Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching
by Redis
Use the datastore already in your stack as the vector index, the agent's memory, and the semantic cache in front of your LLM bill.
Overview
Redis for AI and search is Redis's consolidated documentation hub for using Redis as the data layer behind RAG systems and agents, rather than as a cache bolted onto one. The sections run: what it is and why, AI agents and the Context Engine, How to's, Concepts, Quickstarts, Tutorials, Ecosystem integrations, video tutorials, Benchmarks, Best practices, and Redis University courses. On the retrieval side it documents storing vector embeddings alongside structured metadata in hashes or JSON, creating secondary indexes with FLAT or HNSW vector fields, and querying with KNN, vector range queries, metadata filters and runtime filter-mode tuning, plus the newer vector sets with their `VADD`/`VSIM` commands. Client-side walkthroughs exist for redis-py, node-redis, Jedis, go-redis and NRedisStack, and RedisVL — the MIT-licensed Python client, Python 3.10+, at github.com/redis/redis-vl-python — adds schema management, semantic caching, LLM memory, semantic routing, reranking, an admin CLI and an MCP server over your indexes. On the agent side it covers a two-tier memory model (session-scoped working memory plus long-term memory retrieved by semantic search), an interactive agent builder, and the managed Context Engine suite (Redis Iris): LangCache for semantic caching, Redis Agent Memory with Python and TypeScript SDKs plus a REST API, Context Retriever for governed structured-data tools, and CDC-based Data Integration. Tutorials build LangGraph and Google ADK agents with persistent memory, agentic RAG with LlamaIndex and Bedrock, and RAG evaluation with Ragas.
At a Glance
- Topic
- RAG
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~30 minute vector quickstart; ~4-6 hours across the concept pages, notebook collection and Redis University courses
- Provider
- Redis
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Create FLAT and HNSW vector indexes over Redis hashes or JSON documents with metadata filters
- ✓Write KNN and vector range queries and tune filter mode for query execution at runtime
- ✓Give an agent two-tier memory: session-scoped working memory plus long-term semantic recall
- ✓Cut LLM spend with semantic caching that reuses responses for semantically similar prompts
- ✓Use RedisVL for schema management, semantic routing, reranking and LLM memory from Python
- ✓Wire Redis into LangGraph, LangChain, LlamaIndex and Google ADK as the persistent memory layer
- ✓Use vector sets and the VADD/VSIM commands for lighter-weight similarity search
- ✓Evaluate the resulting RAG pipeline with Ragas using the provided notebook collection
Highlights
- •Uses infrastructure most teams already run, avoiding a second datastore purely for vectors
- •Agent memory is documented as a first-class two-tier design, not improvised on top of a vector index
- •Semantic caching and RAG retrieval live in the same system, so the cache and the index share a deployment
- •Same API from local development through Redis Cloud, with clients in Python, JavaScript, Java, Go, .NET and PHP
- •Caveat worth knowing: the Context Engine services (LangCache, Agent Memory, Context Retriever) are managed Redis Cloud offerings, while vector search and RedisVL are open source and self-hostable
Who It’s For
Best For
- ✓Backend engineers already running Redis who need vector search without a new dependency
- ✓Teams building agents that must remember users across sessions with sub-millisecond reads
- ✓Engineers trying to cut LLM API spend through semantic caching
- ✓LangGraph or Google ADK users looking for a production persistence layer
Prerequisites
- •Working knowledge of Redis data types, especially hashes and JSON documents
- •Python or JavaScript proficiency for the client walkthroughs
- •Understanding of embeddings, vector similarity and basic RAG architecture
FAQ
What is Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching?
Redis's consolidated AI documentation, covering vector indexing and KNN search, two-tier agent memory, semantic caching, and the RedisVL Python client. Aimed at engineers who already run Redis and would rather not stand up a separate vector database, it documents the full path from a hash-backed HNSW index through LangGraph agents with persistent memory to production benchmarks and cost-reduction patterns.
Is Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching free?
Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching is free to access.
What level is Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching for?
Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching is aimed at a intermediate audience. Recommended background: Working knowledge of Redis data types, especially hashes and JSON documents, Python or JavaScript proficiency for the client walkthroughs, Understanding of embeddings, vector similarity and basic RAG architecture.
How long does Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching take?
Expect roughly ~30 minute vector quickstart; ~4-6 hours across the concept pages, notebook collection and Redis University courses. Most learners work through it at their own pace.
What will I learn from Redis for AI and Search — Vector Search, Agent Memory and Semantic Caching?
You'll learn: Create FLAT and HNSW vector indexes over Redis hashes or JSON documents with metadata filters; Write KNN and vector range queries and tune filter mode for query execution at runtime; Give an agent two-tier memory: session-scoped working memory plus long-term semantic recall; Cut LLM spend with semantic caching that reuses responses for semantically similar prompts; Use RedisVL for schema management, semantic routing, reranking and LLM memory from Python; Wire Redis into LangGraph, LangChain, LlamaIndex and Google ADK as the persistent memory layer; Use vector sets and the VADD/VSIM commands for lighter-weight similarity search; Evaluate the resulting RAG pipeline with Ragas using the provided notebook collection.
Topics
Sources
This page was written from 2 sources, 1 on domains other than redis.io.