AIHawk: a self-hosted AI agent with its own Linux VM and a patched Firefox browser
Open-source AI browser agent for web automation: a web browsing agent and computer-use agent in plain English. Browser MCP for Claude Code and Gemini CLI.
At a glance
- What is it?
- AIHawk's README describes a self-hosted agent, called a Dot, that gets its own Linux VM and a patched Firefox browser. It needs Node 24+, Go 1.25+, Git, x86-64 hardware virtualization and an OpenRouter key, and it is MIT licensed.
- Who is it for?
- Adopt it if you want a self-hosted agent with its own Linux VM and a browser whose identities keep their own cookies and logins, and you can meet the requirements: Node 24+, Go 1.25+, Git, x86-64 hardware virtualization and an OpenRouter key. Do not adopt it if you cannot run a hardware-accelerated VM, or if you are unwilling to let an agent act with your accounts and from your connection, which the README's disclaimer warns about.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 13, 2026, and from our analysis. They are not legal advice.
Star history
- feder-cr/AIHawk32,510 ★ · +1,813
| Repository | 2026-09-13 | 2026-11-13 |
|---|---|---|
| feder-cr/AIHawk | 30.7k | 32.5k |
1,813 Stars in the last 61 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The gap AIHawk fills: pages that only exist after a click
Most LLM tooling reaches the web through an HTTP fetch and a text extraction step, which cannot handle pages whose content only appears after interaction. The README presents the project as an open-source, self-hosted agent for anything that needs a computer and a person's judgement. One of its example prompts asks the agent to check one-way fares from Milan to Lisbon every morning for the next two weeks, write them to ~/workspace/fares.csv and report the cheapest day.
Two entry points, one package, and the engine underneath
The README's quickstart is not a one-line package install. It needs Node 24+, Go 1.25+, Git, hardware virtualization (x86-64) and an OpenRouter key, then builds the CLI from a clone of the invisible_dots repository and runs a setup step followed by a server step. A Dot is the README's name for an agent; it has its own hardware-accelerated VM with a Linux desktop and a persistent disk, which sleeps when idle and wakes for the next message, task or automation.
Install: uv, a fetched engine, and a key you should not type on the command line
On Windows the README installs Node.js LTS, Go and Git with winget; on Ubuntu 24.04 it uses the NodeSource setup script, apt and snap. Both then clone https://github.com/feder-cr/invisible_dots, run npm ci, build the CLI workspace, and run the CLI's setup --all and server commands. setup --all installs QEMU and turns on the accelerator (KVM, or the Windows Hypervisor Platform), then builds or downloads the Dot's images, and can be run again after a restart. Afterwards you open http://127.0.0.2:3000, paste your OpenRouter key and create your first Dot.
The pinned binary is a constraint, not a convenience
The README's feature table lists what a Dot has: a VM with a persistent disk, a browser where each identity keeps its own cookies and logins, memory and skills it writes in its home folder, MCP servers you declare, permissions set per tool as allow, ask or deny (an ask waits in your Inbox), tasks and automations, and several ways to reach it: web UI, command line, HTTP API, Telegram, and WhatsApp as an opt-in. It also says a restart or a kill -9 keeps every accepted message and task.
Where it is the wrong tool
The README's disclaimer says the project is provided as-is, with no warranties, to be used at your own risk and in compliance with the laws of your jurisdiction. It adds that a Dot acts with your accounts and from your connection, so you should respect the terms of the sites it visits and their robots.txt. The permission model in the feature table, where every tool is allow, ask or deny, is the control to use before letting an agent act on sites where a wrong click has consequences.
The alternative: write the Playwright script yourself
The README links the pieces of a Dot's browser: invisible_playwright_mcp, built on invisible_playwright and invisible_core, and says the engine is a fork of nanobot. The browser itself is described as Firefox patched in C++ with the fingerprint set inside the engine. The README does not compare AIHawk with writing Playwright scripts yourself.
Maintenance, licence, and what the version history implies
The README says the project is MIT licensed, see LICENSE, and points to THIRD_PARTY_NOTICES.md for third-party components and their licences. The repository metadata also reports MIT. The recent releases listed are named golden images, the newest being golden-9c2b3c5b93ef on 2026-10-31, and the last push was on 2026-10-04.
Editorial conclusion
Adopt it if you want a self-hosted agent with its own Linux VM and a browser whose identities keep their own cookies and logins, and you can meet the requirements: Node 24+, Go 1.25+, Git, x86-64 hardware virtualization and an OpenRouter key. Do not adopt it if you cannot run a hardware-accelerated VM, or if you are unwilling to let an agent act with your accounts and from your connection, which the README's disclaimer warns about. The README states the project is MIT licensed and provided as-is, with no warranties.
Frequently asked questions
What is an AIHawk?
An open source, MIT-licensed, self-hosted AI browser agent. Its README describes each agent, a Dot, as having a hardware-accelerated Linux VM with a persistent disk, a patched Firefox browser, memory and skills, and permissions you set. The repository description also presents it as a browser MCP for Claude Code and Gemini CLI.
how to use aihawk
Following the README quickstart, install Node 24+, Go 1.25+ and Git, clone https://github.com/feder-cr/invisible_dots, run npm ci and build the CLI workspace, then run the CLI's setup --all and server commands. Open http://127.0.0.2:3000, paste your OpenRouter key and create your first Dot.
Do AI job searches really work?
The README does not mention job searches. Its example prompts include opening a browser identity and reading a page's title, checking one-way fares from Milan to Lisbon every morning, and logging in to a shop to download invoices.
Official sources
Was this useful?