An open-source autonomous agent that runs deep research on web and local documents using any LLM provider, producing cited reports.
agent0ai/agent-zero
An open agent framework where the AI agent gets a full Dockerized Linux desktop, a browser, live document editing, and a bridge to your host machine.
- Stars
- 19,510
- Forks
- 3,826
- Open issues
- 134
- Last commit
- 4 days ago
What is it?
Agent Zero is an open agent framework for tasks that go beyond chat. The agent runs inside a Dockerized Linux environment with an XFCE desktop, so it can operate real GUI applications, terminals and files, and you can watch or take over the desktop at any time.
The built-in browser supports DOM annotation: you click page elements to have the agent inspect, change, copy or review them. A Canvas editor lets you and the agent edit Markdown, LibreOffice Writer, Calc and Impress files together. Projects keep files, instructions, secrets, memory and model presets separate per workspace.
The A0 CLI Connector links a running instance to your host machine so the agent can work in local repositories. The project also offers a Plugin Hub with community extensions, MCP and A2A support, customizable prompts and tools, and Time Travel snapshots for reviewing and reverting workspace changes.
Key features
- 01Runs inside a Dockerized Linux XFCE desktop where the agent can use GUI applications.
- 02Built-in browser with DOM annotation for inspecting, changing and commenting on page elements.
- 03Live Markdown and LibreOffice document editing shared between user and agent.
- 04Projects isolate files, instructions, secrets, memory and model presets.
- 05A0 CLI Connector bridges the agent to local host repositories.
- 06Plugin Hub, MCP and A2A support, with editable prompts and tools.
Use cases
- Driving desktop software such as Blender or LibreOffice without an API.
- Annotating a web page and having the agent re-implement a UI component in your own stack.
- Working on local repositories from a Docker-isolated agent through the A0 CLI.
- Producing editable spreadsheets, documents and presentations with an agent.
Topics
FAQ
- What is agent-zero?
- An open agent framework where the AI agent gets a full Dockerized Linux desktop, a browser, live document editing, and a bridge to your host machine. Agent Zero is an open agent framework for tasks that go beyond chat. The agent runs inside a Dockerized Linux environment with an XFCE desktop, so it can operate real GUI applications, terminals and files, and you can watch or take over the desktop at any time.
- Is agent-zero free to use? What license does it have?
- agent-zero is open source under the Other license. Check the license terms before using it in commercial products.
- Is agent-zero still maintained?
- Actively maintained. The last commit to agent-zero was on Oct 14, 2026.
- What can I use agent-zero for?
- Driving desktop software such as Blender or LibreOffice without an API. Annotating a web page and having the agent re-implement a UI component in your own stack. Working on local repositories from a Docker-isolated agent through the A0 CLI. Producing editable spreadsheets, documents and presentations with an agent.
- Where can I get agent-zero?
- The source code and installation instructions are on GitHub: https://github.com/agent0ai/agent-zero
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
AutoGPT is an open-source platform for building, running, and scheduling AI agents that complete multi-step workflows from plain-English descriptions or a visual builder.
An agentic coding tool that runs in your terminal, reads your codebase, and handles coding tasks, code explanations, and git workflows through natural language.
An MCP server that gives coding agents IDE-level symbol search, semantic editing and refactoring over language servers or JetBrains IDEs.
A collection of starter projects showing how to build deployable applications on the Claude API, from support agents to browser and computer-use agents.
An open-source AI coworker for your desktop that completes finished tasks like security reviews, documents, and inbox triage, with approval gates and a sandbox.