What it does

Agent Zero is a framework for agent work that goes beyond chat. It runs in Docker and gives the agent its own Linux desktop in a side canvas, where it can use desktop software, terminals and files while you watch and step in. A built-in browser lets the agent open, read, click and type on pages, and an annotate mode lets you click page elements to inspect or change them, lift a component into your own stack, or leave review comments for the agent to act on. Documents can be edited live by you and the agent together, including a Markdown editor and LibreOffice Writer, Calc and Impress files. Projects keep files, instructions, memory, secrets and model choices separate per piece of work, skills load on demand, and agents can create subordinate agents to split up tasks. It can be extended through a Plugin Hub with 100+ community plugins, custom prompts and tools, MCP servers and A2A connectors. The A0 CLI Connector links a running instance to another machine so the agent can work in real local repositories, and Time Travel keeps snapshot history of the agent's workspace so changes can be inspected or reverted. You can install through the A0 Launcher desktop app, a terminal installer or a docker run command, then configure your LLM provider in the Web UI. The page advises running it inside Docker or another isolated environment.

Capabilities

Linux desktop in the Canvas that both the agent and the user can control
Built-in browser with element annotation
Live co-editing of Markdown, spreadsheet and presentation files
Plugin Hub with 100+ community plugins, plus MCP and A2A support
Per-project isolation of files, memory, secrets and model presets
Subordinate agents and Time Travel snapshots of the workspace

When to use it

You want an agent that can drive real GUI software and a browser, not just chat
You want an agent to work in your local repositories while it runs isolated in Docker
You want to inspect and edit the prompts and tools the agent uses
You need client or project work kept separate