Skip to main content

​
What happened

Conductor has been acquired by Meta. If you’re looking for a replacement, Ref is where your team runs its coding agents together, with one shared doc per project that carries the context, so work hands off between people and agents.

​
Why Ref

Ref is where your team runs its coding agents together.
  • Hand off without pasting chat logs. Each project has one shared doc that carries the context. A teammate, an agent, or you tomorrow can pick up where the last one stopped.
  • Run agents anywhere. Run Claude Code on your laptop under your own login, with Codex and OpenCode over ACP rolling out. Or use a cloud harness like Cursor, Devin or Warp OZ.
  • One doc per project, not a pile of threads. The top ref shows where things stand. Each workstream is a sub-ref with its own agent, which you can see and steer from the browser or from Slack.
  • Context that outlives the chat. Decisions, diagrams and what each agent tried stay in the doc for the next person or agent.
  • Review when it matters. Approve Tasks before code starts, and check each PR before it merges.
The rest of this page takes what you did in Conductor and shows how to do it in Ref.

​
Set up Ref

1

Install

On the laptop you want agents to use, run this from your repo:
It installs the ref CLI, connects Claude Code and Codex to Ref, and opens your browser to sign in. It needs claude or codex on your PATH. To connect by hand instead, see Claude Code or Codex.
2

Pair your laptop

Approve the laptop in the browser, then run ref launch enable in each repo you want agents to work in.
3

Write your first ref

Open Ref, describe the work, and ask for Tasks. See Your first plan.

​
Run parallel agents

In Conductor, you manage agents one by one. Each workspace was a git worktree on its own branch. In Ref, you can do the same thing. You can also work in a notebook to manage multiple agents working on the same shared context. See Multi-agent workflows.

​
Use your existing subscriptions to Claude Code, Codex or OpenCode

Ref drives those agents over the Agent Client Protocol (ACP). Claude Code runs on your paired laptop or cloud machine under your own login, so its usage counts against your own Claude plan. Support for Codex and OpenCode is rolling out. Ref starts each agent and messages it while it works.

​
Run in the cloud, with your laptop when you need it

In Conductor Cloud, agents ran on Conductor’s machines. In Ref, the Ref agent runs in the cloud and can send work to your paired laptop or cloud machine. Ref also integrates with cloud harnesses as well: Cursor, Devin or Warp OZ. To get an always-on machine of your own, set up an exe.dev box and pair it the same way as a laptop. Ref includes a hosted Claude Code agent which supports bring-your-own-key.

​
Review and steer

In Conductor, you left comments in the Diff Viewer and they went back to the agent. In Ref, you comment and review on the ref or the PR, a reviewer agent checks the PR, and your comments go to the agent doing the work.

​
Work as a team

In Ref, a ref is shared with your team with view or edit access. Teammates see which agents are running and where each Task stands, comment, and run their own agents on the same ref. @mention Ref in Slack to turn a thread into a ref. You can’t yet read or message a teammate’s agent conversation; hand off through the ref and its comments.

​
Bring your repo setup

Ref uses your existing Codex or Claude Code installation so there is no additional setup.

​
Get help

If you have any issues, please reach out to help@ref.tools. We’re happy to help you get started!