An orchestrator, not a lab
We do not train models. We call models from several AI labs (OpenAI, Anthropic, Google, Meta and Alibaba) through their providers’ APIs and put them into a structured argument.
// About
Council is a small project with one idea behind it: the questions that matter most are the ones with no single right answer, and those are exactly the questions a confident chatbot handles worst.
It started as an argument with a chatbot. Ask an AI whether to take a job, or whether an idea is worth building, and it answers in the same assured voice it uses for the boiling point of water. That voice is the problem. It hides the fact that reasonable people, and reasonable models, would disagree.
Three models are given three different jobs and answer independently. A fourth reads all three and issues one verdict, with the disagreement left visible underneath. Where a claim can be checked against the outside world, it is.
// What we actually are
We do not train models. We call models from several AI labs (OpenAI, Anthropic, Google, Meta and Alibaba) through their providers’ APIs and put them into a structured argument.
This is not a funded team. It is a working product with a free council you can use right now, and a roadmap we are honest about.
Other tools compete on sounding capable. We would rather be the one that tells you when it is a coin flip, because that is the information you were missing.
Multi-model panels are a live idea and others are building in this space. We think the gate, the roles and verification make ours different, but we did not invent the category.
// Where it is going
Shipped
Fact gate, three roles, chairman verdict, external checking. Free, no login.
Next
Every debate and verdict saved, so Council can draw on what you have already decided.
The real goal
Asking, months later, how the decision actually went, and learning which reasoning led you right.
Models are frozen; they do not learn from you. But a system built around them can remember what you decided and what happened next, and that record gets more useful the longer you use it. That is the version of Council worth building.
Questions, criticism, or a bug? Especially criticism. Write to hello@council.app. Built in Delhi.
// Plain about it
Council is built to argue. If your plan has a weak spot, one seat’s whole job is to find it.
It is a structured second opinion. For legal, medical or financial decisions, use it to get clear on the trade-offs and the questions to ask an expert.
We do not train models. We call models from several labs and structure the argument between them. What we build is the argument.
Every answer is labelled with the exact model that wrote it, so nothing on the site is a claim we cannot back up.
// How we build
A smooth consensus hides the most useful part. Where models split is often the real decision.
High, medium or low confidence on every verdict. Facts get one line; coin flips are called coin flips.
Every claim about which model answered can be checked on the answer itself.
The free preview runs the same structure as the paid plans. Paid plans add depth, not honesty.
// So far
Shipped
Fact gate, three seats, a chairman, confidence levels and the exact model on every answer. Free, no login.
Shipped
Ask the council from Claude Desktop, Claude Code, Cursor and Windsurf.
Next
Saved councils, projects, uploads and the Plus, Pro and Ultra plans.
Later
Council remembers what you decided and asks, months later, how it went.
// Questions
A small, early team in Delhi. Council is not funded by a model lab, which is why it can be honest about where each model is weak.
Through paid plans once accounts launch. There are no ads, and we do not sell your questions.
No. We use their models through their public APIs, like any other developer. Model names are their trademarks.
Use it on a real decision and tell us where it was wrong. Criticism is the most useful thing you can send to hello@council.app.