Knowa – Open-Source LLM Context Optimizer (github.com)

Knowa has launched as an open-source LLM context optimizer designed to enhance the efficiency of AI-powered applications by dramatically reducing the amount of irrelevant context sent to language models (LLMs) in production environments. Instead of processing entire large documents, which for a 1,000-page knowledge base could mean handling 2–4 million tokens and incurring significant costs, Knowa indexes documents once and retrieves only the most relevant chunks for each query. This approach achieves a 90–99% reduction in the token count transmitted to the LLM without compromising answer quality, making it a critical tool as AI applications transition from prototypes to widespread use across organizations. The architecture of Knowa allows it to function as both a hybrid retrieval library and a knowledge base server, ingesting documents from various sources and storing them effectively in a PostgreSQL database. Key features include hierarchical chunking for precision, an entity-rich knowledge graph built during indexing for advanced retrieval capabilities, and the option for entity enrichment using LLMs in a cost-effective manner. With Knowa, teams can manage document indexing and querying through a user-friendly CLI or REST API, empowering them to sustain scalable AI features while significantly optimizing their querying costs and response times.