REDIS-STACK

Redis Stack vector search RedisSearch module.

Definition

Redis Stack vector search features: (1) Vector storage: HASH or JSON documents with vector field type (VECTOR FLAT or VECTOR HNSW), dimensions + distance metric (L2, IP, COSINE) defined index creation. (2) Index: FT.CREATE command index defines vector field + scalar fields metadata, HNSW (default) or FLAT (brute-force). (3) Query: FT.SEARCH with KNN clause '*=>[KNN 10 @embedding $vec AS score]', returns top-K similar items + metadata. (4) Hybrid search: combine vector KNN + scalar field filters (text search BM25 + numeric ranges + tag filters) single FT.SEARCH query. (5) Range queries: '@embedding:[VECTOR_RANGE $range $vec]' returns all vectors within distance threshold. (6) In-memory performance: Redis low-latency vector search ~sub-millisecond typical for small-medium datasets <10M vectors. (7) Persistence: RDB snapshots + AOF append-only file optional. SDKs: redis-py, ioredis (Node.js), Lettuce/Jedis (Java), go-redis, etc. Redis Cloud SaaS managed Redis Stack offering AWS + Azure + GCP regions. Customers: ~10000+ Redis Stack deployments 2024 vector search use cases.

Origin

RediSearch module 2017 ; vector search added RediSearch 2.4 2022 ; Redis Stack unified distribution 2022 ; HNSW added 2023 ; Redis Inc. license change SSPL 2024 (controversy).

Example in context

Real-time recommendation system e-commerce: Redis Stack stores ~10M product embeddings HASH format, user request -> embedding generated -> FT.SEARCH KNN top-5 similar products ~sub-millisecond, integrated with existing Redis cache layer ; same Redis instance handles cache + sessions + vector search.

Related terms

• pgvector — SQL-based alternative.