Conn

Your agent works in your terminal. You keep the keyboard.

conn.eggp.dev · Docs · English · 한국어

Conn is a terminal you share with the AI agent you already use. The agent picks up the shell **you** are in: the server you logged into, the environment you activated, the directory you chose. You watch every command as it runs, approve the risky ones, and take over by typing.

• It starts where you are. Open the SSH session, container or virtualenv yourself, then let the agent continue in that live shell. A password typed at a hidden prompt never reaches it.
• Nothing runs behind your back. Every command the agent submits carries a one-line reason. By default rm, sudo, git push --force, kubectl delete and file overwrites wait for your approval, and one approval covers one action. Inside an SSH session, or any program Conn cannot inspect, every agent command waits for your review.
• Type to take over. Any keystroke takes control back before the agent's next input reaches the shell. Fix what is wrong, then tell it to read the screen and continue.
• Your agent, your model. Works with Codex, Claude Code, Cursor and GitHub Copilot over MCP. Conn runs no model and needs no extra subscription.

[](https://conn.eggp.dev/#film)

*You log in to your server. Your agent asks to delete the whole cache. You say no, type the fix yourself, and it finishes the job.* [Watch the full film](https://conn.eggp.dev/#film) ([file](docs/a