| GitHub: https://github.com/Hormold/cognition-wheel | npx mcp-cognition-wheel While building agent flows, I kept hitting a gap:
sequential-thinking helps models reason step-by-step — but what about cross-checking? cognition-wheel runs 3 top-tier models (Claude-4, Gemini 2.5, OpenAI o3) in parallel, anonymizes them (Alpha / Beta / Gamma), and uses a random judge model to fuse the answers into one take. The pain:
• Models contradict each other — which one do you trust? • Brand bias skews output • One model fails? You’re stuck The fix: • 3 LLMs think in parallel • Names masked to avoid bias • Random judge model synthesizes a final answer How it fits: 1. Call sequential-thinking for step-by-step reasoning 2. Pipe the result into cognition-wheel for fast consensus check
→ deep logic + fast cross-validation Built this because I wanted multi-model reasoning without orchestration hell.
Would love feedback, weird edge cases, or ideas on smarter judge selection. |