Better Vector Search for Long Documents: Chunking Inside Manticore Search
Summary
This post explains how Manticore Search adds chunking to vector search to handle long documents. It introduces five chunking strategies (truncate, mean, fixed, recursive, sentence), shows how to configure model-backed vector columns with chunk_strategy, and provides benchmarks, recipes, and guidance for choosing the right approach. The piece also discusses limitations and practical considerations for deploying chunked search in real-world docs and knowledge bases.