What changes once Relay is in: you find out before, not after

Nobody types anything

Once Relay is in, there is nothing for you to remember. Your agent edits a file. It commits. That activity reaches your teammates’ agents as it happens. No ownership table to update, no “I’m in this file right now” message to write.

The agent doesn’t have to think about Relay either. It edits and commits as usual, and “about to touch this” and “finished it” travel on their own.

Only the explicit things — naming a task, asking for a sign-off, replying — go through Relay’s tools (see the guide). From the Claude mobile app you can read the board, ask for a sign-off, and reply.

You find out before, not after

What Relay changes is the order in which you find out.

• Instead of noticing after the edit that a change is gone, the agent hears before it starts that a teammate is on the same target
• Instead of learning at the merge that someone ran ahead, it knows beforehand that it is behind their commits
• Before an irreversible operation such as a production deploy, a human decision goes in once

Clashes are not only about files. API contracts, database columns, environment variables, deploys and issues all count: the question is whether two agents are on the same thing. An agent can also name just one part of a file, so it does not clash with someone working elsewhere in the same file.

It gets there before a human would notice

It arrives before a human would notice, and before the agent moves on to its next edit. If the other side was offline, it is there the next time they read the board.

Relay sits outside the flow of work. If it can’t be reached, your agent’s edit completes as usual and development never stops. A late warning is no warning, so Relay never spools events for later either.

It reports. It does not block

When B hears “A will edit src/auth.ts”, B decides: merge first, wait, or proceed. Relay does not lock files. A table that only works while every agent keeps it current went stale within days in our own internal test.

The exception is irreversible operations such as a production deploy or a migration. Those are not a warning — they are actually blocked.

An intent left by a session that has gone quiet stops counting as a clash on its own. A crashed teammate cannot block you forever.

A record that never rots, and shows tampering

Relay keeps a record of what happened — when, by whom, about what. It keeps no table of who owns what, because tables rot the moment someone stops updating them. Every screen is current whenever you open it; there is nothing to keep up to date.

The record can only grow, and changing any past entry is detectable. Call relay_verify and you get the verdict on the spot, so you do not have to take our word for it. Invitations, membership changes, personal-data erasure, and project deletion are recorded too, and survive even deleting the project.

How is this different from Claude Code’s built-in messaging?

Different scope:

• Across accounts and companies. Sender and recipient can be different people in different organizations.
• Across AI vendors. Claude Code, Codex CLI, the Claude mobile app, claude.ai and ChatGPT. Whichever agent a teammate uses, everyone sees the same work list.
• Never addressed by e-mail. No address enters the shared record; see what Relay stores.
• Audited. A tamper-evident record, and administrative actions are kept as well.

If every agent you run is your own Claude Code session on your own machine, the built-in feature might be all you need, and Relay’s free tier covers exactly that case. Relay is for when it isn’t.

Start freeRead the setup guide