BroClaw executes multi-step agentic tasks right in your browser — search the web, open pages, fix code, save full projects — without leaving your machine.
No signup · No subscription · API keys never leave your device · Auto-saves chats
BroClaw runs until the job is done. No fixed step limits — the AI decides what to do next: search, read, write, test, iterate — until your task is complete.
Connects to Brave Search, SerpAPI, or your own SearXNG instance. Opens any URL in a sandboxed reader and extracts clean content automatically for the agent to reason over.
Generates full multi-file projects — HTML, CSS, JS, Python, JSON — and saves them directly to your local disk via the File System Access API. Preview instantly in-app.
Claude's extended thinking and o3/o4-mini reasoning tokens are fully streamed and displayed in real time so you can follow the agent's logic step by step.
Runs entirely in your browser. Your AI provider API keys are AES-GCM encrypted in localStorage and never sent to BroClaw servers. Your chats and files stay on your own machine.
Switch between Anthropic, OpenAI, Google Gemini, NVIDIA NIM, or any Ollama/LM Studio endpoint — all from one interface, mid-task with no restart required.
Save custom Skills that teach the agent your domain, tone, and preferred output formats — reusable across any task.
Preview generated websites and apps directly inside BroClaw. Edit, re-run the agent, and see changes instantly without switching between tools.
Connect to any local Ollama instance, LM Studio, or vLLM endpoint. Full model routing with automatic fallback — switch providers mid-task with no restart.
Connect any Model Context Protocol server over Streamable HTTP — local or cloud-hosted, including stdio servers wrapped by mcp-proxy. Add auth headers per server and the agent gets new tools instantly.
Sign in once to Google Workspace for Gmail, Drive, and Calendar, or paste a token to connect GitHub, Notion, and Slack — the agent can then read and act on your real accounts.
| Model | Speed | VRAM / RAM | Best for |
|---|---|---|---|
| SmolLM2 1.7B | ⚡⚡⚡ Ultra | 2 GB | Quick answers |
| Qwen3 0.6B | ⚡⚡⚡ Ultra | 1 GB | Fast tasks |
| Qwen3 1.7B | ⚡⚡ Fast | 2 GB | Coding |
| Gemma 3 1B | ⚡⚡ Fast | 1.5 GB | General use |
| Llama 3.2 3B | ⚡ Balanced | 2.5 GB | Instructions |
| Phi-4 Mini | ⚡ Balanced | 3 GB | Reasoning |
| Gemma 3 4B | 🔴 Powerful | 4 GB | Quality output |
| Qwen3 4B | 🔴 Powerful | 5 GB | Complex tasks |
Downloaded once and cached in your browser — no re-download needed.
No signup. No credit card. Open the app and start in 30 seconds.
🦞 Open BroClawDesktop Chrome 113+ or Edge 113+ required for WebGPU local models