# Retrieval Redis Vector Search and HNSW Indexes
Status: public
Confidence: medium (0.685) (verified)
Last verified: 2026-06-03
Generation: ai_structured


## TL;DR

Redis vector indexes let agents inspect retrieval schema, vector similarity settings, and HNSW tradeoffs inside Redis-backed RAG systems.

## Core Explanation

When Redis is used as a retrieval backend, agents need more than the key name. They should inspect the index schema, vector field definition, dimensions, distance metric, index type, query filter, and whether the search path uses approximate or exact behavior.

HNSW settings affect the latency and recall profile of vector search. A source-mapped diagnosis should connect user-visible retrieval failures to concrete evidence: index definition, vector shape, filter predicate, memory pressure, and measured results for representative queries.

## Source-Mapped Facts

- Redis documentation describes vector search as finding vectors that are similar to a query vector in the vector space. ([source](https://redis.io/docs/latest/develop/ai/search-and-query/vectors/))
- Redis documentation lists FLAT and HNSW as vector index types. ([source](https://redis.io/docs/latest/develop/ai/search-and-query/vectors/))
- Redis documentation says schema fields in an index can include vector fields. ([source](https://redis.io/docs/latest/develop/ai/search-and-query/indexing/))

## Further Reading

- [Redis Vector Search](https://redis.io/docs/latest/develop/ai/search-and-query/vectors/)
- [Redis Indexing](https://redis.io/docs/latest/develop/ai/search-and-query/indexing/)