SQL Anywhere 0.3.0 — vector-native edge

SQL Anywhere 0.3.0 — vector-native edge

SQL Anywhere is one of the few engines to ship native vector search and
bi-directional edge replication in the same file-compatible SQLite fork. 0.3.0
turns that into a batteries-included toolkit for local-first / edge RAG —
retrieval with no extra infrastructure and no network round-trip.

Highlights

• Hybrid search (vector + FTS5) — fuse DiskANN similarity with full-text
  relevance via Reciprocal Rank Fusion in one query.
• Vector quantization — compress_neighbors=float16|float8|float1bit
  shrinks the DiskANN index up to 5.5× (measured) for constrained devices.
• embed() — sqlanywhere::embed(text, dims) turns text into a vector
  inline (vector32(embed(...))), no external pre-compute. Dependency-free
  hashing-trick embedder (lexical; swap in a real model for semantic search).
• Local RAG example — cargo run -p sqlanywhere --example local_rag runs the
  whole pipeline (embed → quantized index → FTS5 → hybrid RRF) end to end.

Everything is additive and opt-in; databases that don't use the new features
stay byte-compatible with stock SQLite.

Docs

• Guide (why / how / examples): docs/VECTOR_SEARCH.md
• Full changelog: https://github.com/kwhorne/sql-anywhere/blob/main/CHANGELOG.md

Notes

CRDT offline-merge (cr-sqlite) remains on the roadmap — it needs a pinned
nightly toolchain + build-std + C linking and is tracked separately.