A terminal cockpit for coding agents

Keep your coding agents in context.

Run Claude Code, Codex, and Cursor in one terminal. Move a task to the next ready session through a handoff you can inspect, trim, confirm, or cancel.

  • Run Claude Code, Codex, and Cursor in one cockpit.
  • Review every context bundle before it moves.
  • Keep ready sessions live for a quick hand-back.

A reviewed handoff, whichever agent is next.

Three deliberate moments turn a task switch into a reviewable path to the right ready session.

  1. Prepare context

    Kitten assembles recent conversation context, touched files, and pending changes.

  2. Choose and review

    The developer chooses a ready target, then inspects the redacted preview and edits or removes context before sending.

  3. Confirm and continue

    Only explicit confirmation sends the bundle; the receiving agent then continues the task.

Install Kitten with your package manager.

Choose npm, pnpm, or Bun. Each route installs the published CLI and the matching macOS or Linux binary.

npm install --global @matheusbbarni/kitten

Copy this command into a terminal. It installs the Kitten CLI and the matching supported platform binary.

What you need

Kitten uses your existing coding-agent installations and runs from inside a Git repository.

Questions before you try it

The short version of Kitten's scope, control boundary, and fallback behavior.

What moves between agents?
A bounded bundle of recent conversation context, touched files, and pending changes — not every piece of task history.
Which coding agents work with Kitten today?
Claude Code, Codex, and Cursor. Install and authenticate the local agent CLIs you want Kitten to run.
Will Kitten support more agents?
More Agent Client Protocol (ACP) connections are planned. Kitten adds a connection only after its integration is supported and reviewed.
Does Kitten send anything automatically?
No. The handoff opens in a preview where you can edit, trim, confirm, or cancel it. Only confirmation sends it.
Does redaction replace human review?
No. Kitten redacts recognized secrets before preview, and your review remains the final safeguard before sending.
What happens if one agent is unavailable?
Kitten marks that session unavailable and leaves the other ready sessions usable instead of taking down the cockpit.
Does the showcase track visitors?
V1 emits no automatic website analytics. Kitten's application telemetry is separate, local, content-free, opt-in, and off by default.