# memista

> memista is a lightweight vector search library for Rust. It combines
> SQLite for metadata storage with USearch (HNSW) for vector similarity
> search, exposing a small HTTP API via Actix-web. The project is
> experimental (v0.1.x) and is licensed GPL-3.0. Source is at
> https://github.com/Skelf-Research/memista; crate at
> https://crates.io/crates/memista.

## Key facts

- Language: Rust (1.56+)
- Index backend: USearch 2.19.x (HNSW algorithm, SIMD via `simsimd`)
- Distance metric: Inner Product (`MetricKind::IP`)
- Quantization: F32 (`ScalarKind::F32`)
- Embedding dimensions: hardcoded to 2 in current crate; users fork
- Metadata storage: SQLite (one `chunks_<database_id>` table per partition)
- Persistence: SQLite WAL + `<database_id>.usearch` index file per partition
- HTTP server: Actix-web, default `127.0.0.1:8083`
- API: `POST /v1/insert`, `POST /v1/search`, `DELETE /v1/drop`
- OpenAPI documentation via apistos at `/swagger`, `/redoc`, `/rapidoc`, `/scalar`
- Tested scale: under ~100k vectors (per project README)
- License: GPL-3.0

## Site

- Home: https://memista.skelfresearch.com/
- About: https://memista.skelfresearch.com/about/
- Blog: https://memista.skelfresearch.com/blog/
- Compare: https://memista.skelfresearch.com/compare/
- Docs: https://docs.skelfresearch.com/memista/
- Rustdoc: https://docs.rs/memista
- RSS: https://memista.skelfresearch.com/rss.xml
- Full LLM dump: https://memista.skelfresearch.com/llms-full.txt