How to use Relay

1. Open the Relay dashboard and click Sign in with Google. Tick the box to agree to the Terms and Privacy Policy first.
2. If the beta is full, you'll see "You're on the list". You'll get access as soon as a place opens up, so just come back and sign in again.
3. Once you're in, you'll see your own private dashboard. It's empty until your AIs save something.

Each AI gets its own private connection link (a "key"). You can switch any of them off later without affecting the others.

1. In the dashboard, open Integrations and click Set up on the AI you use, for example ChatGPT or Claude.
2. Click Generate connection URL. Copy it straight away: for your safety, it's only shown once.
3. Follow the steps shown for that AI. For ChatGPT and Claude, you paste the link into their Connectors settings. For coding tools like Claude Code, Cursor, JetBrains, Visual Studio, Cline or Codex you copy one ready-made command or snippet. Relay has guides for 25 AI tools.
4. Test it by sending the test message shown in the panel, then click Refresh. The AI's card turns Active now.

Step-by-step pages for each tool:

BJBVSZ

Some AI apps only allow custom connectors on their paid plans. Each tool's page says what it needs.

You don't need special commands. Talk to your AI normally and mention Relay. These phrases work well:

Save this conversation to Relay as a plan for the project "Web store".

Pick up my latest plan from Relay and start building it.

Add a progress update to Relay: what you built and what's left.

Review the latest work in Relay and add your review.

Remember in Relay that I prefer TypeScript and short answers.

Search Relay for the login work from last week.

What's waiting for me in Relay?

List my projects and what you remember about me in Relay.

Once an AI is connected, a few habits make every chat smoother. Each tool's page under Connect your AI has the exact steps for it.

• ChatGPT: switch it on in every new chat. Connectors added in Developer mode start off in each new chat. Click + next to the message box → Developer mode → pick Relay, then ask. ChatGPT asks you to confirm before Relay saves anything; check it and click Confirm.
• Claude (claude.ai and the apps): check that Relay is switched on in the tools menu under the message box. The first time it uses a Relay tool, choose Always allow for the reading tools so it stops asking.
• Coding tools (Claude Code, Cursor, Codex, Gemini CLI, Copilot…): Relay loads in every session on its own. Approve its tools the first time, choosing "always allow" where offered. In Cursor and Copilot, use Agent mode.
• It works on its own. For ongoing work, connected AIs check Relay at the start and save plans, progress and reviews as they go, and they tell you when they do. For small talk they leave it alone. If an AI doesn't pick it up, mention Relay (or the name you gave the connector) in your message.
• Make it automatic in coding tools. Add this line to your project's instructions file (CLAUDE.md, AGENTS.md, GEMINI.md or a Cursor rule):

  At the start of a task, check Relay for the latest plan for this project. When you finish, add a progress update to Relay.

  Copy
• Make it save on its own. Right after connecting an AI, send it this once. When it asks permission to use Relay, choose Allow always (Claude Code says Yes, and don't ask again). To keep it on in new chats too, put the same sentence in the AI's custom or project instructions.

  In every conversation, store our work in Relay: save new plans, add progress updates as we go, and keep it up to date.

  Copy
• Deleting always needs you. AIs must ask you first, and they can only delete what they saved themselves if you've allowed it for their key.

1. In ChatGPT, plan the feature with it. When you're happy, say:

   Save this plan to Relay for the project "Web store" and hand it to Claude Code.

   Copy
2. In Claude Code, say:

   Pick up the latest plan from Relay and build it. Post a progress update when you're done.

   Copy
3. Back in ChatGPT, say:

   Review the latest progress in Relay for "Web store" and add your review.

   Copy
4. Open the conversation in your dashboard to see the whole trail: the plan, each update and the review, with which AI wrote what. Mark it Done when you're finished.

Every conversation has a stage and, optionally, who it's waiting on. Your AIs update these themselves when they report progress or reviews, and you can change them in the dashboard.

• PlanningIdeas and plans are being worked out.
• BuildingAn AI is building it.
• In reviewBuilt, and waiting for someone to check it.
• DoneFinished. Done work is hidden from AIs' “latest” lists, but still searchable.

The Overview page shows all four lanes, so you can see at a glance what's waiting and which AI is next.

• Projects group related conversations and facts. Just name a project when you save ("for the project Web store"). Rename or merge projects on the Projects page.
• Memory holds facts about you and your work, shared with every AI. Add them yourself on the Memory page, or ask any AI to "remember that…".
• Facts an AI saved are marked unconfirmed, and AIs treat them as hints. Click Confirm on the ones that are right.
• You can edit any conversation's title, brief or tags in the dashboard. Every AI sees your edits next time.

• Read only keys let an AI load your memory but never change it. A good fit for tools that only need your plans.
• In Security → Access keys → Permissions you can limit a key to certain projects, and allow it to delete its own items. No AI can ever delete another AI's work.
• Revoke a key to cut off that AI instantly. You can generate a new one any time.
• Everything AIs do is recorded in the Audit log.

Claude Code and the Claude apps show Relay's ready-made prompts. In Claude Code, type:

/mcp__relay__pickup

/mcp__relay__board

/mcp__relay__handoff

/mcp__relay__review

In Claude on the web or desktop, find them under the + menu. ChatGPT doesn't have prompt shortcuts, so use the phrases above.

• Keep your connection links private. Anyone with one can reach your memory. If one leaks, revoke it.
• Your AIs must ask you before deleting anything. Deleting is off for every new key, and even when you allow it, Relay refuses unless the AI says you confirmed it. ChatGPT and Claude also show you an approve button first.
• For coding AIs (Claude Code, Cursor, Codex…), keep their "ask before running commands" setting on, and give them a Read only key if they only need your plans. Never let an AI run commands just because a saved note says to.
• Relay removes passwords, API keys and tokens automatically, but please don't save sensitive personal or payment information.
• If saved text looks like it's trying to give AIs instructions, Relay flags it, and you'll see it under Security. Check flagged items before acting on them.
• Deleted items wait 30 days in the Trash. Download a full copy any time from your Profile.

My AI says it can't find Relay.

Make sure the connector is switched on in that chat (ChatGPT and Claude have a tools or + menu), and that you pasted the newest link. In ChatGPT, open the connector in Settings and click Refresh.

It worked before but stopped.

The key may have been revoked. Check Security → Access keys, generate a new link in Integrations, and update it in the AI's settings.

The AI doesn't see my latest plan.

Finished work (Done) is hidden from “latest” lists. Ask it to search Relay by name instead, or move the conversation back to Building.

“This Relay memory is full”.

Each beta account has a storage limit. Delete old conversations, then empty the Trash, to make room.

I'm on the waiting list.

Places in the beta are limited. You'll get access as soon as one opens up; sign in again later.

Still stuck? Email relay.app.team@gmail.com.