Jasper

6 chapters · index reference

Concepts

These six chapters describe one Jasper index from the inside out. They read in order — the graph chapter assumes you know what a segment is, the scoring chapter assumes you know how the graph answers a query — but each one also stands alone if you already know the vocabulary and want a single reference.

Every chapter that touches the on-disk format or the query path includes working Rust and Python, using the same catalog example throughout: a 768-dimensional embedding collection with roughly four million rows.

  1. Index Anatomy 2 min read What a .jasper file actually contains: manifest, vector store, graph store, and write-ahead log.
  2. Building the HNSW Graph 2 min read Layers, M, and ef_construction: how vectors join a hierarchical navigable small world graph.
  3. Quantization Tiers 2 min read Trading precision for memory with scalar and product quantization, and reranking to get it back.
  4. Hybrid Keyword Scoring 2 min read Fusing an inverted-index keyword score with vector similarity into one ranked list.
  5. Query Tuning 2 min read The ef_search, oversampling, and filter knobs that trade recall against latency at query time.
  6. Persistence and Recovery 2 min read The write-ahead log, checkpoints, and what happens the moment Jasper reopens after a crash.