It prompts your agents better than you do.
Relay knows every command and the best prompting practice for Claude Code, Codex, OpenClaw, Hermes and OpenCode. It runs them on your computer, asks when it needs you, and keeps going.
No subscription · Open source
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.
Two cameras, two speakers, one mic.
Say its name and talk. The answer comes back in your ear. No screen, and nothing runs on them.
Talk to them from anywhere.
Say the name you picked. It reaches the session you left open on your desk, the same one, still running, hands free.
They keep working after you leave.
Start a job and walk out. A build, a test suite, a refactor across forty files, it tells you when it is done.
Your day becomes their context.
The glasses are on, so the meeting, the hallway promise and the number someone said out loud are already written down. That is the context your agents never had.
You are not briefing them any more. They were there.
It goes where you go.
The capture does not stop at your desk. Walking, driving, in a client meeting, it is the same memory filling up.
It stays on your computer.
Notes and audio land on your own machine. Read them, export them, or wipe them. We throw the audio away once it is written down.
Every agent can read it.
One memory the whole set shares, so what you settled this morning is there tonight, in a different agent, without saying it twice.
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, and sets them up for you.
| Ray-Ban Meta | Relay | |
|---|---|---|
| Music, calls, photos | Yes | Yes |
| Runs your coding agents | No | Yes |
| Sets them up for you | No | One command |
| 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 |
| Open source | No | Yes |
Running in three steps.
About ten minutes, once.
curl -fsSL https://relay.glass/install | sh
| Runs on | macOS or Linux, arm64 or amd64 |
|---|---|
| The one requirement | It does not sleep. A Mac mini, a Linux box, a laptop with the lid open |
| Disk | About 300 MB, most of it the local embedder |
| Agents | Whichever you already have. It finds them |
-
Step 1
Run the command.
On your own computer, or any machine you leave on. It sets itself up and prints a code.
No machine to spare? We run one for you, $39 a month, tokens included.
-
Step 2
Install the app.
Type in the code it printed. No account, nothing in between.
-
Step 3
Put the glasses on.
They pair with your phone. Pick a name for your agent, then just say it.
The glasses are yours to program.
Every control is exposed, all 92 commands in the protocol. Build something that drives the glasses, drives your own server, or that nobody has thought of yet.
Paste it into any agent you already run. It carries the whole SDK.
Just glasses.
No screen, no glow. Nobody can tell.
What's inside.
| Storage | 4 GB, days of audio |
|---|---|
| Cameras | Two, 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 | $249, and nothing after it |
Pick your lens.
All three fit every Relay. Same price, chosen at checkout.
Sunglasses
Dark all the time.
Photochromic
Clear inside, dark in the sun.
Blue light
Clear. Cuts screen glare.
Founders Edition.
The first hundred, then the price goes up.
The orchestrator runs on your machine, with your accounts and your keys.
One payment. Nothing after it.
- Relay One glasses
- All the software, free
- Runs on a computer you already own
We keep a machine awake for you, with the models included.
First year included, then $429.
- Everything in Relay One, plus
- A year on a machine of ours
- Tokens included, first month free
- A heavy month moves you to cheaper models. We never cut you off
The software is free because the hosted plan pays for it. Run your own machine and you pay us once, then never again.
Check it before you buy it.
The whole stack is on GitHub. Clone it and run the tests right now, with no glasses attached.
If we disappear tomorrow, the daemon keeps running and you still have the source.
$ relay setup pick models, start on boot detect which agents are installed doctor re-check every credential pair print a pairing code status what is configured $ npm test tests 292 suites 56 pass 292 fail 0 duration 423ms
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.
Do I need the $39 plan?
Most people do not. If you own a machine that stays on, run one command and pay us nothing, you bring your own agent accounts. The plan is for people with no machine to spare: we run one, and the models come with it. There is nothing in between.
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.
What do I call it?
Whatever you want. You pick the name when you set it up, and that name is what wakes it. Say it and start talking. If your hands are free and you would rather not say a name, tap the temple instead.
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.
Where do my words go?
To your machine. You can read them, save them or wipe them. We throw the audio away once it is written down.
Can I build my own apps?
Yes, and the kit works without the glasses, so you can start today.