They answer alone
Every member gets your question and nothing else. No model sees another's reply first, so nobody is anchored on whoever happened to go first.
A Caligo Arts product
Put a question to Claude, GPT, Gemini and Grok together. They answer independently, read each other, correct what is wrong or out of date, and come back with one answer — plus exactly where they disagreed. No subscription: pay for one session, or make a free account and ask a question a day.
The table will hold this to exactly the standard it holds each other to. If you are wrong, it will say so.
How a round works
Not one model asked four times. Four different companies' flagship models, each answering on its own before it has seen anybody else's work — which is the only way four answers can genuinely disagree.
Every member gets your question and nothing else. No model sees another's reply first, so nobody is anchored on whoever happened to go first.
Now every member reads every answer, including its own, and says plainly what is wrong, out of date or risky. This is the round you cannot get from one chatbot — and it is where most of the value is.
A single answer, written from the transcript, with the count of who agreed. The count is arithmetic, not a model's opinion of its own conversation — and a tie is reported, never quietly resolved.
Why it is worth your time
Most good ideas die because there was nobody to think them through with. Not nobody at all — nobody who wanted to talk about that, right then, without a stake in the answer. This is a think tank you can open at eleven at night: a brain trust for the complicated ones, where you can state your position and have it argued with properly.
No ego, no envy, no politics, no one quietly steering it towards the thing that suits them. Four models with no stake in the outcome and nothing to protect, arguing about whether the idea is any good.
Add your own position and the table is instructed to hold it to the same standard it holds each other to — and to say so plainly if it is wrong, out of date or risky. It is not there to agree with you.
Members that can search the live web do, on questions where the answer goes stale — so you find out the thing you were about to build shipped last spring, before you build it.
Four subscriptions to compare them properly, or one session here. Same question, four replies, side by side — you will work out quickly which one you actually like talking to.
A single assistant that is confidently wrong sounds exactly like one that is right. Three models correcting the fourth does not.
If something genuinely would change the advice, the table asks — after it has settled everything it can between itself, not by interrogating you a question at a time.
What it costs
Several questions at the table, open for 24 hours. No account, no subscription, no card kept on file — it is a one-off, and it is the whole thing.
Payment is handled by Stripe on their own page. No card detail ever reaches this site.
Make an account and the table takes one question a day, on the house. No card and nothing to cancel — it is the same four models, and nothing is held back.
When the council should know who you are
This page deliberately knows nothing about you. That is right for a question you can state in a paragraph — and wrong for the ones worth having a council for, which are usually about your own plans, your own week and the thing you decided three months ago and half remember.
Tick exactly which of your notes the table may read — and see, in a glance, what it can see. Nothing goes to a model that you did not put there.
Every round is written into a note in your own vault, so you can scroll back to what the table said in March, edit it, search it and carry it forward.
The council reads its own history, so the next session picks up where the last one left off instead of meeting you for the first time again.
Watch four of them answer the same question and you find out quickly which one you actually like thinking with. Inside AIOS you can open a one-to-one with that member and keep the conversation going — without four subscriptions to four companies.
The ordinary prompt box lets you address one model by name — type “Hey Claude” or press its chip — and keeps the cheap, fast default for everything else. The council is the expensive, careful option, for the questions that deserve it.
A daily note that becomes your tasks, your calendar and a recommended plan, with no syntax to learn. The council is the part people talk about; this is the part they use every morning.
AIOS is the rest of it: a daily note that turns into your tasks, your calendar and a recommended plan without you learning any syntax — with the council inside it. $15 a month, or $144 a year, and the council is included.
Already on a plan? Sign in and this page uses it — no session to buy.
The questions people ask
Whichever is currently the strongest model each provider offers. They are named on the chips at the top of this page, with the exact model id each seat is sending — so you can see what you are getting before you spend a round, rather than taking our word for it.
Not to pay. A session is bought and used with no account at all. An account is what the free question a day is attached to — four flagship models answering is the most expensive thing this site does, so the one it gives away is tied to somebody rather than to a browser. An account is also what context control needs, because the table reading your own notes means there has to be somewhere to keep them.
It goes to the models, and the last couple of rounds are kept on our server so the conversation can continue. There is no email, no name and no account attached to it, and the session is deleted after it expires. We do not keep your IP address.
Because they answer before they have seen each other's work. A relay — where each model reads the one before it — produces four versions of the first answer. This does not.
You are told they are split, and what the split was about. The council cannot break its own tie; that decision is yours and it says so.
No — and that is worth saying. A question with one right answer does not need four models. This is for the ones you are turning over: is this worth doing, what am I not seeing, is this already solved.