Claude Code and Codex, side by side.
You already pay for Claude Code or Codex. Turret runs them in one window, alongside Gemini and DeepSeek, on your own machine and your own repositories. Sessions hand work to each other, across engines, and keep going while you're away from the desk.
Several sessions at once
Every session is its own process, working in a folder you pick. The sidebar lists them all, with the ones they started nested underneath. Close Turret and open it again, and each session picks up where it stopped.
A worktree each
Tick Open on a new worktree when you start a session, and it works on its own branch in its own checkout. Your folder's current checkout stays as it was. If your repository has its own WorktreeCreate hook, Turret uses it.
Sessions that plan, build and review
A session can start another session, send it a message while it works, and read what it did. You ask in plain words. One session writes the plan, two children take a phase each, and the parent reads their work when they finish. A child can start a child of its own, and that's as deep as it goes.
Codex and Gemini beside Claude Code
Pick the engine for each session: Claude Code, Codex, Gemini or DeepSeek. A child can run on a different engine from its parent, so a Claude Code session can hand a phase to Codex.
Claude Code and Codex run on your own sign-in. Turret runs the CLI's own sign-in and never reads the login. Gemini and DeepSeek take an API key, which Turret stores encrypted with your operating system's own encryption and never shows back.
Every engine gets the same sidebar, transcript and Turret tools. Each keeps its own permission model:
- Claude Code asks before it acts, with Allow once, Allow always and Deny. You can write rules, and put a tool on a list it is never allowed.
- Codex runs in the sandbox you pick: read only, workspace write or full access.
- Gemini and DeepSeek run read only or full. Read only withholds every tool that changes something.
Keeps your CLIs current
When a newer Claude Code or Codex is out, the session shows a banner. Press it, and Turret runs the CLI's own update.
Automations
Save a prompt with a folder, an engine and a schedule: by hand, daily, on chosen days, or every few minutes. Each run opens a fresh session, on a fresh worktree if you want one, and the run is kept for you to read. Turret has to be running for a schedule to fire. If it was closed, a missed run can catch up once when it opens.
Computer use
Switch computer use on and a session can look at the screen, click and type in another app. It works on macOS, Windows and Linux. On a Mac you grant two permissions. On Linux it needs GNOME on Wayland, or an X11 session. A session never clicks inside Turret itself.
Your other machines
Pair a second desktop running Turret. Each sees the other's sessions, and either can start a session on the other, on a folder or a GitHub repository. Sign both in to the same Tailscale account and they reach each other from any network.
