Llm Council
The LLM Council works together to answer your hardest questions
Multiple LLMs debate and peer-review each other to produce a trustworthy answer. The peer review is anonymized to prevent bias among models. Confidence scores are calibrated using isotonic regression against real human dispositions.
Per-model and per-stage token and cost accounting is provided on every response. The product runs as a single package with four surfaces: MCP server, HTTP API, Python library, and CLI. It supports any model and router, including GPT, Claude, Gemini, Grok, DeepSeek, and local Ollama via various providers.
The process has three parallel stages: deliberate, anonymized peer-review, and synthesis.
Summary drafted from the project's own website. Every sentence is backed by text on that page and was reviewed before publishing.