AutoGPT: Build, Deploy, and Run Autonomous AI Agents
AutoGPT is an open-source platform for creating, deploying, and managing continuous AI agents that automate multi-step workflows. Agents built on it run persistently, respond to external triggers, and chain actions together rather than waiting for a fresh prompt each turn.
Self-Hosting the Platform
Self-hosting AutoGPT is a technical setup aimed at developers. A cloud-hosted option is offered separately for teams that would rather run a managed environment.
System Requirements
Hardware
– CPU: 4+ cores recommended
– RAM: 8 GB minimum, 16 GB recommended
– Storage: at least 10 GB free
Software
– OS: Linux (Ubuntu 20.04 or newer), macOS (10.15 or newer), or Windows 10/11 with WSL2
– Docker Engine 20.10.0+
– Docker Compose 2.0.0+
– Git 2.30+
– Node.js 16.x+ and npm 8.x+
– VS Code 1.60+ or any modern editor
Network
– Stable connection with outbound HTTPS
– Access to the ports Docker configures during setup
Setup
The maintained, current instructions live in AutoGPT’s official self-hosting documentation; consult it for the full walkthrough. The steps assume Docker, Git, npm, and an editor are already installed.
For a local install on macOS or Linux, the one-line script installs dependencies, configures Docker, and launches a local instance in one pass:
curl -fsSL https://setup.agpt.co/install.sh -o install.sh && bash install.sh
Platform Architecture
AutoGPT separates the builder-facing interface from the execution backend.
Frontend
The Frontend is where users design, run, and monitor agents:
- Agent Builder — a low-code interface for designing and configuring agents.
- Workflow management — agents are assembled by connecting blocks, where each block performs a single action.
- Deployment controls — manage an agent’s lifecycle from testing to production.
- Ready-to-use agents — a library of pre-configured agents that run without building anything.
- Monitoring and analytics — track agent performance and refine workflows over time.
Server
The Server runs deployed agents. Once live, an agent can be triggered by external sources and operate continuously. It holds the core agent logic, the supporting infrastructure for reliable and scalable execution, and a marketplace of pre-built agents.
Example Agents
Two illustrations of what AutoGPT agents can do:
Generate short-form video from trending topics. The agent reads topics on Reddit, identifies what is trending, and automatically produces a short-form video from the content.
Surface quotable moments from your videos. The agent subscribes to a YouTube channel; when a new video is posted, it transcribes the video, uses AI to pick the most impactful quotes, and drafts a social post to publish automatically.
Licensing
- Polyform Shield License — all code and content inside the
autogpt_platformfolder. - MIT License — everything outside
autogpt_platform, including the original stand-alone AutoGPT Agent and projects such as Forge, agbenchmark, and the AutoGPT Classic GUI.
AutoGPT Classic and Related Tools
- Forge — a toolkit for building your own agent application. It handles most of the boilerplate so development effort goes into what makes an agent distinct.
- agbenchmark — works with any agent that supports the Agent Protocol, providing a strict, objective environment for autonomous performance testing.
- UI — the frontend for controlling and monitoring agents; it connects over the Agent Protocol, so it works with many agents.
- CLI — a command-line interface at the repo root that ties the tools together. After cloning, install dependencies with
./run setup.

