Skip to content
Hydra

Install Hydra on your server

Hydra runs on one Linux server that you own or rent. During a pilot we install it with you, on your server, and you keep the result. This page says what to have ready.

The server

  • Ubuntu 24.04, or a recent Debian. Hydra runs as a normal system service under its own user.
  • 4 vCPU and 6 GB of RAM run Hydra for a team of staff today. Heavy work (OCR, speech, upscaling) goes to remote drivers, so the server stays small. Add memory if engineers will run many terminal seats on it.
  • Disk: 40 GB to start. Chats, uploads, pictures and the audit log grow with use; the Monitor page shows how fast.
  • A domain name for it, and TLS in front of it (your reverse proxy, or Hydra's own HTTPS).

Sign-in

Staff sign in with Google Workspace. You create an OAuth client in your Google admin console and give Hydra its id and secret. You choose which domains may sign in, and whether people from your domain start active or wait for an admin.

Model accounts

Hydra talks to the model providers under your accounts: Claude and Codex subscriptions for chat and terminals, and an OpenRouter key for the models you route there. You connect them from the admin pages after the install; nothing is shared with us.

How an install goes

  1. You give us SSH access to the server, or a screen share while you type.
  2. We install Hydra and its service, point your domain at it and check /login answers.
  3. You sign in as the first owner. Hydra makes you the owner because your address is on the owner list.
  4. You connect your model accounts and invite the first people.
  5. We remove our access. From here the server is yours alone.

A packaged installer for teams who want to do this without us is on the way; until then the pilot is the install.