— 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
nenenejej•4mo ago
However the "sqlite" approach is also cool. Docker while handy is another thing to fiddle with and sqlite is cool for it's simplicity of deployment. Single file for the executable (or use a driver library) and single file for DB.
lokahdev•4mo ago
The inspiration was SQLite, a tiny, embedded, zero-ops, single-file database you ship inside any app, not a networked server you operate next to an app.