All Tools
D
Vector DBFreeOpen Source
DISKANN
Graph-based vector search that runs on disk, not RAM
MIT
ABOUT
Traditional vector search indexes must fit in RAM to serve queries at low latency, which limits dataset size to available memory. DiskANN uses a compressed Vamana graph structure that lives primarily on disk, keeping only a compact representation in memory. This lets it index billions of vectors on a single machine — datasets that would otherwise require a cluster of RAM-heavy servers. It supports real-time insertions and deletions (no full rebuild), filtered search with arbitrary predicates, and streaming index building. The library is used in production at Microsoft for Bing, Azure Search, and Cosmos DB.
INTEGRATION GUIDE
1. Index billions of image or document embeddings on a single server without paying for RAM-heavy hardware
2. Serve filtered vector search queries at low latency — e.g., "find similar products in category X under $50"
3. Build a real-time recommendation engine that updates the index as new items arrive, without downtime
4. Deploy vector search on edge devices or in memory-constrained environments using the disk-resident index
5. Replace an in-memory ANN index that can't scale to the next order-of-magnitude dataset size
TAGS
vector-searchapproximate-nearest-neighborgraph-baseddisk-basedmicrosoftsimilarity-searchcpp