Coding & Development

Nanobrowser

Open-source Chrome extension for multi-agent AI web automation

Free and open source (Apache 2.0); no subscription and no paid tier listed on the site. You pay your chosen LLM provider directly for API usage, or nothing if you run local models (e.g. via Ollama).
Visit Nanobrowser →
Pricing
Free and open source (Apache 2.0); no subscription and no paid tier listed on the site. You pay your chosen LLM provider directly for API usage, or nothing if you run local models (e.g. via Ollama).
Best for
Developers, tinkerers and privacy-minded users comfortable with API keys who want an open-source browser agent.
Official site
nanobrowser.ai
Last checked
2026-10-27

Nanobrowser suits developers and technically comfortable users who want a free, inspectable AI browser agent running inside their own signed-in Chrome or Edge. Skip it if you want a polished plug-and-play product, use Firefox or Safari, or will not manage API keys and costs. It is a side-panel extension: you describe a task and it clicks, types and reads pages in sites you are already logged into, using a Planner and a Navigator agent that can each be given a different model. The site lists OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter, xAI Grok, Mistral and local models, plus OpenAI-compatible endpoints. The software is free, but model usage is not, and costs depend on your provider and how many steps tasks take. The default Ask mode prompts before submitting, sending, deleting or uploading, though the site warns it may not catch every such step. The GitHub repo shows about 13.8k stars and 1.4k forks (the site shows 14.2k and 1.5k).

Best for

Developers, tinkerers and privacy-minded users comfortable with API keys who want an open-source browser agent. Not for Firefox/Safari users (only Chrome and Edge are officially supported), non-technical users wanting zero setup, or anyone who needs vendor support or guaranteed reliability on sensitive actions.

Key features

01

Multi-agent architecture

A Planner agent handles high-level reasoning and task breakdown while a Navigator agent executes concrete browser actions (clicking, typing, scrolling); the Planner can dynamically re-instruct the Navigator when it hits obstacles mid-task.

02

Per-agent model assignment

Users can assign a different LLM to the Planner versus the Navigator, letting them balance cost and capability (e.g., a stronger reasoning model for planning, a cheaper model for routine navigation steps).

03

Broad LLM provider support

Works with OpenAI, Anthropic, Gemini, Groq, Cerebras, Llama, Ollama for local models, and any custom OpenAI-compatible API endpoint.

04

Interactive side panel

A chat-style sidebar inside the browser shows real-time status updates as agents work, and lets users watch progress rather than treating automation as a black box.

05

Follow-up questions

After a task completes, users can ask contextual follow-up questions about the results without starting a new session from scratch.

06

Conversation history

Past agent interactions and task runs are saved and accessible, making it easier to revisit or reuse previous automations.

07

Local-first, privacy-focused execution

The extension runs entirely client-side in the browser; no browsing data or credentials are sent to a Nanobrowser-operated cloud service.

08

Manual and store installation paths

Available via the Chrome Web Store for stable releases, or as a manually loaded unpacked extension from GitHub release zips for access to the newest features ahead of store review.

Pricing breakdown

Free & Open Source

$0
No subscription; Apache 2.0 licensed, self-hosted as a browser extension
  • Full multi-agent automation feature set with no paywalled tiers
  • Bring your own LLM API key — actual cost is whatever the connected provider (OpenAI, Anthropic, Gemini, Groq, Cerebras, etc.) charges for usage
  • Zero ongoing cost when paired with local models via Ollama
  • Source code, issue tracker, and Discussions fully open on GitHub
  • Optional voluntary support via GitHub Sponsors — not required to use the extension

Pros and cons

Pros

  • No subscription cost at all: the software itself is free, so the only ongoing expense is whatever LLM API usage the user chooses to connect, which can be minimized or eliminated by running local models through Ollama.
  • Splitting the Planner and Navigator roles across different models gives users fine-grained control over the cost/performance tradeoff, something closed agent products don't expose.
  • Because everything executes inside the user's own Chrome or Edge session, there's no separate cloud agent seeing or storing browsing activity, which is a meaningful privacy improvement over hosted browser-agent services.
  • The Apache 2.0 license and public GitHub repository mean the automation logic is fully auditable rather than a black box, which matters for users automating sensitive or authenticated web sessions.
  • An active Discord and GitHub Discussions community shares prompt patterns, local-model benchmarks, and configuration tips, which shortens the learning curve for tuning agent behavior.
  • Manual installation from GitHub releases gives early access to new capabilities ahead of the (sometimes delayed) Chrome Web Store review process.

Cons

  • Browser support is narrow: Chrome and Edge are officially supported, while Firefox, Safari, and other Chromium-based browsers like Opera or Arc are explicitly unsupported, which rules it out for non-Chromium users.
  • There's no managed backend or default model included, so a new user has to obtain and configure their own API keys before the extension does anything useful — it isn't a plug-and-play experience.
  • Using cost-effective model configurations (e.g., cheaper Planner/Navigator pairings) is documented by the project itself as producing less stable output and requiring more iterations on complex tasks.
  • Local models require noticeably more careful, explicit prompt engineering than cloud models, according to the project's own guidance, which adds friction for non-technical users trying to run it fully offline.
  • As a community open-source project without a company behind it, support depends on volunteer maintainers and Discord/GitHub responsiveness rather than an SLA-backed help desk.
  • Real-world reliability of any browser-automation agent depends heavily on how frequently target websites change their DOM/layout, and Nanobrowser's task success is bounded by the underlying LLM's web-navigation reasoning quality.

Frequently asked questions

Is Nanobrowser really free?

The extension is free under Apache 2.0 with no paid tier listed. You pay your LLM provider for API usage, or $0 with local models.

Which browsers work?

Chrome and Edge are officially supported. Firefox, Safari and other Chromium forks such as Opera and Arc are not supported, though it may work on them.

Which models can I use?

Per the site: OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter, xAI Grok, Mistral, local models, and any OpenAI-compatible service.

Is it safe to let it act on logged-in sites?

It runs in your own browser session. Ask mode confirms before submitting, sending, deleting or uploading, but the site says it may not catch every case, so supervise important tasks.

Does it work out of the box?

No. You must supply a model provider key or a local model first.

Ready to try Nanobrowser?

Head to the official site to explore pricing and start a free trial where available.

Visit Nanobrowser →