I built [VectorLiteDB (https://github.com/vectorlitedb/vectorlitedb)
— a simple, embedded vector database that stores everything in a single file, just like SQLite.
The problem:
If you’re a developer building AI apps, you usually have two choices for vector search
– Set up a server (e.g. Chroma, Weaviate)
– Use a cloud service (e.g. Pinecone)
That works for production, but it’s overkill when you just want to:
– Quickly prototype with embeddings
– Run offline without cloud dependencies
– Keep your data portable in a single file
The inspiration was *SQLite* during development — simple, local, and reliable.
The solution:
So I built VectorLiteDB
– Single-file, embedded, no server
– Stores vectors + metadata, persists to disk
– Supports cosine / L2 / dot similarity
– Works offline, ~100ms for 10K vectors
– Perfect for local RAG, prototyping or personal AI memory
Feedback on both the tool and the approach would be really helpful.
– Is this something that would be useful
– Use cases you’d try this for
Comments URL: https://news.ycombinator.com/item?id=45319922
Points: 1
# Comments: 0
Source: github.com
