Coming from Conductor
Keep the agents. Add the part that was missing.
Conductor Cloud shuts down on 20 November. You can keep running parallel Claude Code agents exactly as you do today — and give that work objectives, acceptance criteria, and a signed record of who approved what.
What is actually happening, and when
Conductor built something people liked — running several coding agents in parallel, each in its own worktree. That is not in question here. These are simply the dates that matter if your work is still in the cloud.
- 4 November 2026Meta confirmed its acquisition of Conductor (Melty Inc.).
- 13 November 2026Conductor Cloud workspaces went read-only.
- 20 November 2026Conductor Cloud shuts down and deletion of cloud user data begins.
- Early 2027Meta has committed to model updates and bug fixes for the local Mac app into early 2027.
Dates as reported by DevX. If your work is only in Conductor Cloud, export it before 20 November.
What carries over, and what you get back
You keep
- —Parallel Claude Code agents, still running side by side
- —Git worktrees per agent, so nothing collides
- —A review step before anything lands
- —Your repos, your branches, your machine — unchanged
You add
- —Objectives with acceptance criteria, so an agent knows what finished means
- —A human acceptance gate: nothing is done until a person says so
- —A hash-chained, signed record of who approved what and when
- —Team seats, so the work is visible to more than one laptop
Moving over takes about five minutes
- 1
Export anything you still need from Conductor Cloud
Cloud data deletion begins on 20 November. Whatever is only in the cloud should come out first — this part is time-boxed and nothing else on this page is.
- 2
Connect Planwright to Claude Code
One command. OAuth opens in your browser for GitHub sign-in; there is no token to create or paste.
claude mcp add planwright --transport http https://mcp.planwright.tools/mcp
- 3
Turn your open work into objectives
Ask your agent to create an objective per piece of work in flight. It writes them through the MCP, so each one lands on the board with acceptance criteria you can actually check against.
claude "Read my open branches and create a Planwright objective for each piece of work in flight"
- 4
Carry on, with a trail
Agents claim objectives from the board, record their diffs, and request acceptance. You approve. Every transition is signed.
Using Cursor, Codex or Amp instead? Setup for every agent →
Fair questions
Is Planwright a drop-in replacement for Conductor?
Not exactly, and it would be dishonest to say so. Conductor was a great harness for running several Claude Code agents at once. Planwright is the layer above that: what the agents are supposed to achieve, and proof of who approved the result. You keep running agents the way you already do — through Claude Code, Cursor, Codex or Amp — and Planwright gives that work objectives and an audit trail.
Do I have to move off the local Mac app right now?
No. Only Conductor Cloud shuts down on 20 November; Meta has said the local app keeps getting model updates and fixes into early 2027. The urgency is about cloud data, not about your local setup. You can connect Planwright today and keep using whatever harness you like.
Does my work have to live somewhere new?
Your code never leaves your machine or your GitHub. Planwright holds objectives, acceptance criteria and the signed audit chain — not your repositories. Agents commit with your own credentials, exactly as they do now.
What does it cost to start?
Free covers a single user, which is enough to connect an agent and run real work through it. Team is $15/seat/mo when you want more than one person on the board.
Is there lock-in?
Identity is GitHub OAuth, and the audit verifier is open source — your evidence bundle is a portable, signed artifact you can check without us. Having just been moved off a tool you did not choose to leave, that seems worth saying plainly.
Your agents already work. Give them something to aim at.
Free covers a single user. No card, no trial clock.