Your persistent backend for every AI tool — messages, knowledge, todos, and memory that travel with you across every session.
Shared knowledge bases
Wiki-style notes with tags, member roles, and invitations. Share a single knowledge base across your AI sessions, your team, or your closest friends.
Try saying…
Structured tables, inside knowledge bases
Actual SQL tables for app-like data. Define schemas, insert rows, run queries — directly through MCP tools.
| id | name |
|---|---|
| 1 | Carbonara |
| 2 | Pesto |
| recipe_id | name |
|---|---|
| 1 | pasta |
| 1 | pancetta |
| 2 | basil |
Try saying…
A classic REST API, beyond MCP
Every table is also reachable over plain HTTPS — so scripts, cron jobs, and apps can read and write your data outside any AI session. Scoped API keys (read-only or read-write, one KB or all) are created in Settings, and your coding agent gets built-in docs to wire it up for you.
$ curl .../api/v1/kbs/$KB_ID/query \
-H "Authorization: Bearer rk_a1b2c3d4..." \
-d '{"sql": "SELECT name, minutes FROM recipes"}'{"rows": [{"name": "Carbonara", "minutes": 30},
{"name": "Pesto", "minutes": 20}]}Try saying…
Custom dashboard pages
Ask your AI to write a small HTML page that reads your KB data live — portfolio dashboards, expense charts, browseable note indexes. Renders in a sandboxed iframe right inside the Relay web app.
- AAPLstock35%
- BTCcrypto18%
- CASHcash15%
Try saying…
The Relay Library
Vectorized document storage. Upload your PDFs, Word files, and text documents once — then ask your AI questions about them in plain language. It searches by meaning, not keywords, and answers with the exact passage, quoted and cited down to the page.
“…either party may terminatethis agreement by giving three months' written notice to the end of a calendar month.”
“…the flat is to be returned in swept-clean condition on the final day of…”
Not one word of the question appears in the clause that answers it.
Try saying…
Todos & persistent memory
Keep a private todo list and let AI tools save durable facts about you. Both survive across sessions, devices, and tools.
- Buy milk
- Book flights
- Reply to Alice
- Schedule dentist
- Prefers dark mode
- Lives in Berlin
Try saying…
Encrypted messaging
Talk to other Relay users by username or alias. Every AI tool you use can read and reply on your behalf.
Try saying…
The brain for your agent
Point an autonomous agent at a knowledge base and let it live there. The KB holds the agent's instructions, working memory, and structured data — and the agent rewrites its own entries as it learns. A scheduled routine wakes it up; everything specific it needs, it looks up in its brain.
A portfolio management agent
A day in the loop
- 1A daily routine firesA scheduled trigger sends one short instruction: "do your run — your brain is this knowledge base." No strategy in the trigger itself.
- 2The agent reads its brainIt loads its operating instructions, its watchlist, its tables, and the lessons it has written for itself so far.
- 3It acts and recordsEvery run writes a row to its decision log and updates its structured state — a full, queryable audit trail of what it did and why.
- 4It improves itselfIt appends new lessons to its playbook, grooms its notes, and refines its own filters — editing the very brain it will read next time.
The brain evolves, so tomorrow's run starts smarter than today's.
What lives in the brain
Watch it work
- openNew position — momentum + volume confirmationtoday
- holdThesis intact, no better edge available1d
- swapRotated into a stronger setup3d
- flatNo setup cleared the bar — logged why4d
The agent authors its own dashboard page that reads its tables live — so you can monitor its latest actions, decisions, and performance without touching the data.
Try saying…
Every document you own, answerable
The Relay Library is vectorized document storage: your PDFs and Word files, chunked and embedded, searchable by meaning. You stop filing things away to never find them again — you just ask, and get the exact clause back, quoted and cited.
Asking your paperwork a question
1 · Upload once, tag as you go
Drop a stack of files in at once. Each gets tags and a content date — the date the document is about, not the day you uploaded it. Those two become the filters your agent searches with.
2 · Ask in plain language
Your agent narrows to the right shelf before it reads a single page — the tag keeps the lease and the manual out of the way:
search_library({
query: "water damage from a burst boiler — is it covered?",
tags: ["insurance"],
date_from: "2024-01-01",
top_k: 5,
})3 · Get the passages back, cited
“§7.2 Escape of water — damage caused by water escaping from fixed heating installations, including boilers and connected pipework, is covered up to the sum insured.”
“§9.1 Claims must be notified within 14 days of the insured event. The excess payable per claim is €250.”
Verbatim, ranked, and each one carrying its document and page. The clause that answers the question never says “burst” or “flooded” — semantic search finds it anyway, and the second hit surfaces the deadline you didn't know to ask about.
4 · Then chain it into the rest of Relay
The Library is where the facts come from; the KB, todos, and email are what you do with them. One sentence from you, and the retrieval, the deadline, and the paper trail all land in the same backend.
Try saying…