Prompt Analytics

Prompt Analytics turns collections of LLM prompts into reusable knowledge about what works and when. Our system, Facet, identifies recurring guidance in prompts and measures how its effects vary across models and tasks. Drawing on 4,475 prompts and 6.5 million controlled interventions, we show how this evidence can improve existing text-to-SQL pipelines and guide automatic prompt optimization.

The library now have >35K prompts and is growing. Use Facet locally to analyze the prompts your agent is making, use our web service to optimize your prompts, and contribute to the library!