Certamen

Tournament-based AI decision synthesis framework where models compete and critique each other to extract maximum knowledge.

Installation

pip install certamen-core

Quick Start

# Auto-discover Ollama models
make discover-ollama

# Run tournament
certamen --config config.yml

Configuration

config.yml is a slim file: question + models + workflow reference.

question: "What is the best approach to ..."

models:
  gpt:
    provider: openai
    model_name: gpt-5.2-pro
  claude:
    provider: anthropic
    model_name: anthropic/claude-opus-4-6
  gemini:
    provider: google
    model_name: gemini/gemini-3.1-pro-preview
  grok:
    provider: xai
    model_name: xai/grok-4-1-fast-reasoning

workflow: diamond-tournament

The workflow defines the tournament logic; config.yml only declares what to ask and who to ask. Built-in workflows ship with the package (diamond-tournament); custom workflows live under ~/.certamen/workflows/ or can be passed via -w path.yml.

Architecture

See CLAUDE.md for detailed architecture and development guidelines.