LanceDB: Native Full-Text Search (FTS) for S3 + Hybrid Search Weighting & Table-Optimization

[AmirBraham]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Add native full-text search (FTS) support, tunable hybrid search, and one-line table-maintenance helpers to the LangChain ↔ LanceDB integration:

• FTS baked into the LanceDB VectorStore (no external Tantivy wrapper) that stores its shards alongside .lance fragments and works the same on local disk and S3. Current FTS indices are local-only, which blocks remote workflows.

• Simple, first-class hybrid queries exposed via search_type="hybrid"

• optimize_table() helper that compacts fragments, vacuums deletes, and (optionally) rebuilds vector + FTS indices—saving users from writing their own maintenance scripts.

Motivation

Almost every LangChain how-to about hybrid search uses external retrievers or ensemble hacks. A built-in solution would match other providers (Elastic, Astra DB, Qdrant), also many teams persist Lance tables to S3; without remote-friendly FTS they must deploy a separate search engine.

Proposal (If applicable)

No response