An AI agent such as Claude Code can now work with your servers through Serverus, right in your visible tabs, so you see everything it does. Its commands are typed into the server's real terminal tab (the one you have open, or a new one opened in the background), and it gets back the exit status and clean output. Shell state carries over between you and the agent: working directory, environment, sudo, virtualenvs. You can take the terminal over at any moment and hand it back when you're done. File tools work on SFTP, FTP and S3, and uploads, downloads, deletes and permission changes run in the tab's transfer queue, where you can follow and cancel them.
Agent access is off by default. You set it per connection or per folder, and folders pass it down. The levels are Off, Read-only, Ask and Full. With Ask, every change and command waits for your approval in Serverus, and you can allow it for 15 minutes at a time. There is also a switch for full access to everything. If you allow it, the agent can add connections you describe to it, passwords or keys included. It never chooses the new connection's access, and one that would get Full access asks you first. Passwords, keys and connection notes never reach the agent. While an agent is connected, the vault doesn't lock on idle or sleep. The lock button still locks it.
To connect Claude Code, turn agent access on in Settings → AI Agent and run the command shown there once. Starting a Claude Code session doesn't open Serverus: it starts when the agent first uses one of its tools. Agent access works on macOS and Linux and isn't available on Windows yet.