agent0ai/agent-zero
AI EngineeringAn open-source AI agent with its own sandboxed Linux desktop, browser and terminal for research, coding and document work.
What it is
Agent Zero is an open-source AI agent framework that gives the agent a real Linux workspace rather than only a chat loop. Inside a Docker sandbox it can use a terminal, a browser and desktop applications (for example Blender or LibreOffice), and you can watch or take over the desktop yourself. It is aimed at people who want an agent to research, write code, edit documents or drive a GUI, while keeping the work visible and recoverable.
Around the agent, the project provides a Canvas panel with browser and document views, a browser with DOM annotation, projects that isolate files/instructions/memory/secrets, on-demand skills, agent profiles, model presets, multi-agent delegation and a plugin hub. Prompts live in prompts/ and tools in tools/ or plugins, so behavior can be inspected and extended project overview.
Quick start
You need Docker, plus a supported AI account, a provider API key or a local model server.
- On a desktop machine, download and open A0 Launcher v1.9 for your OS from the desktop setup section. Supported builds include Windows x64, macOS Apple Silicon and Linux x64 AppImage.
- Let the Launcher guide Docker setup, then create an instance: choose Download Agent Zero or Run Agent Zero, keep the default version and persistent workspace, and wait for Running.
- Click Open UI and send a message. When asked, connect a model: choose cloud provider, AI account or local model, pick a Main model and a Utility model, then finish setup.
If Docker is already installed, the README also gives a one-line container start:
docker run -d -p 127.0.0.1:5080:80 -v a0_usr:/a0/usr agent0ai/agent-zero
Then open http://localhost:5080 on the Docker host Docker instructions. For a server or SSH session without a desktop, the README points to the interactive A0 Install script; on remote hosts, read the VPS deployment guide before connecting from your browser.
Constraints and safety model
The agent works inside Docker and should stay isolated; the README explicitly warns against mounting your whole home directory and against granting host file or command access to an instance you do not trust. Credentials belong in project secrets or settings, not in prompts or public files. Actions touching accounts, money, production systems or private data should be reviewed, and important workspaces still need backups. Time Travel snapshots under /a0/usr are a practical safety layer, not a replacement for Git or backups safety model.
A model provider is required before chatting. Host-file access through Launcher or the A0 CLI Connector is optional and must be installed on the machine the agent should work on, not inside the container. Plugin behavior and third-party provider integrations should be read before installation. The project states an MIT license.
HotRepo's take
On the Python daily list for 2026-11-03, Agent Zero recorded 38 new stars, reaching position 15 with about 19.7k stars and 3.9k forks. That is a moderately active single-day signal, not evidence of long-term growth; no broader ranking history is available here. The distinguishing technical fact is the full Linux desktop inside the sandbox and the inspectable prompts/tools, which make it a heavier but more controllable alternative to chat-only agent tools.
Written by AI from the project's README, updated . The project's own docs are the source of truth.