Multi-agent web tasks without the $200 cover charge

What it does

Nanobrowser is a Chrome extension that automates web tasks through a multi-agent system running directly in your browser. You supply the LLM API keys, and specialized agents—Planner, Navigator, and others—collaborate to browse, click, and extract data on your behalf. It keeps credentials local and positions itself as a free alternative to OpenAI Operator.

The interesting bit

The project treats agent specialization as a first-class feature, letting you assign different models to different roles—put a heavy reasoner on planning and a cheap fast model on navigation. It is also largely an assembly of existing open-source tools—Browser Use, Puppeteer, and LangChain—which is a pragmatic way to avoid rebuilding the browser-automation stack from scratch.

Key highlights

• Multi-agent architecture with role-specific models (Planner, Navigator, etc.)
• Bring-your-own-key support for OpenAI, Anthropic, Gemini, Ollama, Groq, Cerebras, and custom OpenAI-compatible endpoints
• Runs locally in the browser; credentials stay on your machine
• Side-panel chat interface with conversation history and follow-up questions
• Officially supports Chrome and Edge only; Firefox and Safari are out of scope

Caveats

• The README admits cost-effective model configurations may produce “less stable outputs” and require more iterations for complex tasks
• Local models demand more specific, cleaner prompts and struggle with high-level ambiguous commands

Verdict

Worth a look if you want browser automation without surrendering your data or signing up for another SaaS subscription. Skip it if you need cross-browser support or expect turnkey reliability from smaller local models.