible servers (mlx_local, etc.) don't use API keys —
+  // they're served from localhost and the apiUrl is already baked into the
+  // provider config (or overridden via env var inside providers.js).
+  if (base.local) {
+    return base;
+  }
+
   const envKey = base.envKey;
   let apiKey = process.env[envKey];
 
@@ -604,50 +612,21 @@ async function initStores(dbPath, verbose = false, customSystemPrompt = '') {
 }
 
 /**
- * Run a single chat message and stream output.
+ * Stream events from an agentLoop iterable to stdout.
+ * Handles thinking markers, text deltas, tool status, and errors.
  *
- * @param {string} message - User message
- * @param {Object} options - CLI options
+ * @param {AsyncIterable<Object>} events - Async iterable of agentLoop events
+ * @returns {Promise<string>} Accumulated assistant text content
  */
-async function runChat(message, options) {
-  const storesObj = await initStores(options.db, options.verbose, options.system);
-  const provider = await getProviderConfig(options.provider);
-
-  let session;
-  let messages;
-
-  if (options.session) {
-    session = await storesObj.sessionStore.getSession(options.session, 'cli-user');
-    if (!session) {
-      console.error(`Error: Session not found: ${options.session}`);
-      process.exit(1);
-    }
-    messages = [...(session.messages || []), { role: 'user', content: message }];
-  } else {
-    session = await storesObj.sessionStore.createSession('cli-user', options.model, options.provider);
-    messages = [{ role: 'user', content: message }];
-  }
-
-  const context = {
-    userID: 'cli-user',
-    sessionId: session.id,
-    providers: { [options.provider]: { apiKey: process.env[AI_PROVIDERS[options.provider]?.envKey] } },
-    ...storesObj,
-  };
-
+async function streamEvents(events) {
   let hasThinkingText = false;
   let thinkingDone = false;
+  let assistantContent = '';
 
   process.stdout.write('Thinking');
   startSpinner();
 
-  for await (const event of agentLoop({
-    model: options.model,
-    messages,
-    tools: getActiveTools(options.sandbox, options.sandboxAllow),
-    provider,
-    context,
-  })) {
+  for await (const event of events) {
     switch (event.type) {
       case 'thinking':
         if (event.text) {
@@ -669,6 +648,7 @@ async function runChat(message, options) {
           thinkingDone = true;
         }
         process.stdout.write(event.text);
+        assistantContent += event.text;
         break;
       case 'tool_start':
         if (!thinkingDone) {
@@ -689,11 +669,55 @@ async function runChat(message, options) {
         stopSpinner('error');
         break;
       case 'error':
+        stopSpinner();
         console.error(`\nError: ${event.error}`);
         break;
     }
   }
 
+  return assistantContent;
+}
+
+/**
+ * Run a single chat message and stream output.
+ *
+ * @param {string} message - User message
+ * @param {Object} options - CLI options
+ */
+async function runChat(message, options) {
+  const storesObj = await initStores(options.db, options.verbose, options.system);
+  const provider = await getProviderConfig(options.provider);
+
+  let session;
+  let messages;
+
+  if (options.session) {
+    session = await storesObj.sessionStore.getSession(options.session, 'cli-user');
+    if (!session) {
+      console.error(`Error: Session not found: ${options.session}`);
+      process.exit(1);
+    }
+    messages = [...(session.messages || []), { role: 'user', content: message }];
+  } else {
+    session = await storesObj.sessionStore.createSession('cli-user', options.model, options.provider);
+    messages = [{ role: 'user', content: message }];
+  }
+
+  const context = {
+    userID: 'cli-user',
+    sessionId: session.id,
+    providers: { [options.provider]: { apiKey: process.env[AI_PROVIDERS[options.provider]?.envKey] } },
+    ...storesObj,
+  };
+
+  await streamEvents(agentLoop({
+    model: options.model,
+    messages,
+    tools: getActiveTools(options.sandbox, options.sandboxAllow),
+    provider,
+    context,
+  }));
+
   process.stdout.write('\n\n');
   process.exit(0);
 }
@@ -864,68 +888,14 @@ async function runRepl(options) {
   const handleMessage = async (text) => {
     messages.push({ role: 'user', content: text });
 
-    let hasThinkingText = false;
-    let thinkingDone = false;
-    let assistantContent = '';
-
-    process.stdout.write('Thinking');
-    startSpinner();
-
     try {
-      for await (const event of agentLoop({
+      const assistantContent = await streamEvents(agentLoop({
         model: options.model,
         messages: [...messages],
         tools: getActiveTools(options.sandbox, options.sandboxAllow),
         provider,
         context,
-      })) {
-        switch (event.type) {
-          case 'thinking':
-            if (event.text) {
-              if (!hasThinkingText) {
-                stopSpinner('');
-                process.stdout.write('\n');
-                hasThinkingText = true;
-              }
-              process.stdout.write(event.text);
-            }
-            break;
-          case 'text_delta':
-            if (!thinkingDone) {
-              if (hasThinkingText) {
-                process.stdout.write('\n...done thinking.\n\n');
-              } else {
-                stopSpinner('');
-              }
-              thinkingDone = true;
-            }
-            process.stdout.write(event.text);
-            assistantContent += event.text;
-            break;
-          case 'tool_start':
-            if (!thinkingDone) {
-              if (hasThinkingText) {
-                process.stdout.write('\n...done thinking.\n\n');
-              } else {
-                stopSpinner('');
-              }
-              thinkingDone = true;
-            }
-            process.stdout.write(`[${event.name}] `);
-            startSpinner();
-            break;
-          case 'tool_result':
-            stopSpinner('done');
-            break;
-          case 'tool_error':
-            stopSpinner('error');
-            break;
-          case 'error':
-            stopSpinner();
-            console.error(`\nError: ${event.error}`);
-            break;
-        }
-      }
+      }));
 
       if (assistantContent) {
         messages.push({ role: 'assistant', content: assistantContent });

package/core/agent.js

@@ -31,7 +31,7 @@ const OLLAMA_BASE = "http://localhost:11434";
  * @param {Array} options.tools - Tool definitions from tools.js
  * @param {AbortSignal} [options.signal] - Optional abort signal
  * @param {Object} [options.provider] - Provider config from AI_PROVIDERS. Defaults to Ollama.
- * @param {Object} [options.context] - Execution context passed to tool execute functions (e.g. databaseManager, dbConfig, userID).
+ * @param {Object} [options.context] - Execution context passed to tool execute functions (e.g. providers, userID).
  * @yields {Object} Stream events for the frontend
  */
 export async function* agentLoop({ model, messages, tools, signal, provider, context, maxTurns }) {
@@ -174,7 +174,7 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
       };
     };
 
-    // Local providers (ollama, dottie_desktop): direct fetch, no failover
+    // Local providers (ollama, mlx_local): direct fetch, no failover
     if (provider.local) {
       const { url, headers, body } = buildAgentRequest(provider);
       response = await fetch(url, { method: "POST", headers, body, signal });
@@ -210,8 +210,9 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
       const result = yield* parseAnthropicStream(response, fullContent, toolCalls, signal, activeProvider.id);
       fullContent = result.fullContent;
       toolCalls = result.toolCalls;
-    } else if (activeProvider.id === "dottie_desktop") {
-      // Dottie Desktop serves local models which may use:
+    } else if (activeProvider.id === "mlx_local") {
+      // Local MLX-style OpenAI-compatible server. Models served this way
+      // may emit output in one of three formats:
       // 1. gpt-oss channel tokens (<|channel|>analysis/final<|message|>)
       // 2. Native reasoning (delta.reasoning from parseOpenAIStream)
       // 3. Plain text (LFM2.5, SmolLM, etc. — no special tokens)
@@ -235,6 +236,21 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
         if (done) {
           fullContent = value.fullContent;
           toolCalls = value.toolCalls;
+          // Flush buffered plain-text responses that never hit the
+          // CHANNEL_DETECT_THRESHOLD. Happens for short greetings and
+          // small-talk from models that don't emit gpt-oss channel tokens
+          // (Gemma 4 E2B, LFM2.5, SmolLM, etc.). Without this flush, the
+          // rawBuffer is silently discarded and the downstream consumer
+          // never receives any text_delta — the UI renders an empty bubble.
+          // Skip if the buffer contains tool call markers so the existing
+          // post-loop parseToolCalls() below can handle them.
+          if (!usesPassthrough && !usesNativeReasoning && !analysisStarted && !finalMarkerFound && rawBuffer.length > 0) {
+            if (!hasToolCallMarkers(rawBuffer)) {
+              const textEvent = { type: "text_delta", text: rawBuffer };
+              validateEvent(textEvent);
+              yield textEvent;
+            }
+          }
           break;
         }
 
@@ -270,7 +286,7 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
         // the model doesn't use gpt-oss format (e.g. LFM2.5, SmolLM).
         // Flush buffer and switch to passthrough for remaining tokens.
         if (!analysisStarted && !finalMarkerFound && rawBuffer.length > CHANNEL_DETECT_THRESHOLD) {
-          console.log("[dottie_desktop] no channel tokens after", rawBuffer.length, "chars — switching to passthrough");
+          console.log("[mlx_local] no channel tokens after", rawBuffer.length, "chars — switching to passthrough");
           usesPassthrough = true;
           const textEvent = { type: "text_delta", text: rawBuffer };
           validateEvent(textEvent);
@@ -285,7 +301,7 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
             if (aIdx !== -1) {
               analysisStarted = true;
               lastThinkingYieldPos = aIdx + ANALYSIS_MARKER.length;
-              console.log("[dottie_desktop] analysis marker found at", aIdx, "| yieldPos:", lastThinkingYieldPos);
+              console.log("[mlx_local] analysis marker found at", aIdx, "| yieldPos:", lastThinkingYieldPos);
             }
           }
 
@@ -295,7 +311,7 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
             if (endIdx !== -1) {
               const chunk = rawBuffer.slice(lastThinkingYieldPos, endIdx);
               if (chunk) {
-                console.log("[dottie_desktop] thinking (final):", chunk.slice(0, 80));
+                console.log("[mlx_local] thinking (final):", chunk.slice(0, 80));
                 const thinkingEvent = {
                   type: "thinking",
                   text: chunk,
@@ -309,7 +325,7 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
             } else {
               const chunk = rawBuffer.slice(lastThinkingYieldPos);
               if (chunk) {
-                console.log("[dottie_desktop] thinking (incr):", chunk.slice(0, 80));
+                console.log("[mlx_local] thinking (incr):", chunk.slice(0, 80));
                 const thinkingEvent = {
                   type: "thinking",
                   text: chunk,
@@ -325,7 +341,7 @@ export async function* agentLoop({ model, messages, tools, signal, provider, con
           // Check for final channel marker
           const fIdx = rawBuffer.indexOf(FINAL_MARKER);
           if (fIdx !== -1) {
-            console.log("[dottie_desktop] final marker found at", fIdx, "| bufLen:", rawBuffer.length);
+            console.log("[mlx_local] final marker found at", fIdx, "| bufLen:", rawBuffer.length);
             finalMarkerFound = true;
             lastFinalYieldPos = fIdx + FINAL_MARKER.length;
             const pending = rawBuffer.slice(lastFinalYieldPos);
@@ -740,8 +756,9 @@ export async function getOllamaStatus() {
 }
 
 /**
- * Check if Dottie Desktop is running and list available models.
- * Uses the OpenAI-compatible /v1/models endpoint.
+ * Check if a local OpenAI-compatible model server is running and list
+ * available models. Defaults to the MLX LM server convention
+ * (http://localhost:1316/v1) and can be overridden with MLX_LOCAL_URL.
  *
  * @returns {Promise<{running: boolean, models: Array<{name: string}>}>}
  */
@@ -765,8 +782,8 @@ function stripGptOssTokens(text) {
   return text.replace(TOKEN_RE, "").trim();
 }
 
-export async function getDottieDesktopStatus() {
-  const baseUrl = (process.env.DOTTIE_DESKTOP_URL || 'http://localhost:1316/v1').replace(/\/v1$/, '');
+export async function getMlxLocalStatus() {
+  const baseUrl = (process.env.MLX_LOCAL_URL || 'http://localhost:1316/v1').replace(/\/v1$/, '');
   try {
     const res = await fetch(`${baseUrl}/v1/models`);
     if (!res.ok) return { running: false, models: [] };

package/core/cdp.js

@@ -184,35 +184,22 @@ export class CDPClient {
     return result.result?.value;
   }
 
-  /**
-   * Get the page title.
-   * @returns {Promise<string>}
-   */
+  /** Get the page title. */
   async getTitle() {
     return this.evaluate('document.title');
   }
 
-  /**
-   * Get the current URL.
-   * @returns {Promise<string>}
-   */
+  /** Get the current URL. */
   async getUrl() {
     return this.evaluate('window.location.href');
   }
 
-  /**
-   * Get text content of the page body.
-   * @returns {Promise<string>}
-   */
+  /** Get text content of the page body. */
   async getBodyText() {
     return this.evaluate('document.body?.innerText || ""');
   }
 
-  /**
-   * Get text content of an element by CSS selector.
-   * @param {string} selector - CSS selector
-   * @returns {Promise<string>}
-   */
+  /** Get text content of an element by CSS selector. */
   async getText(selector) {
     const escaped = selector.replace(/"/g, '\\"');
     return this.evaluate(`document.querySelector("${escaped}")?.innerText || ""`);
@@ -308,26 +295,6 @@ export class CDPClient {
     });
   }
 
-  /**
-   * Click an element by CSS selector.
-   * @param {string} selector - CSS selector
-   */
-  async clickSelector(selector) {
-    const el = await this.querySelector(selector);
-    if (!el) throw new Error(`Element not found: ${selector}`);
-    await this.click(el.x, el.y);
-  }
-
-  /**
-   * Click an element by visible text.
-   * @param {string} text - Text content to find
-   */
-  async clickText(text) {
-    const el = await this.getByText(text);
-    if (!el) throw new Error(`Element with text "${text}" not found`);
-    await this.click(el.x, el.y);
-  }
-
   /**
    * Type text character by character.
    * @param {string} text - Text to type
@@ -453,9 +420,7 @@ export class CDPClient {
     });
   }
 
-  /**
-   * Close the connection.
-   */
+  /** Close the CDP connection. */
   close() {
     if (this.ws) {
       this.ws.close();
@@ -490,24 +455,6 @@ export class CDPClient {
     throw lastError;
   }
 
-  /**
-   * Wait for an element to appear in the DOM.
-   * @param {string} selector - CSS selector
-   * @param {Object} options - Wait options
-   * @param {number} options.timeout - Timeout in ms (default: 5000)
-   * @param {number} options.interval - Poll interval in ms (default: 100)
-   * @returns {Promise<{x: number, y: number, nodeId: number}>} Element info
-   */
-  async waitForSelector(selector, { timeout = 5000, interval = 100 } = {}) {
-    const start = Date.now();
-    while (Date.now() - start < timeout) {
-      const el = await this.querySelector(selector);
-      if (el) return el;
-      await new Promise(r => setTimeout(r, interval));
-    }
-    throw new Error(`Timeout waiting for selector: ${selector}`);
-  }
-
   /**
    * Wait for network to be idle (no requests for a period).
    * @param {Object} options - Wait options

package/core/compaction.js

@@ -12,7 +12,7 @@ const CONTEXT_LIMITS = {
   openai: 120000,
   xai: 120000,
   ollama: 6000,
-  dottie_desktop: 6000,
+  mlx_local: 6000,
 };
 
 /** Number of recent messages to always preserve verbatim. */