ADR-00022: SearchIndex and VectorSearchBackend stay separate interfaces even if a future engine could serve both¶
Context¶
SearchIndex (FTS5) and VectorSearchBackend are two distinct Python interfaces. An engine bundling both capabilities natively exists in principle (LadybugDB was raised as a candidate: native full-text search and a vector index in one).
Decision¶
The two interfaces remain separate regardless of which concrete engine(s) back them. A single engine may still end up implementing both later — interface separation and implementation consolidation are different axes — but only if it earns that through the same diligence sqlite-vec received (verifying its FTS engine's fidelity, its vector index type and persistence guarantees, not just its license and star count).
Alternatives considered¶
- Coupling functions together behind one engine (e.g. one engine serving both full-text and vector search) — rejected: any future hosted/remote swap of either piece would then require a provider supporting all the coupled functionality, narrowing the option space for no necessary reason.
Consequences¶
None beyond the Decision above — no engine has yet earned bundled adoption through the diligence bar this ADR sets.