dair-ai LLM Council (Fireworks AI)
Detailed Guide
Read the detailed guide before executing this skill. It retains the complete procedure and reference material. Treat its safety, prerequisites, and validation requirements as mandatory. For focused work, load the relevant sections; for end-to-end work, read the guide completely.
When to Use
Use when this workflow matches the user request: Use this skill for its documented workflow.
Source: dair-ai/dair-academy-plugins (MIT).
This skill implements Karpathy's LLM Council concept where multiple open-weight LLMs deliberate on a query, powered entirely by Fireworks AI:
- Phase 1: All models respond to the query independently (parallel)
- Phase 2: Models rank each other's anonymized responses
- Phase 3: A Chairman LLM synthesizes the final answer
All inference runs through Fireworks AI using open-weight models. The speed and pricing of Fireworks makes it practical to run multi-model deliberation that would be slow or expensive on other providers.
CRITICAL RULES
- ALWAYS use AskUserQuestion to let the user select council models (multiselect) and the Chairman model
- ALWAYS save raw responses to files - never summarize or truncate API outputs
- ALWAYS show full transparency - display all individual responses, all rankings, AND the final synthesis
- NEVER skip the ranking phase - it is essential to the council deliberation process
- Read from files for display - ensures content is shown unmodified
- ALWAYS display the final output to the user after Phase 3 completes
Limitations
- Requires the upstream tool, account, API key, or local setup when the workflow names one.
- Does not authorize destructive, production, paid, or external-message actions without explicit user approval.
- Validate generated artifacts or recommendations against the user's real sources before treating them as final.