A self-hosted AI agent platform for your own server

Agenhood is an open-source, self-hosted AI agent platform. Run many AI agents on one server with Docker, each in its own sandbox. Your keys and data never leave your machine.

What is a self-hosted AI agent platform?

Agent frameworks help you write agent code. Chat and workflow builders help you build apps around a model. A self-hosted AI agent platform does the part in between: it runs the agents on your server, keeps each one isolated, keeps its files and shows you what it did.

That is what Agenhood does. You pick a ready-made agent, such as Claude Code or Codex, and give it work. Agenhood handles the rest.

What you get

  • A sandbox for every agent. Each agent works in its own locked-down container and reaches the internet only through a filter. How the AI agent sandbox works.
  • Agents that keep their work. Files survive restarts. Idle agents pause, use no CPU or memory, and wake up when the next task arrives. Long-running AI agents.
  • Undo for bad runs. Every task saves a snapshot, so you can roll back what an agent changed. Agents working in a Git repo use its history instead.
  • Schedules. Run a task once or every hour, day, week or month. Scheduled AI agents.
  • A console and an API. Give tasks and watch agents work live in the browser, or do the same from your own code.
  • GitHub and Slack. A comment on an issue or pull request, or a Slack message, becomes a task. The agent replies in the same thread.
  • Your keys stay with you. API keys are stored encrypted on your server and never reach the browser.

Which agents and models can you run?

Claude Code, Codex, Opencode, or Agenhood's own agent. You choose per agent, so one server can run a mix. They use Anthropic and OpenAI models with your own API keys.

Agents can search the web, browse pages, run code and work with files. So they handle research, reports and data work, not only coding. For coding, see self-hosted coding agents and the Claude Code sandbox.

Run AI agents on your own server

Everything runs on one VM with Docker. No Kubernetes and no hosted service. Your data, files and task history stay on that machine.

Web console React app Your code REST API, /v1 Single Docker host (one VM) REST + SSE Postgres tasks, history Control plane API, auth, scheduler Connectors GitHub, Slack Internal network, no gateway provisions, drives Agent containers one per agent /workspace volume Egress proxy only way out SearXNG web search filtered Internet, LLM APIs

An 8 GB VM runs three agents at once, and a 12 GB VM runs five. The sizing guide covers bigger setups.

Open source and free to run

Agenhood is an open source AI agent platform under the MIT license. The full code is on GitHub. There are no licence fees: you pay only for your server and your model usage.

Compare Agenhood

Looking at other options? See how Agenhood compares with Claude Managed Agents, Manus, Daytona, Codex Cloud, Cursor background agents, Hermes Agent and OpenClaw.

What it doesn't do

  • One server. All agents run on one Docker host. To run more at once, use a bigger VM.
  • For your own team. Admins create the accounts. It is not built to be a public, multi-tenant service.
  • You run it. There is no hosted version, so updates and backups are up to you.

Questions people ask

What is the difference between an agent framework and a self-hosted AI agent platform?

A framework is a library for writing agent code. You still have to host, isolate and schedule what you build. A self-hosted agent platform like Agenhood does that part for you, on your own server.

Is Agenhood open source?

Yes. Agenhood is MIT licensed and the full source is on GitHub. There is no hosted version: you run it on your own server.

How big a server do I need to run AI agents?

An 8 GB VM runs three agents at once and a 12 GB VM runs five. Agents with a built-in browser need more memory. The sizing guide on GitHub has the details.

Which models can the agents use?

Anthropic and OpenAI models, with your own API keys. You pick the agent and the model per agent. Opencode agents can also use Opencode's own model plans.

Does any of my data leave the server?

Your files, task history and keys stay on your server. The only things that leave are the prompts sent to your model provider and the web requests an agent makes through its filter.

Can I try it on my laptop first?

Yes. One command, make dev, runs Agenhood on your laptop with nothing to configure. You only need a server and a domain when you want to run it for real.

Run it locally in one command

You need Docker and Python 3.12 or newer. make dev builds the agent image, starts the stack, runs migrations, creates a login and prints the console URL.

Terminal
git clone https://github.com/appssemble/agenhood.git
cd agenhood
make dev