Agent Zero

Agent Zero is an open-source agent framework that gives an AI agent a full Linux computer to work with: a Dockerized Linux desktop, a real browser, a file system and a terminal, all driven from a web UI. Instead of being locked into fixed tools, the agent writes and runs its own code and tools dynamically to complete tasks. The project has gathered roughly 19,500 stars on GitHub and remains under very active development in 2026.

Overview

Agent Zero runs inside Docker and presents a live Canvas where you can watch the agent operate an XFCE desktop, annotate web pages in the built-in browser to direct its work, and co-edit documents together in Markdown or LibreOffice formats. Work is organized into projects with isolated files, instructions, memory and model presets, and a Plugin Hub offers more than 100 community plugins. A CLI connector bridges the agent to your host machine so it can also work in your real repositories, and multiple agents can cooperate by delegating subtasks to one another. Everything, from prompts to tools, is transparent and editable.

Key Features

• Full Dockerized Linux desktop the agent can drive, with terminal, files and GUI applications
• Built-in browser with DOM annotation, so you can click elements and turn them into instructions
• Live document cowork in Markdown and LibreOffice, plus persistent per-project memory
• Plugin Hub with 100+ community plugins, skills, agent profiles and model presets
• Host-machine bridge via the A0 CLI connector, and multi-agent cooperation and delegation
• Runs anywhere Docker runs, from a small VPS to a GPU workstation, with any major LLM provider

💳 Pricing

Agent Zero is free and open source; you install it on your own machine or server and bring your own LLM API keys or subscriptions. Development is supported by sponsors and the community.

Who Is It For

Agent Zero is for tinkerers, developers and automation enthusiasts who want a transparent, self-hosted general-purpose agent that can operate a real computer rather than only chat.