The Council
A multi-agent deliberation engine. Point it at a decision and a panel of models argues it to consensus — with the dissent kept on the record — before you commit.
01 / The build
The argument, not just the answer.
A single model is a confident narrator of one point of view. The Council seats several models at a table, gives each a stance, and lets them read and rebut each other across rounds until the panel converges — or deadlocks on the record.
You get back a verdict, a calibrated confidence, and the dissent that didn't make it — so you can overrule the room knowing exactly what you're overruling.
Why I built it
For decisions that matter — architecture calls, hiring, spend — you want the argument, not just the answer. Copy-pasting a prompt between three chat tabs and diffing it by hand doesn't scale past the second question.
02 / Design decisions
The trade-offs are part of the story.
01Seated debate rounds
Each model holds a stance and reads the others before answering. Perspectives compound across rounds instead of collapsing into one model's bias.
02Dissent on the record
The verdict ships with the objections that lost. You see what the panel overruled, not just what it agreed on.
03Confidence scoring
Every synthesis carries a calibrated confidence — a signal for when to trust the room and when to overrule it.
04Runs on NautGate
Providers are routed through a single OpenAI-compatible gateway, so the panel mixes models without leaking keys into the app.