Redis Vector Search
In-memory vector search in Redis with HNSW, hybrid filtering, vector sets and semantic caching for low-latency AI apps
Overview
Redis, the in-memory data store best known as a cache, includes vector search through the Redis Query Engine. You can index vectors stored in hashes or JSON documents with HNSW or flat indexes, filter on tags, numbers, text and geo in the same query, and get results with very low latency because everything is in memory. Redis 8 also added vector sets, a simpler native data type for similarity search, and Redis offers semantic caching for LLM responses (LangCache) and agent memory patterns.
On licensing, Redis moved to source-available licenses in 2024, then added AGPLv3 as an open-source option starting with Redis 8 in May 2025. You can run it yourself or use Redis Cloud, which holds SOC 2 Type II, ISO 27001 and HIPAA certifications.
The Verdict
Who Should Use Redis for Vectors?
Best For
- Real-time recommendations and search with tight latency budgets
- Semantic caching of LLM responses
- Teams already running Redis
- Agent short-term memory and session state
Not Ideal For
- Very large vector sets where RAM cost dominates
- Cheap long-term storage of embeddings
- Teams wanting a dedicated vector DB with built-in embeddings
What's Great
- Very low query latency from in-memory storage
- Hybrid queries mixing vectors with tag, numeric and text filters
- Caching, sessions and vectors in one system
- Large community and client libraries in every language
- Strong compliance on Redis Cloud
Watch Out For
- RAM is expensive, which limits economical scale
- Licensing changes in 2024 led many to the Valkey fork
- No built-in embedding models
- Self-hosting at scale needs operational care
Pricing
View all features & details
Key Features
- HNSW and flat vector indexes
- Vectors in hashes or JSON documents
- Hybrid filters: tag, numeric, text, geo
- Vector sets data type (Redis 8)
- Vector quantization options
- Semantic caching for LLMs (LangCache)
Integrations
- LangChain, LlamaIndex
- LangGraph checkpoints and memory
- RedisVL Python library
- Clients for every major language
How It Compares
| Feature | Redis | Valkey | pgvector | Qdrant |
|---|---|---|---|---|
| Storage | In-memory | In-memory | Disk (Postgres) | Disk and memory |
| License | AGPL / RSAL / SSPL | BSD | PostgreSQL License | Apache 2.0 |
| Latency | Very low | Very low | Moderate | Low |
| Semantic caching | Built in | No | No | No |
| Cost at large scale | High (RAM) | High (RAM) | Low | Moderate |
Is this your tool? Add a badge