Every feature,
end to end.

When multiple agents try to edit the same file, /.relay uses file locking to prevent destructive conflicts. If one agent is modifying a component, another agent cannot silently overwrite those changes.

A change in one file can affect dozens of others. So /.relay goes beyond simple file locks. Instead of only asking "is this exact file being edited?", it can answer "does this change affect files that another agent is currently working on?" — which lets agents work in parallel while staying safe across connected parts of the project.

  @pony / claude   ── holds ──►  ui/Header.tsx
                                      ▲
                          imported by │
                                      │
  @unnath / cursor ── wants ──►  ui/Nav.tsx      ✗ dep-blocked
  @arjun  / codex  ── wants ──►  lib/date.ts     ✓ granted (disjoint)

Agents are not always running on the same machine. One developer has an agent running locally; a teammate has their own filesystem, their own environment, their own agents. /.relay keeps everyone synchronized using patch updates — instead of constantly copying the entire project, it propagates the changes, so every agent stays on the same version of the codebase.

None of this needs a hosted backend. One person on the team runs relay serve — that machine is the host. Everyone else connects over an ngrok tunnel, straight to that laptop. You are the server. You are the client. Nothing lives on infrastructure you do not own.

  host machine                                   guest machine
  ─────────────                                  ─────────────
  agents ─► hooks ─► lock table  ◄── ngrok ──►  hooks ◄─ agents
                        │                              │
                     dep graph                    lock mirror
                        │                              │
                  Mission Control                Mission Control
                     :3002                            :3002

• host Runs relay serve, owns the authoritative lock table, and mints invite links from the Team tab.
• guest Runs their own local relay, joins with the invite link, and mirrors the host's locks onto their board.
• invite-only Rooms can require a member token; guests without one get read-only room context.
• presence Who is online, which project they are on, and which agent is active — live, per member.
• relay doctor Walks the whole chain — local API, room join, tunnel reachability, host locks, mirror state — and names the broken link.

Started by relay serve on localhost:3002, with the API on 127.0.0.1:3001. It runs on your machine — there is no hosted dashboard and no account to create.

relay clone and relay add write pre-tool (claim), post-tool (release), pre-read, and stop hooks into your project for every supported agent. Bring whatever you already use — nobody on the team has to change how they work.

Pre-read hooks also register interest, so the board can show what an agent is looking at — not only what it is writing.

Hooks cover coordination automatically. MCP is for agents you want to hand structured tools to — and for reading room context from a machine that has no relay installed at all. Run relay mcp-url to print a ready-to-paste config; the room endpoint is read-only for anyone off the host machine.

• requires Node.js 18+ (20+ recommended) and npm. No database, no Docker, no hosted account.
• optional The gh CLI for relay login; ngrok for cross-machine rooms.
• env RELAY_PORT (3001) · RELAY_UI_PORT (3002) · RELAY_UI_ORIGIN