Self-hosting AI agents for document work: installing Open Docy in four minutes

Most AI agent tools assume your documents can go to someone else's cloud. For invoices, contracts and client files, that's often a non-starter. So we built Open Docy: a self-hosted Agent OS where you design agents on a visual canvas, run them on your own server, and keep a person in the loop for decisions that matter.

It's open source (MIT). One honest caveat first: the platform is open source, but the AI plugin is a signed, compiled download. The installer verifies its signature before installing it.

You need Python 3.12 (uv will fetch it), Node 20.19+ or 22.12+, and pnpm.

git clone https://github.com/opendocyai/Open-Docy cd Open-Docy uv venv --python 3.12 cp settings.ini.sample settings.ini && cp.env.sample.env Enter fullscreen mode Exit fullscreen mode Edit settings.ini and.env and replace every change-me value. One gotcha: the secret must be exactly 16 characters, or sign-up fails with a "passwords are tampered with" error. Then:

uv sync uv run docy install-ai uv run docy install-libreoffice uv run docy build-web uv run docy run Enter fullscreen mode Exit fullscreen mode install-ai downloads the AI plugin and checks its signature. On macOS, install-libreoffice only prints the Homebrew command ( brew install --cask libreoffice ), so run that yourself. When the server is up, open http://127.0.0.1:3000. A health check is at /api/v1/common/health.

The first account you create becomes the workspace owner. Passwords need at least 12 characters. A starter invoice agent is already there.

I recorded the whole install on a clean Mac. It takes about four minutes, and the long waits are sped up in the video: opendocy.com.

We're looking for feedback on the install experience and on which document workflows to build first. Questions and ideas are welcome in GitHub Discussions: github.com/opendocyai/Open-Docy