itsaryanchauhan/winnow repository overview

⁠Winnow

Open-source prompt compression middleware for RAG pipelines.

Sits between your vector DB and LLM - compresses retrieved chunks using LLMLingua-2, cutting token count by ~50% with less than 3% accuracy loss.

⁠Quick Start

docker run -p 8000:8000 itsaryanchauhan/winnow

⁠API

POST /v1/compress → compress a single context POST /v1/compress/batch → compress multiple contexts GET /health → health check

⁠Links

GitHub: https://github.com/itsaryanchauhan/Winnow⁠ Docs: https://itsaryanchauhan-winnow.hf.space/docs⁠