USE CASES
Legal analysis. Financial guidance. Medical information. Technical architecture. When the decision matters, counsel it.
A SaaS company reviews a vendor agreement for liability exposure. A single model misses a force majeure clause buried in an addendum. The council catches it because three of four models flag the inconsistency against standard terms in the same industry. The 22% disagreement contained the most valuable insight.
A research team analyses a mid-cap industrial for a long position. Single-model recommendation: buy. Council recommendation: buy on a 12-month horizon with one dissent on inventory-build risk that the synthesis converts into an entry-price ceiling. Same direction, materially different position sizing.
A health platform generates drug-interaction explainers. A single model produces clean prose but misses a contraindication for patients on a specific anticoagulant. The council catches it: two members independently flag the interaction; the chairman incorporates it as a prominent caveat at the top of the explainer.
An engineering team evaluates microservices vs. modular monolith for a 30-person startup. Single-model recommendation: microservices. Council recommendation: modular monolith with two named extraction triggers (team-size threshold, deploy contention). The synthesis turns a binary recommendation into a decision rule.
Use a single model — for simple queries, chat, translation, summarisation. Fast and cheap. One perspective is enough.
Use a router — for high-volume workloads where cost matters more than depth. Let a classifier pick the right model.
Use Counsel — for decisions that compound. Where a hallucination would be costly. Where the disagreement between models is the signal you need.
Any decision where a single hallucination would be costly to undo: contracts, investments, patient-facing content, irreversible system-design choices.
Yes, but it's overkill. The honest framing: route easy queries to a single model, send high-stakes ones through the council.
Yes — the Code Review preset combines models with complementary strengths across paradigms. Catches bugs one model would miss.