Token-aware chunking for RAG, in one API call

Naive chunking splits on a fixed character count and slices sentences — and sometimes words — in half. Embeddings of half-thoughts retrieve poorly.

Split where the document already breaks

/v1/extract with chunk: true segments markdown on heading boundaries, packing each chunk up to your chunk_max_tokens budget with a configurable chunk_overlap_tokens so context carries across the seam.

You get back an array ready to embed — no parsing library, no recursive splitter to tune. Pipe meta.chunks straight into your vector store. Details in the Extract reference.