Motet is an agent that sits above your agents. It listens to the Claude Code, Codex and Cursor sessions you are running, tells you which one needs you, and cues the right one when you say the word.
The demo opens in your browser, with made-up agents. Motet itself is free, open source, and runs on your Mac.
The room, with Motet ready at the bottom. Shown with the built-in demo: the agents and projects are made up.
A motet is a piece of music for several independent voices. That's what running agents is. Motet is the one listening to all of them.
Motet is an agent of its own. It follows what every one of your agents records, so you ask one assistant instead of opening six windows. And when you want something done, it cues the right desk.
Motet follows each agent's log as it is written: what it was asked, what it ran, what it changed, where it stopped.
What did I miss? Is anything stuck? Who changed this file? It answers from the records, links what it used, and says what it can't confirm. In Auto mode it picks the model for you: a quick one for everyday questions, a stronger one for summaries and handovers.
Ask Motet to put a question to an agent. It carries your words to that desk and brings the answer back. On verified versions of the Codex app it can also steer or stop a running turn.
"Ask Esme whether the old rows need migrating too."
Finds Esme's desk, sends your question, and shows whether it was delivered.
Answers in her own chat, where the work is.
Her answer arrives in Motet, marked as her words and not Motet's guess.
Motet acts only when you ask. It does not yet act by itself; that is the biggest item on the roadmap. A cue is queued for the agent and does not interrupt it. Cueing works with eligible Codex chats and with Claude Code chats that have the plugin connected. Questions and cues through Motet use your own Anthropic API key.
Your Claude Code chats and your Codex chats live in separate apps. Motet gives them a way to talk. One agent asks another a question in plain words and the answer comes back to it, so the work carries on without you copying notes from window to window.
"Ask Bram whether the rounding fix passed review."
You tell Ada, in her own chat"It passed. One thing: the table still rounds each item."
Bram answers, in its own chatThe answer is filed against Ada's question, and she collects it when she needs it. It does not wake a chat that has stopped.
It works between eligible Claude Code and Codex chats on your Mac today. Agents from the same company keep using that app's own tools.
An agent contacts another only when you have asked it to. A message from another agent never gives it new permissions.
A question that is waiting, failed or answered is shown as what it is. Requests and replies are kept for 24 hours.
The bridge is optional. Set it up once in each app; a Claude Code chat also needs the plugin connected and agent messaging switched on. Connected agents can read the shared record, so keep secrets out of it.
Motet is not another model plan, and it does not sit between you and one. Your agents keep running in their own desktop apps, on your own Claude, Codex, Cursor and Antigravity subscriptions. Motet connects to those apps on your Mac.
Use Claude Code, Codex, Cursor and Antigravity the way you do now. The work still happens in those apps, under whatever plan you already use them with.
It reads the session records those apps already write on your Mac, and it reaches the apps through their own desktop connections. It does not ask for your logins to them.
Every session sits together, whichever company's agent it is. Ask Motet about any of them, and cue the ones it can reach.
The one separate cost is Motet's own key. Questions you ask Motet, and cues you give through it, use your own Anthropic API key, and Anthropic bills for what they use. Watching, messaging an agent directly, sending files and starting chats do not need it.
Your recent sessions on this Mac, grouped by project, each one showing what it is doing right now.
When an agent stops to ask permission or a question, it rises to the top and the bell tells you. You see the exact command it wants to run before you say yes.
Open an agent's desk. The conversation is on the left and its work is on the right: the edits, the commands, the files and the pages it opened. Long runs of steps fold away until you want them.
Write to an agent from its desk. Motet says plainly whether the message was queued, delivered or never confirmed.
Drop a PDF or a picture on an agent. You see where the copy goes before it is sent.
Press the plus beside a project to open a fresh Codex chat in that folder. More apps are on the roadmap.
We would rather name the gaps than let you find them. This is what works today, what doesn't yet, and what we are building toward.
A loop of its own. Motet watches for the cue you care about, follows up with an agent, plans the day with you, and keeps two agents working on the same piece in step. You stay the conductor and decide how much it may do.
A roadmap is a direction, not a promise or a date. The full version, with what has been proven and what is only an idea, is in the repository.
You already have tools that say "done" when they mean "started". Motet shows what the records contain and says so when they run out.
A command with no recorded result is shown as started, never as done.
"Probably needs you" means probably. It is worked out from what the agent said, and it can be wrong.
"No log entry for ten minutes" means the log is silent, and nothing more.
An agent's private thinking isn't in its log, so Motet never shows or paraphrases it.
Motet reads the session logs your agents already write, and it never changes them. It listens only on your own computer, behind a private key kept in your user folder.
The Motet app sends none of this to us. It runs only on your computer, and there is no Motet cloud.
You need a Mac with Git and Python 3.9 or later. Start with the demo, then open your own room.
$git clone https://github.com/glo6al/motet.git && cd motet$python3 src/server.pyMotet onhttp://127.0.0.1:8793/— open it and press Unlock
Two commands. No packages to install, no build step.
Open 127.0.0.1:8793/?demo. Made-up agents on made-up projects. Click everything; nothing is sent anywhere.
Open 127.0.0.1:8793 and press Unlock. Sessions active in the last twenty minutes appear.
Yes. Motet is open source under the Apache 2.0 licence: free to use at home and at work, and free to change. The licence does not grant rights to use the Motet name or logo as your branding.
No. Your agents keep running in their own apps. Motet watches them and carries your messages, files and answers to them.
It runs on macOS today. On Windows only the automated tests have run so far, so treat it as untested there.
Not for your agents. Motet works with the desktop apps you are already signed in to, on your existing subscriptions. Only Motet itself uses an Anthropic API key: for its answers and its cues, which Anthropic bills by use, and to fetch the list of models. Skip the key and everything else still works.
No. It follows coding sessions that were recently active on this computer. It does not import your account history.
Not yet. Motet acts when you ask: it answers, and it carries your message to one agent at a time. A loop of its own, where it follows up and coordinates without being asked, is the biggest item on the roadmap.
Motet is open source under the Apache License 2.0. Use it at home or at your company, change it, and build on it.
Tell us what would help: getting set up, support, or something Motet does not do yet.
Your message goes privately to the people who make Motet. Nothing is posted in public, and you are not added to a mailing list. How we handle your details.
Still not working? Write to hello@usemotet.com.
Thank you, . We have your message and will reply to .
Nothing was posted in public, and you were not added to a mailing list.
Every agent in view. A clear cue when one needs you.