1. Project Overview
Nanobrowser is an open-source Chrome browser extension that uses multi-agent AI to automate web browsing tasks directly inside the user's own browser, offering a free, privacy-preserving alternative to paid AI browser-automation services.
2. Background & Positioning
Nanobrowser was created as a response to commercial "AI browser agent" products (such as OpenAI Operator) that charge substantial monthly subscription fees while requiring users to hand over their browsing session and credentials to a remote cloud service. The project's core mission is to make autonomous web automation accessible, transparent, and cost-free by running entirely as a local Chrome extension.
Compared to similar AI browsing agents, Nanobrowser differentiates itself in three ways:
- Multi-agent by design โ instead of a single model trying to plan and act at once, it splits reasoning ("what to do next") from execution ("how to click/type/navigate") between dedicated agents.
- Bring-your-own-model โ users are not locked into one AI vendor; they can mix and match providers and even assign a cheaper model to one agent and a stronger model to another to control cost and quality.
- Local-first privacy โ all browsing, authentication state, and task execution stay in the user's own browser session rather than being proxied through a third-party cloud automation service.
3. Feature Categories
๐ค Multi-Agent Automation โ 2 core agents working together (Planner and Navigator) to break down and execute tasks.
- Planner agent for reasoning, task decomposition, and strategy
- Navigator agent for concrete browser actions (click, type, scroll, navigate)
- Agent hand-off loop that re-plans when a step fails or the page changes
- Task automation across arbitrary websites, not a fixed set of integrations
๐ฌ Interactive Side Panel โ a persistent in-browser UI for controlling and monitoring agents.
- Real-time status updates while agents are working
- Chat-style interface for issuing tasks and follow-up instructions
- Conversation history so past tasks and results can be revisited
- Contextual follow-ups that continue an existing browsing session
๐ Flexible LLM Integration โ supports 7+ LLM providers so users can choose per-agent models.
- OpenAI (e.g., GPT-4o)
- Anthropic Claude (e.g., Claude Sonnet, Claude Haiku)
- Google Gemini
- Groq, Cerebras (fast inference providers)
- Ollama for fully local, self-hosted open-weight models
- Any custom OpenAI-compatible endpoint
๐ Privacy & Local Execution โ keeps sensitive data out of third-party clouds.
- No mandatory subscription or remote automation backend
- Credentials and session cookies remain inside the user's own browser
- Users choose which LLM provider (including fully local Ollama models) receives task data
4. Key Highlights
- Multi-agent Planner/Navigator architecture โ separating strategic reasoning from tactical execution produces more reliable multi-step task completion than a single monolithic agent loop.
- Per-agent model assignment โ a stronger, more expensive model can drive planning while a cheaper, faster model handles routine navigation steps, balancing cost and capability.
- Zero-cost alternative to paid browser agents โ avoids the ~$200/month price point of commercial operator-style products by running as a free browser extension.
- Wide LLM provider support โ works with major hosted APIs (OpenAI, Anthropic, Gemini, Groq, Cerebras) as well as fully local models via Ollama.
- Chrome-native side panel UX โ task control and history live in the browser's own side panel rather than a separate app or web dashboard.
- Open governance โ Apache-2.0 licensed with an active Discord community and public contribution process.
5. Use Cases by Role
- General developers: automate repetitive browser research or QA tasks (e.g., "extract the top headlines from a tech news site" or "find trending repositories on GitHub matching a topic") without writing custom scraping scripts.
- Project managers / non-technical users: delegate comparative research tasks, such as shopping comparisons across multiple sites by price, features, and specifications, through natural-language chat instructions instead of manual browsing.
6. Getting Started
Find what you need โ browse the README and Discord community for example prompts and recommended model configurations:
https://github.com/nanobrowser/nanobrowser
Install / integrate โ install from the Chrome Web Store, or build from source for the latest development version:
git clone https://github.com/nanobrowser/nanobrowser.git
cd nanobrowser
pnpm install
pnpm build
Then load the dist output as an unpacked extension via chrome://extensions/ with Developer Mode enabled. Requires Node.js v22.12.0+ and pnpm v9.15.1+.
Contribute โ read the contribution guide, then open a pull request or share prompts/feedback via Discord:
CONTRIBUTING.md
Security issues should be reported privately through GitHub Security Advisories rather than public issues.
7. Project Structure
nanobrowser/
โโโ chrome-extension/ # Extension shell: manifest, background script, build config
โ โโโ manifest.js
โ โโโ src/
โโโ pages/ # UI surfaces of the extension
โ โโโ content/ # Content scripts injected into web pages
โ โโโ options/ # Extension options/settings page
โ โโโ side-panel/ # Main chat/side-panel UI where agents run
โโโ packages/ # Shared internal packages (monorepo)
โ โโโ shared/ # Shared types/utilities across the extension
โ โโโ storage/ # Local storage/state management
โ โโโ i18n/ # Internationalization
โ โโโ ui/ # Shared UI components
โ โโโ vite-config/ # Shared build tooling
โโโ CONTRIBUTING.md
โโโ SECURITY.md
โโโ package.json
chrome-extension/ holds the browser extension entry point and manifest; pages/side-panel/ is where the Planner/Navigator chat experience lives; packages/ contains the shared libraries reused across the extension's different UI surfaces.
8. Related Ecosystem
- Chrome Extensions platform โ Nanobrowser is built on Chrome's Manifest V3 extension APIs and the side panel API.
- LLM providers it integrates with: OpenAI, Anthropic Claude, Google Gemini, Groq, Cerebras, and Ollama for local open-weight models (e.g., Qwen, Falcon, Mistral).
- pnpm / Turborepo tooling โ the codebase is organized as a pnpm workspace monorepo built with Turborepo-style task orchestration.
- Complementary category: sits alongside other AI browser-automation and computer-use agent projects, distinguished by its free, extension-based, multi-agent approach.
9. License
Licensed under Apache License 2.0.
- โ Free to use, modify, and distribute, including for commercial purposes
- โ Can be forked and built into derivative browser extensions with attribution
- โ No warranty is provided; the project disclaims liability for misuse
- โน๏ธ The project explicitly disclaims responsibility for any cryptocurrency-related derivative projects built on top of it
10. FAQ
Q: Do I need to pay for a subscription to use Nanobrowser?
A: No. Nanobrowser itself is free and open-source; you only pay for usage of whichever LLM API you choose to connect (or nothing at all if you run local models via Ollama).
Q: Which LLM providers are supported?
A: OpenAI, Anthropic Claude, Google Gemini, Groq, Cerebras, and self-hosted models through Ollama, plus any custom OpenAI-compatible API endpoint.
Q: Is my browsing data sent to Nanobrowser's own servers?
A: No. Nanobrowser runs locally as a Chrome extension; task execution happens in your own browser session, and data is only sent to the LLM provider you configure.
Q: How do I install it from source?
A: Clone the repository, install dependencies with pnpm install, run pnpm build, then load the build output as an unpacked extension at chrome://extensions/.
Q: Can I use different models for planning versus browsing actions?
A: Yes. Nanobrowser's Planner and Navigator agents can each be assigned a different LLM model, so you can pair a stronger model for reasoning with a cheaper/faster model for routine navigation.
11. Quick Links
- Repository: https://github.com/nanobrowser/nanobrowser
- Contributing guide: https://github.com/nanobrowser/nanobrowser/blob/master/CONTRIBUTING.md
- Security policy: https://github.com/nanobrowser/nanobrowser/blob/master/SECURITY.md
- Privacy policy: https://github.com/nanobrowser/nanobrowser/blob/master/PRIVACY.md
12. Summary
Nanobrowser brings multi-agent AI automation to everyday browsing by running entirely as a free, open-source Chrome extension rather than a paid cloud service. It is best suited for developers, researchers, and general users who want to automate repetitive web research and browsing tasks while keeping control over which AI model handles their data and how much it costs.