Coding agents on a server your team shares

How a thread works

• 

  Start from a prompt

  Pick a repository and an agent, then say what you want done. Eureka makes a branch and an Incus container for the thread, clones the repository into it and starts the agent.

• 

  Anyone on the team can reply

  Every member can open a thread while the agent works and send the next message. The transcript shows who wrote each prompt, and the avatars in the header show who is reading now.

• 

  The agent keeps going after you close the tab

  Turns run on the server, so closing the tab does not stop one. A thread that sits idle for 30 minutes goes to sleep. The next prompt wakes it and picks up the same agent session.

• 

  Changes are pushed as they happen

  Every turn ends with a commit and a push to the thread's branch. The page lists the changed files and links to the diff on GitHub. Open the pull request from the same page when you are ready.

Use the agent you already pay for

Eureka runs Codex, Claude, Cursor and OpenCode. Each member connects their own subscription or API key in settings. The model and effort pickers show whatever the agent reports, so there is no list to keep up to date.

• Codex
• Claude
• Cursor
• OpenCode

Runs on one Linux host

Eureka is one Elixir release and one SQLite file. It needs Incus for the sandboxes and a GitHub OAuth app for sign-in and pull requests. Nothing has to face the internet; the guide assumes a private network such as Tailscale.

Eureka is pre-alpha and the repository is private, so the guide and the source need GitHub access.