needlestack 0.0.7

A distributed vector search microservice.

Needlestack is a distributed vector search microservice.

• Free software: MIT license

• Documentation: https://needlestack.readthedocs.io.

Features

• gRPC server for kNN vector search

• Shard vectors over multiple nodes

• Replicate shard over multiple nodes

• Retrieve vectors by ID

Caveats

These are some limitations of the current alpha builds

• Vectors must be sharded and indexed before loading to Needlestack

• To reload updated vectors, all MergerServicer and SearcherServicer nodes must be restarted

• Vectors must be saved locally to disk for Needlestack to load them

• Only kNN library supported currently is Faiss

Next Steps

These are valuable features to make Needlestack usable in production

• Endpoint to index vectors in realtime

• Update vectors from pre-built data sources without restarting the cluster

• Allow vectors to be loaded from data sources like S3 or GCS

• Add support for other kNN libraries