$ memista

Prototyping vector search

memista is a fast way to prototype vector search: cargo add, run the server, insert vectors, query. Browsable OpenAPI docs and plain-file storage make it easy to inspect and throw away. Best under ~100k vectors.

For: Engineers prototyping retrieval before committing to a stack

Before committing to a vector database, you want to see retrieval working on your own data. memista is a fast way to spike that.

Why it’s good for a spike

  • cargo add memista, run the server, and you have three endpoints on 127.0.0.1:8083.
  • Browsable OpenAPI docs at /swagger, /redoc, /rapidoc, /scalar let you poke the API without writing a client.
  • Storage is a .db and a .usearch file — inspect them, or delete them and start over.

Then decide

If the prototype confirms you need vector search at a scale beyond memista’s ~100k tested vectors, you’ve learned that cheaply and can move to a distributed store with confidence. If it confirms a single binary is the right shape, you’re already done. Compare the options: vs USearch, vs pgvector, vs Qdrant.

Related: quickstart, features.

Frequently Asked Questions

Is memista meant for production?

It is experimental (v0.1.x) and tested under ~100k vectors. It is excellent for prototyping and small production workloads where a single binary and two files are the right shape; validate carefully before larger scale.

Related

Try memista

A single crate, GPL-3.0. Two files on disk and three endpoints.