VectorDB - In-memory vector database with swappable indexing
Posted by doganarif@reddit | Python | View on Reddit | 3 comments
What My Project Does
It's a lightweight vector database that runs entirely in-memory. You can store embeddings, search for similar vectors, and switch between different indexing algorithms (Linear, KD-Tree, LSH) without rebuilding your data.
Target Audience
This is for developers who need vector search in prototypes or small projects. Not meant for production with millions of vectors - use Pinecone or Weaviate for that.
Comparison
Unlike Chroma/Weaviate, this doesn't require Docker or external services. Unlike FAISS, you can swap index types on the fly. Unlike Pinecone, it's free and runs locally. The tradeoff: it's in-memory only (with JSON snapshots) and caps out around 100-500k vectors.
GitHub: https://github.com/doganarif/vectordb
virtualadept@reddit
This is really cool - I've been looking to play around with something like this for a while now. Thanks!
Mysterious-Rent7233@reddit
That's cool but I was hoping it was going to be in-process, like SQLite or FAISS. I'd love to replace FAISS but I don't want an external server.
thallazar@reddit
I'm confused on your chroma comparison. It doesn't require docker or an account to use locally. What benefits does it offer over using chroma in a local folder capacity?