Multi-agent deliberation engine

Put a question to several models and get an answer you can audit.

Polyvise runs a question through a structured format — adversarial debate, consensus finding, or an advisory panel — using several independent models at once. What comes back is a verdict where every claim links to the source behind it and every step names the model that produced it.

Start a run →See past runs

deliberation modes

models across 6 providers

rounds per debate

run recorded

Mode registry

Choose how the models deliberate

Every mode renders into the same run surface.

Live

Hybrid Council

Two agents argue for, two argue against, and a neutral judge scores the result. Six rounds, ending in a verdict with a confidence figure.

Adversarial5 agents

Consensus

Agents answer independently, then revise over several rounds. Reports the range they settle into and which agents held their original position.

ConvergentN agents

Advisory Panel

Four named perspectives — economist, ethicist, operator, skeptic — each give advice separately. A chair writes up where they agree and where they conflict.

AdditiveLens-based

Live

Model Lab

The same prompt sent to every configured model, with the answers side by side and the latency, cost and token count for each.

ComparativeBenchmark

Engine pipeline

What happens after you start a run

Frame

Turns your question into a resolution the agents can argue for or against, and classifies the topic and its stakes.

Scout

Agents stake out positions and list their assumptions independently, before any of them see each other's work.

Evidence

Retrieves sources, grades each one primary, expert or methodology, and screens out anything promotional or undated.

Deliberate

Opening, cross-examination, rebuttal and closing rounds. Each turn records which claims and sources it rests on.

Judge

A neutral model scores five dimensions, writes the verdict, and lists the evidence that would change it.