documentdb-vector-search

Vector Search — Azure DocumentDB (cosmosSearch)

Azure DocumentDB's native vector index type is cosmosSearch. Pick the sub-type by scale:

Index sub-type	Scale sweet spot	Tier
vector-diskann (recommended)	Up to 500k+ vectors	M30+
vector-hnsw	Up to ~50k vectors	M30+
vector-ivf	Under ~10k vectors	M10+

Similarity options: COS (cosine), L2 (Euclidean), IP (inner product).

Rules

• vector-choose-index-type — Prefer DiskANN for production; use HNSW up to 50k, IVF under 10k.
• vector-create-diskann-index — Create a vector-diskann index with correct dimensions, similarity, maxDegree, and lBuild.
• vector-knn-query — Query with $search + cosmosSearch; tune lSearch and k; combine with pre-filters.
• vector-product-quantization — Shrink high-dimensional vectors (up to 16,000 dims) while preserving recall.
• vector-half-precision — Halve vector memory with fp16 indexing and minimal recall loss.
• vector-normalize-embeddings — Normalize embeddings when using cosine similarity; store model + dimensions alongside vectors.