For people who already run an agent.
The landing page says what Relay does. This page says how, and shows the one take that proves it.
He says its name. His computer starts the work.
Sixty-two seconds, one take, no cut.
Raw, uncut. It can work on your own computer, too.
Hermes
Your agents run on your computer.
The glasses see, listen and answer. Your phone connects them to your computer. That is the whole system.
It runs the agents.
Claude Code, Codex and the rest run here, the way they do now. Same sessions, same files, same accounts. Leave it on and you can reach it.
It connects the two.
Bluetooth to the glasses. Internet to your computer. It stays in your pocket. Anything you can say, you can also tap.
A camera, two speakers, one mic.
Say its name and talk. The answer comes back in your ear. No screen, and nothing runs on them.
It knows each agent better than you do.
Every agent has its own commands, its own flags, its own way of being asked well. Relay knows all of them, writes the prompt properly, and comes back to you only when it actually needs you.
# what Relay actually sent
/resume payments
Run only what is failing: npm test -- --failed
Read the diff before editing. Smallest change
that makes it pass. Do not touch unrelated files.
Loop: re-run after each edit, stop when green
or after 3 tries, then report.
It writes the prompt.
You say a sentence. It sends the slash commands, the flags and the loop that agent responds to, because it knows them and you should not have to.
It picks who answers.
Continue a session, start a fresh one, or call a tool. When it is not sure, it asks instead of guessing.
Shared across all of them.
One memory every agent reads and writes, and any tool you add shows up in all of them. Settle it in Claude Code, and Codex has it tomorrow.
Not another camera.
Ray-Ban Meta is a camera you talk to. Relay runs the agents you already work with.
| Ray-Ban Meta | Relay | |
|---|---|---|
| Music, calls, photos | Yes | Yes |
| Runs your coding agents | No | Yes |
| Wake it with your own name | No. “Hey Meta” only | Yes |
| Use your own AI | No. Theirs only | Yes |
| One memory across agents | No | Yes |
Running in three steps.
About ten minutes, once.
| Runs on | macOS or Linux, arm64 or amd64. On Windows, inside WSL2 |
|---|---|
| The one requirement | It does not sleep. A Mac mini, a Linux box, or a laptop — setup offers to keep a Mac awake with the lid closed |
| Disk | About 300 MB, most of it the local embedder |
| Agents | Whichever you already have. It finds them |
-
Step 1
Set up a machine.
Your own computer, or any machine you leave on. Setup prints a code when it is done.
No machine to spare? The Relay membership runs one for you, with the glasses included: try it free.
-
Step 2
Install the app.
Sign in with your email, then enter your code.
-
Step 3
Put the glasses on.
They pair with your phone. Pick a name for your agent, then just say it.
What's inside.
| Storage | 4 GB on board. The day is recorded by your phone, not the glasses |
|---|---|
| Camera | Photo and video |
| Mic | One, with noise cut |
| Speakers | Two, for music and calls |
| Wake | Say its name, or tap the temple |
| Controls | Touch and two buttons |
| Charging | Magnetic USB‑C. Works while worn |
| On-device AI | None. Every model runs on your computer, not the glasses |
| Cost | $299, and nothing after it |
Questions.
I already run Claude Code on my laptop. Why do I need this?
You do not, at your desk. Relay is for the rest of the day: the session stays open while you walk out, and you reach it by voice with no laptop to open. It also writes down the meeting you just left, so the agent has the context you would otherwise have to type back in.
Which models do the included tokens use?
Claude, Kimi, Qwen, DeepSeek and OpenAI. We pick the one that fits the job. A big job starts on the best model we have. If your month runs up a large bill, we move you to cheaper models. We do not stop you. You can always ask for more.
Can I connect my own Claude account?
Yes, on any plan. We start the login on your computer and send you the link. You sign in yourself and send the code back. We never see your password. Your Claude account is then used in Claude Code only. For OpenAI, Kimi and the rest, paste in an API key. That use is billed to you, and only your own plan limits it.
Which agents can I run?
Claude Code, OpenClaw, Hermes, Codex, OpenCode, and anything else you can install. It is a real computer. Use your own account, or ours.
Can I build my own apps?
Yes, and the kit works without the glasses, so you can start today.
Already have Relay Cloud? Your account is at console.relay.glass.