Which models are supported in the council?

The skill supports a variety of top-tier open-weight models including DeepSeek V3.2, Llama 4 Maverick, Qwen3, and others available via Fireworks AI.

What is an LLM Council?

An LLM Council is a deliberation pattern where multiple AI models provide independent answers, rank each other's work, and a final model synthesizes the results into a consensus answer.

Can I see the individual responses from each model?

Yes, the skill is designed for full transparency and will display the raw responses and rankings from every participating model before the final synthesis.

Who acts as the 'Chairman' in the deliberation?

You can select the Chairman model via an interactive prompt; it is the model responsible for reading all peer rankings and synthesizing the final output.

Do I need a Fireworks AI account to use this skill?

Yes, this skill requires a Fireworks AI API key to perform high-speed, parallel inference across various open-weight models.

LLM Council

Name: LLM Council
Author: dair-ai

bydair-ai

•

데이터 과학 및 ML

Orchestrates a multi-model deliberation process using Fireworks AI to generate high-quality, synthesized answers through consensus and peer ranking.

The LLM Council skill implements Andrej Karpathy's 'Council of LLMs' concept directly within Claude Code, leveraging the speed and affordability of Fireworks AI. It coordinates a three-phase workflow where multiple open-weight models (like DeepSeek, Llama, and Qwen) respond to a query in parallel, rank each other's anonymized outputs, and finally submit their findings to a 'Chairman' model for synthesis. This approach is ideal for complex reasoning tasks, reducing individual model hallucination and providing a balanced, multi-perspective conclusion that surpasses the capability of any single open-weight model alone.

주요 기능

01Parallel multi-model inference via Fireworks AI

02Final response synthesis by a user-selected Chairman model

03Interactive model selection supporting DeepSeek, Llama, Qwen, and more

0485 GitHub stars

05Anonymized cross-model peer ranking and evaluation

06Full transparency with raw response logging and ranking displays

사용 사례

01Reducing individual model bias through collective deliberation and peer review

02Benchmarking and comparing multiple state-of-the-art open-weight models

03Generating high-consensus answers for complex technical or philosophical queries

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dair-ai/dair-academy-plugins llm-council

For use in Claude.ai and ChatGPT

Download Skill