One memory.
Every AI tool.

A persistent backend your AI connects to over MCP. Notes, documents, tables and todos that outlive the chat window.

Get startedLog in
ClaudeGrokCursorCron jobsYour scriptsChatGPT

…and any other MCP client.

Persistent

Nothing dies with the session.

Portable

The same data in every client.

Open

Reachable over plain HTTPS, not just MCP.

The Relay Library

A vectorized database for your own documents. Upload your PDFs once, then ask in plain language — it answers with the passage, cited to the page.

“What’s the notice period if I move out?”
Lease_2024.pdfp. 7

“…either party may terminate this agreement by giving three months’ written notice to the end of a calendar month.”

Knowledge bases

Notes your AI writes, tags and re-reads — plus real SQL tables for structured data an agent can query. Share one with your team.

Server architecture#infra
Project A docs#work
Report template#report

Todos

One list. Every tool. Still there tomorrow.

  • Buy milk
  • Book flights
  • Reply to Alice

Relay agents

Build your own agent inside the app — give it a purpose, pick any model, and chat with it. It works on your Relay data.

Research agentanthropic/claude-sonnet-5 ▾
Check the watchlist and log what changed.
Three moved. Wrote the rows and updated my playbook note.

Live dashboard pages

Ask for a dashboard, get a real one — HTML your AI writes, reading your data live.

Portfolio
5Positions
100%Weight
3dRebalance

Messaging

Reach other Relay users by name. Your AI reads and replies.

@alice · Got the Q3 numbers?
Sent — they’re in the shared KB.

Tables & a REST API

Your KB tables are reachable over plain HTTPS too — scoped keys, no AI required.

$ curl relaymcp.xyz/api/v1/kbs/$KB/query \
    -H "Authorization: Bearer rk_a1b2c3…" \
    -d '{"sql":"select name, minutes from recipes"}'
→ [{"name":"Carbonara","minutes":30}, …]

Just say it

“Save these meeting notes to my work KB.”“What’s the notice period in my lease?”“Build me a dashboard from my positions table.”