BGram

BGram: Open-Source Database and Search Engine for Large-Scale Embeddings

A scalable database optimized for storing and querying dense vectors, enabling fast nearest neighbor search for big data applications.

What is BGram?

BGram is an open-source database and search engine optimized for storing and querying billions of embeddings/vectors. It is designed to efficiently store dense vectors and enable fast nearest neighbor search on those vectors.

Some key features of BGram:

• Stores embeddings/vectors in a compressed, memory-efficient format
• Supports indexing billions of embeddings for fast nearest neighbor search
• Has a simple client API for data loading, vector search, and aggregation
• Horizontally scalable with sharding support
• Embeddings can be uploaded from machine learning frameworks like TensorFlow and PyTorch

BGram can be useful for applications that need to store and query large embedding datasets, such as image search, document search, recommendation systems, and semantic search. It allows running complex vector similarity queries over billions of embeddings interactively.

Compared to alternatives like FAISS and SPTAG, BGram focuses specifically on fast nearest neighbors over dense embeddings and has a simpler API. It offers competitive performance and scalability for embedding search workloads. Overall, BGram is an efficient open-source vector database suited for big data applications.

BGram Features

Official Links