FAQ

Is vcal-core a database?
No. It’s an in-process vector index library. Embed it directly into your service for fast, local semantic search.

Cosine vs Dot?
Use Cosine for most text embeddings. Use Dot if your embedding pipeline is designed for it.

Can I delete items?
Yes: delete(id) marks the item as a tombstone (soft delete). Deleted items are skipped during search and are physically removed during rebuild or snapshot re-save.

How big can the index get?
It is limited by your process memory (RAM). For larger or persistent deployments, consider using a higher-level system like VCAL Server.

Does it support SIMD or hardware acceleration?
No. As of v0.1.2, the library is fully safe Rust (#![deny(unsafe_code)]) and does not use SIMD. This prioritizes portability, safety, and predictable behavior across platforms.

Batch search?
batch_search(&[&[f32]], k) allows multiple queries in a single call, reducing per-query overhead.