agent-zero

Agent Zero is a framework for building and running AI agents that can use a real Linux computer to get work done. Instead of being a single-purpose chatbot or a fixed list of skills, it gives a large language model a working environment: a terminal, a file system, a web browser, memory of past projects, and the ability to write and run code. The agent figures out how to use those pieces to solve whatever task you give it, and can build new tools for itself along the way.

The agent runs inside a Kali Linux environment (a Linux distribution) packaged in Docker, with a Web UI on top. From there it can inspect and edit files, install software, run scripts, search the web, and adapt its plan as a task evolves. A right-side panel called the Canvas makes the work visible: you can watch the browser session, see documents being edited, inspect workspace history, and step in before a mistake compounds. A native Browser lets the agent click, type, take screenshots, and use Chrome extensions, while an annotation mode lets you point at parts of a page and leave instructions. A Desktop surface adds a full Linux desktop with LibreOffice (Writer, Calc, Impress), so you and the agent can edit the same documents side by side. A separate A0 CLI Connector bridges a running Agent Zero instance to your host machine when you want it to work outside the container.

You bring your own large language model, connecting through providers or, with the new OAuth flow, through an OpenAI Codex plan. Agent Zero would suit someone who wants an autonomous, transparent assistant that can take open-ended goals and operate a real computer rather than just generate text. The project is written in Python.