Redis
Redis is an in-memory data store used as a vector database, semantic cache and memory layer for AI and agent applications.
Overview
Redis
Redis is an open-source, in-memory data store used as a database, cache and message broker, known for sub-millisecond reads and writes. Created in 2009, Redis is now widely used as a vector database, semantic cache and memory layer for AI applications. Redis stores embeddings for retrieval-augmented generation, caches large language model responses to cut inference cost and latency, and holds short- and long-term memory for agents. It is general-purpose infrastructure, not an AI product, but itsvector sets, RedisVL library and agent memory server have made it common plumbing in production GenAI stacks.
Production credibility: The open-source Redis project was created by Salvatore Sanfilippo (antirez) in 2009; the commercial company Redis (formerly Redis Labs) was founded in 2011 by Ofer Bengal and Yiftach Shoolman, headquartered in Mountain View, CA. Raised approximately $350M+, including a $110M Series G (2021, led by Tiger Global) at a reported $2B+ valuation. Reached roughly $300M ARR with about 12,000 paying customers and 50+ accounts spending over $1M/year; named customers include OpenAI, Uber and Lovable. antirez rejoined in 2024; Redis 8 (2025) added the AGPLv3 open-source license and native vector sets.
Key Features
- Vector sets and vector search for similarity over embeddings, built into Redis 8
- RedisVL client library for building RAG and GenAI features in Python
- Semantic caching (LangCache) that reuses prior LLM answers to cut cost and latency
- Agent Memory Server for short- and long-term agent memory with topic and entity extraction
- Sub-millisecond in-memory reads and writes for low-latency retrieval
- Core data structures (strings, hashes, sorted sets, streams) plus pub/sub messaging
- Redis Cloud managed service across AWS, Azure and Google Cloud
- Integrations with LangChain, LangGraph and LlamaIndex for agent and RAG pipelines
Ideal Use Case
AI teams use Redis as a vector database for retrieval, a semantic cache that reuses prior LLM answers to cut cost and latency, and a memory layer that stores short- and long-term context for agents.
How Redis differentiates
As a vector store, Redis competes with Pinecone and pgvector. Pinecone is a managed, vector-native service that removes operations but locks you into one workload; Redis adds vector search to a store many teams already run for caching and queues, so it can serve embeddings, a semantic cache and agent memory from one system. Against pgvector, Redis trades SQL and disk durability for in-memory speed, which suits low-latency retrieval and caching. The trade-off is that Redis is memory-bound and less suited to very large vector corpora than a purpose-built vector database, and durability needs careful configuration.
FAQ
Q: What is Redis used for? A: Redis is an in-memory data store used as a database, cache and message broker. In AI apps it is also used as a vector database, a semantic cache for LLM responses, and a memory layer for agents.
Q: How is Redis used as a vector database for AI? A: Redis stores embeddings using vector sets and the RedisVL library, then runs similarity search for retrieval-augmented generation. It can also semantically cache LLM responses to cut cost and latency and hold short- and long-term agent memory, often from the same instance used for caching.
Q: Redis vs Pinecone for vector search: which is better? A: Pinecone is a managed, vector-native database that removes operations and scales to very large corpora. Redis adds vector search to a store many teams already run, so it serves embeddings, caching and agent memory together with in-memory speed. Pinecone suits huge dedicated vector workloads; Redis suits low-latency retrieval consolidated with existing infrastructure.
Q: Who founded Redis? A: The open-source Redis project was created by Salvatore Sanfilippo (antirez) in 2009. The commercial company, formerly Redis Labs, was founded in 2011 by Ofer Bengal and Yiftach Shoolman. antirez rejoined the company in 2024.
Q: How much has Redis raised, and is it open source? A: Redis has raised roughly $350M+, including a $110M Series G in 2021 at a reported $2B+ valuation. After a 2024 license change, Redis 8 (2025) re-added the OSI-approved AGPLv3 open-source license alongside its source-available options.
tl;dr
Redis is a fast in-memory data store that has become common infrastructure for AI apps, used as a vector database, semantic cache and agent memory layer. It is general-purpose plumbing rather than an AI product. Redis 8 added native vector sets and the AGPLv3 open-source license, and the company reportedly reached about $300M ARR.
Why Use Redis
Editorial Review
Our take on Redis.
Redis serves as an in-memory vector database and semantic cache for AI agents, trading persistence for speed.
What works
- Sub-millisecond vector and semantic lookups for AI applications
- Familiar API surface if you already use Redis for caching
- Freemium model lowers barrier to experimentation
What doesn't
- In-memory architecture limits total data size to available RAM
- Requires architectural planning when scaling beyond single node
Redis is an in-memory data store that's become a practical infrastructure layer for AI applications needing fast retrieval and caching. It handles vector storage, semantic caching, and memory management—three jobs that matter when you're building agents or retrieval-augmented generation (RAG) systems where latency kills user experience. The appeal is straightforward: keeping your working set in RAM means sub-millisecond lookups, which beats disk-bound alternatives when response time matters.
The tradeoff is real, though. In-memory means you're managing RAM as a constraint. If your vector corpus or cache gets larger than your available memory, you hit a ceiling. Redis works best as a layer in a larger stack—typically fronting a persistent vector store or traditional database, not replacing one. For prototyping or for applications where your hot data genuinely fits in memory, it's clean. For systems that need unlimited scale without architecture gymnastics, you might find yourself building around its boundaries.
The freemium structure and high community rating suggest it's well-adopted for this use case. If you're familiar with Redis from cache duty, the vector and semantic features feel like a natural extension rather than a new tool entirely.
User Reviews
Similar Tools
