microsoft/DiskANN

Health v2 · maintenance only

Adoption not part of the score

1949 stars · 465 forks observed · 2026-10-20

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

DiskANN3 is a composable Rust library for scalable, accurate vector indexing and nearest-neighbor search, designed to be embedded into databases via a DataProvider trait. It supports real-time updates, filtered and diversity-aware search, multiple quantizers, and memory tiers from in-memory to disk-based providers.

Use cases

• add vector similarity search to my database
• billion-scale nearest neighbor search on a single node
• fast filtered vector search with real-time updates
• embed an ANN index in a key-value store
• benchmark vector search throughput and recall
• larger-than-memory vector index on disk

When to choose

• you need scalable vector search embedded in your own storage engine or database
• you need fresh real-time updates without index rebuilds
• you need disk-based indexes for datasets larger than memory
• you want high throughput and recall compared to HNSWlib or managed vector DBs

When to avoid

• you just want a turnkey managed vector database with an API out of the box
• your project is not in Rust or C++ and you cannot implement the DataProvider trait
• you need non-x86/aarch64 SIMD support

Facets

library · maturity active

vector-database search-engine database databases machine-learning developer-tools rust windows cpp ann nearest-neighbor-search vector-index diskann hnsw-alternative quantization similarity-search search linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="microsoft/DiskANN")