Chonkie

Chunking is the unglamorous center of every RAG pipeline, and Chonkie packs the common strategies into one small dependency. Chunkers include token, fast, sentence, recursive, semantic, code, neural, table, late chunking, and an LLM-driven slumber chunker, with refineries that merge overlapping chunks or attach embeddings afterward. More than 32 integrations plug in across tokenizers, embedding providers such as OpenAI and Cohere, LLMs, and vector databases, while the purely structural chunkers run with no network access at all. The library is deliberately lean, about a 505 KB wheel and a 49 MB installed footprint versus 80 to 171 MB for comparable frameworks, and installs with pip install chonkie. A REST API server with Docker support opens it to non-Python consumers, documentation lives at docs.chonkie.ai, and the maintainers run a hosted cloud alongside the library. The core is MIT licensed with about 4,600 stars, and the repository now sits under the feyninc organization as the company behind it operates as Feyn Labs. RAG developers reach for it when heavyweight frameworks are overkill and tested chunking primitives are enough.