Nanobrowser is an open-source Chrome extension that automates web tasks with a multi-agent system running inside your browser. You describe a task in a side-panel chat, such as extracting the top headlines from a news site or finding a product under a price limit, and the agents carry it out across websites. It positions itself as a free alternative to OpenAI Operator.
Two agents divide the work: a Planner reasons about the task and corrects course when it hits an obstacle, and a Navigator acts on the page. You assign a different model to each agent to balance capability and cost, for example a larger Claude model as Planner and a smaller Claude or Gemini model as Navigator.
Nanobrowser uses your own API keys, so you pay only the providers you call, and credentials stay in the local browser rather than a cloud service. Supported providers are OpenAI, Anthropic, Gemini, DeepSeek, Grok, Azure OpenAI, OpenRouter, Ollama, and custom OpenAI-compatible endpoints. Local models such as Qwen3-30B-A3B, Qwen 2.5 Coder, Falcon3, and Mistral Small run through Ollama with no API cost, and need more explicit, step-by-step prompts.
Chrome and Edge are officially supported. Firefox, Safari, and other Chromium variants are not. The extension builds on Browser Use, Puppeteer, WXT, and the Vercel AI SDK.
Features
- Multi-agent system: a Planner and a Navigator collaborate on multi-step web workflows
- Per-agent models: assign a different LLM to each agent
- Side-panel chat: a chat interface with real-time status updates
- Follow-up questions: ask contextual questions about completed tasks
- Conversation history: access and manage past agent interactions
- Provider choice: OpenAI, Anthropic, Gemini, DeepSeek, Grok, Azure OpenAI, OpenRouter, Ollama, and custom OpenAI-compatible providers
- Local models: run models through Ollama so no data leaves your machine
- Local execution: runs entirely in the browser with your own API keys
