Lead and team
A lead splits the job up, members work on their parts at the same time, and the lead writes one finished report.
Good for research with several angles.
The work you keep
meaning to get to.
Write down a job once — who does what, where you want a say — and it runs the same way every time. Several teams, in order, stopping wherever a person has to decide. It runs on your Mac, on your files.
Version 1.1.3 · Apple silicon Mac (M1 or newer) · macOS 13 or later · Invite code required
Four ways to build a team
Every team hands you one finished result. Choose the shape that fits the work.
In sequence
The work passes through each member in order — writer, editor, proofreader. The last one reviews it and can send it back once.
Good for content that needs a set process.
One agent
One agent understands the task, plans it, does the work, and reviews how it went — keeping lessons for next time.
Good for a self-contained project.
Round table
Every member works on the same problem and answers what the others said, for the rounds you choose, or until the one writing it up says it is ready. Then that member writes the finished result.
Good for a decision that needs several viewpoints. Everyone speaks every round, so it costs more.
Write a procedure down once
Name the steps, the team for each, and where a person has to approve. Edit it later — the name, the budget, the steps themselves.
It runs while you are away from the desk
Start it and walk away, or set a routine to run at the times you choose. Come back and Today tells you, in plain sentences, what ran, what it cost, and what needs you. The Mac has to be awake — leave it open and plugged in. Work does not continue overnight with the lid closed, and we would rather you heard that here than on the first night.
And says so when it doesn't
A routine that was refused says why, instead of sitting there looking like it will run tomorrow.
Email, once you allow it
Connect a Gmail mailbox (only Gmail, for now) and choose what may be read and what may be sent, per agent, with limits per hour and per day. Off until you turn it on. Every message says it came from an agent.
Your files, your machine
Teams, notes, drafts and the record live in a folder on your Mac. Shell commands run inside the operating system's own sandbox, or they do not run.
One switch stops everything
Pause all work at once. Nobody is exempt — not a routine, not a procedure halfway through.
Get the app
Agent Platform is invite-only. A code opens the download; without one, join the waitlist and we'll be in touch. Free for now, including the model usage, up to the included credits set on your code — the figures you see in the app are what that work cost. Pricing comes later.
Install in a few minutes
-
Download
Enter your invite code above to start the download (a .zip file, for Apple silicon Macs). Open it from your Downloads folder. Safari usually opens it for you, so you may find Agent Platform there already instead of the .zip.
-
Move it to Applications
Drag Agent Platform into your Applications folder, then open it from there.
-
Allow it to open
The app isn't signed by Apple yet, so the first time macOS says it can't check it. Close that message with Done (or Cancel, on macOS 13 and 14) — not Move to Trash. Then open System Settings → Privacy & Security, scroll down, and click Open Anyway next to Agent Platform. The button is there for about an hour after you tried to open the app; if it isn't, open the app again first. macOS asks once more: choose Open Anyway (Open on macOS 13 and 14) and, if asked, enter your Mac's password. You won't be asked again.
-
Enter your invite code
Type the code you were sent, your name and your email. That's it — no AI account or key needed. A code has an end date; if it has passed, ask whoever sent it for a new one.
Then what?
Getting started takes it from there: what to set up before anything useful happens, the first thing worth trying, what a rehearsal is, and what to do when something is refused. Read it before you start, or when you get stuck — a tester who can't get started is the most useful message we can get.
Updates
When a new version is out, the app tells you. Signed in with your invite code, press Install update in Settings: the app downloads it and checks it was signed by us. When it's ready, press Restart to install, and it opens again as the new version. The old version goes to the Trash once the new one has started, and if the new one doesn't start, the app offers to go back to it. Your chats, teams and saved work stay exactly as they are. What changed in each version is on What's new.
Why not just a chatbot?
A chatbot answers a question. This does a job you would otherwise do yourself, the same way every time, and stops where you want a say.
Try it with everything switched off
A rehearsal runs the whole job for real — the real reading, the real drafts — with nothing able to write a file, run a command or send a message. You see exactly what it would have done, then decide.
Zapier and Make both document that testing a workflow runs it for real. This does not.
Checked by something that isn't itself
Add a step where a second team checks the work before it moves on — a team on a different AI company's model. One system marking its own homework is not a check.
It sends work back. The example at the top of this page is what a refusal looks like.
It stops at your desk
Put an approval anywhere in the job. Nothing goes out through it until a person says so, and the record shows who that person was.
Only people approve. There is no tool an agent can call to approve itself.
A ceiling that counts money you haven't been billed for
Set a limit per job. It is checked at every step, it counts what is spent but not yet charged, and it holds when several agents work at once.
It applies to models you pay for on your own account too.
A record you can check, not a log you're shown
Every action is written to a chain where each entry seals the one before it. Change one and the chain says so. It names who asked, what ran, and what it cost.
Export it as a spreadsheet for an accountant, or anyone who needs to check it.
Don't take our word for any of it
Each of the app's safety claims is paired with the tests that would fail if it stopped being true. One command runs them and prints every claim proven or unproven — no account, no key, no internet.
See what it printed for this version, dated, from the machine that built the download.
That is still our run on our machine. There is no public repository yet, so the source is not something you can go and download — ask us for it and we will send you the tree this release was built from, so your own IT can run it.
evanwm24@gmail.comopen in your mail app
Privacy, plainly
Saved on your Mac. Your chats, teams, files and saved work are stored on your Mac, not on our servers.
Sent to do the work. When you ask for something, the request — including any files you attach — goes to Anthropic, the company that makes Claude, through our access service. That service counts usage against the limit on your invite code and does not keep your conversations.
Other people's material, in those words. If an agent reads someone's note, their contract or their email in order to do the job, that text goes to Anthropic with the request — the same as anything else you hand it. Anthropic's own terms govern what they keep and for how long. This app cannot change that and does not claim to. What does not leave your Mac is everything around it: their files, the drafts, and the record of what was done for them. Our own server keeps none of that material — it counts tokens and money, and keeps only what you choose to send it from the app: feedback, and a problem report you have read first.
If you join the waitlist, we keep your name, email and note only to contact you about Agent Platform. Ask us and we'll delete them.
What we would want to know, in your position
The documents this app ships with have a section called "what is not true yet". It would be strange to leave it out of the place you are actually reading.
Who runs this, and what if they stop
Agent Platform is built and operated by one person. Its access service — the part that checks your code and passes your work to Anthropic — runs on a single server, with no formal uptime commitment.
If it went away tomorrow, the app on your Mac keeps working on your own Anthropic key: your teams, your files and your record are already on your machine, and were never anywhere else.
What it does not do yet
The Mac has to be awake and plugged in; close the lid and the work stops. Apple silicon Macs only — not Windows, not a phone, not a browser. No SOC 2, no ISO 27001, and no outside security review of any kind, so if your insurer, or anyone you work for, asks for one, we do not have it. No SSO and no SCIM: people are managed inside the app, by email address. An install signed in with an invite code sees Claude's models only — putting the checking team on a different AI company's model needs a gateway key we set up with you.
And the record proves that nothing was edited afterwards. It is not signed, so it cannot prove the machine that wrote it was honest.
Questions?
Stuck on the install, or not sure whether it is for you? Write to us — a person reads it.
evanwm24@gmail.comopen in your mail app