Agent Zero

Agent Zero is an open agent framework from Agent Zero, s.r.o. It runs an autonomous agent inside a Dockerized Linux environment with a Web UI, a built-in browser, a desktop, and tools the agent can use and create. The latest release listed on GitHub is v2.14, and the project is distributed under an MIT license.

What It Is

Agent Zero is a framework for tasks that need more than chat. The agent works in a real Linux environment inside Docker, where it can use a terminal, files, a browser, and desktop applications. Users configure an LLM provider in the Web UI and then give the agent concrete tasks. The project says prompts, tools, plugins, skills, and settings are inspectable and editable.

Workspace, Browser, and Documents

The Canvas hosts an XFCE Linux desktop session in the container, so the agent can operate GUI software while the user watches and intervenes. The built-in Browser supports an Annotate mode, where clicking page elements lets users change, inspect, lift, or comment on them for the agent to act on. A Markdown editor and LibreOffice Writer, Calc, and Impress integration let the agent and the user edit the same documents live. Time Travel adds snapshot history, diffs, and revert for the /a0/usr workspace.

Projects, Plugins, and Multi-Agent Work

Projects isolate instructions, memory, secrets, knowledge, repositories, and model-preset choices. Skills, Agent Profiles, and Model Presets shape how the agent behaves in a given chat. Agents can create subordinate agents to delegate research, coding, or analysis. The built-in Plugin Hub lists community plugins, which the README says number more than 100, and users can publish their own. MCP, A2A, custom tools, and custom prompts are also supported.

Setup Path

The recommended install is the desktop A0 Launcher for macOS, Linux, and Windows, which checks Docker and manages Instances. Alternatives are the A0 Install shell and PowerShell scripts or a direct docker run command. The A0 CLI Connector links a running instance to the host terminal, files, and browser, so the agent can work in local repositories. The README recommends isolating the agent in Docker and granting host access only to trusted machines.