COOKREW an open-source spatial workspace for AI agents

Cookrew / Features / checkpoints

Every agent turn is a checkpoint: scrub, fork, pin

In Cookrew every agent turn is a checkpoint. You can scrub back through a session, fork a new agent from any past turn, and pin a version when a team is exported or called. Exporting or calling never touches the original session.

Get CookrewGet started

RECORDED3 steps, captured from the running app

How it looks

  1. 1

    A card open on its rail

    A card open on its rail
    ● RECClicked Mason open after a morning of work: the finished turn sits above the LIVE line, the live terminal runs under it, and the rail on the right counts the eight checkpoints it has taken in this session.
  2. 2

    Mid-scrub on the rail

    Mid-scrub on the rail
    ● RECPressed the rail at its live end and dragged upward: the checkpoints fanned out with the title each turn was given — the suite run, the 999 B and 1 PB cases, the rename to UNITS, the README line — and the transcript followed the marker.
  3. 3

    What a past turn offers

    What a past turn offers
    ● RECHeld the pointer on one row of the fanned list until its actions surfaced: ROLE, FORK and REWIND — fork starts a new agent from that checkpoint and leaves this one running, untouched.

Questions about checkpoints

What happens to my session when I fork or export?

Nothing. Forking, exporting and serving all copy from a checkpoint; the original session is never touched.

Where do checkpoints come from?

From the harness’s own session record: Cookrew reads the transcript the agent writes, so a checkpoint is the turn as the agent saw it, not a scrape of the screen.

Related

A spatial canvas for AI agents: terminals, notes and browsers

Every black card is a real terminal running a real agent. Notes and browsers sit beside the agent they are about, and the wires are the org chart.

The Board: every agent’s phase, tokens and checkpoint on one screen

What each agent is doing, how many tokens it has spent, which checkpoint it is on. Notes, forks and session cards mix into one activity stream.

The agent team marketplace: open someone’s crew from a browser

The marketplace lists doors, not copies. A team stays on its author’s machine; you sign in, buy a seat once if it is priced, and get a live sandboxed session of your own.

Cookrew is an open-source desktop workspace for running a team of AI coding agents on one canvas. It runs Claude Code, Codex, OpenCode and Pi side by side, records every agent turn as a checkpoint, and can serve a saved team at a cookrew.dev address where anyone can open a live session from a browser or from the Cookrew app. Source on GitHub ↗