Relay
One AI teammatefor your wholecompany.
Every person gets a private workspace. The same agent works in your group chats, projects, and Slack channels, carrying the right memory, files, and permissions into each.
Relay what your company knows, where your team already is, and while nobody is watching.
01
Never explain your company twice
Relay learns the facts as you work. Your private chats stay private, and what the team learns, the whole company gets.
02
It already lives in Slack
Mention Relay in any channel and it answers in the thread. Same memory, same files, same permissions as the web app.
03
It works before you do
Describe the job once and Relay builds the schedule. It runs on time, sends you the result, and keeps a log you can open.
Not a chatbot. A coworker with a memory.
In Slack, on any model, with a computer of its own.
Slack
One agent, in every channel. Mention @relay in Slack and it answers in the thread, with the same memory, files, and permissions it has on the web.
- Threaded replies, no re-tagging
- One memory across Slack and web
- Quiet around external guests
Models
Any model, any harness. Weigh intelligence, speed, context, and cost in the picker, then run it on Pi, Claude Code, Codex, or OpenCode.
- Claude, GPT, and OpenRouter
- Speed, cost, and context compared
- Pi, Claude Code, Codex, OpenCode
Company brain
It remembers, so you never repeat yourself. Drop a file anywhere and it is read, not just filed. Talk, and it keeps what it learns as memory. One company brain that grows as you work.
- Facts kept where they were said
- Documents searched by meaning
- Never visible outside its scope
Sandboxes
Every scope gets a real computer. Not a code interpreter: a machine with a shell, where it runs your tests, opens PRs, and reads logs. Packages and files persist between turns and across days.
- A real shell, not a code interpreter
- Installed tools stay installed
- Untrusted work runs in a scratch box
Scheduled
Cron jobs, set up by asking. Describe the job in a sentence and the agent writes the cron, runs it while nobody is watching, and keeps every run open to read.
- Set up by asking, not by forms
- Run now, pause, or edit
- Every run logged and readable
The agent loop is yours to pick.
Under the model sits the harness: the loop that decides how the agent reads, thinks, and calls tools. Four of them ship, all driving the same core.
Same memory · same files · same permissions
Pi
Relay's own in-process loop
Claude Code
Anthropic's agent CLI
Codex
OpenAI's agent CLI
OpenCode
The open-source agent CLI
A teammate the company has, not a chatbot the company shares.
Most AI products are one user, one chat, and no memory of the company around them. Relay gives every person a private workspace and every room a brain the whole company keeps filling in.
One agent, every room
The same teammate works one-on-one, in a group chat, in a project, and in a Slack channel, carrying the right memory and permissions into each.
Pricing moves to usage-based in Q4.
A company brain that fills itself in
Facts land where they were learned: yours, the room's, and the company's. The org scope is mounted read-only into every turn, so nobody re-explains context.
Your workspacePrivate by construction
Every person and every room is its own scope, with its own memory, files, and keys. Nothing crosses between them without an explicit grant.
Questions
The short answers are here. For anything else, join the waitlist below.
Give your company one teammate.
Private workspaces, shared memory, and an agent that keeps working when nobody is watching. We are onboarding teams in order.