LLMSlim
Open-source context compression for RAG and AI agents
8 followers
Open-source context compression for RAG and AI agents
8 followers
Open-source (MIT) Python library that compresses what an LLM sees (retrieved documents, chat history, tool output) to a hard token budget. It selects query-relevant units, lexically or with an optional 22M cross-encoder, instead of rewriting tokens. In its own reproducible benchmark (Claude Haiku 5.5 reader, NaturalQuestions, ~25% budget) it is level with LongLLMLingua at about 100x the speed per question; full context still scores higher. pip install llmslim
