Redis

If you checked the weather, swiped a credit card, or looked up a flight today, odds are Redis was somewhere in the loop. It started as one developer's 2009 side project, grew into the default caching layer for modern apps, and now claims more than 10,000 customers and a spot in 43% of enterprise AI-agent stacks.

The company was founded in 2011 by Ofer Bengal and Yiftach Shoolman, raised roughly $355M at a $2B valuation, and moved to San Francisco in 2024. The last two years have been the most eventful in its history: a bruising license fight, a rival fork, the return of its original creator, and a hard pivot toward being the memory layer for AI.

Redis keeps data in memory so applications can read and write in under a millisecond, which is why it shows up as the cache, session store, and real-time layer behind so much software. The same engine now does vector search, so teams can run RAG and semantic search without bolting on a separate database.

The business wraps that core in three products: the free open-source project, a managed cloud service, and self-managed enterprise software. Increasingly the pitch is not just speed but memory for AI agents, the thing that keeps a chatbot coherent from one turn to the next.

Redis exists because primary databases are too slow for the hot path. It absorbs the reads a relational database would choke on, powers leaderboards and sessions and streams in real time, and now serves vectors for retrieval. The newer bet is agent memory: giving AI agents fast, consistent context so they do not lose the thread mid-conversation.

The core users are backend and platform engineers who reach for Redis the moment a query needs to be instant. The newer audience is AI and ML engineers building RAG pipelines and agents, and the buyers are enterprise architects who want caching, search, and agent memory in one system instead of three. Redis says it serves customers across financial services, gaming, healthcare, and e-commerce.

The lineup fans out from one idea. Redis Open Source is the free core; Redis Cloud is the managed multi-cloud service; Redis Software is the self-managed enterprise build for teams that need to run their own infrastructure.

The growth story is Redis for AI, a cluster of newer products, Iris, LangCache, Agent Memory, and the Context Engine, aimed squarely at the agent boom. Redis Flex rounds it out by blending RAM and flash so big datasets run fast without paying pure in-memory prices.

Redis runs the classic open-core play. The database is free to self-host, and the money comes from taking the operations off your hands: Redis Cloud is metered by memory and throughput, and Redis Software is licensed by shard count for enterprises that self-manage. The value you pay for is failover, backups, security, high availability, and support, not the code itself.

Redis competes with Memcached and managed caches like Amazon ElastiCache, faster upstarts like DragonflyDB, and a wave of dedicated vector databases on the AI side. Its sharpest rival is one it accidentally spawned: Valkey, the BSD fork that AWS, Google, and Oracle stood up after the 2024 license change.

What Redis still has going for it is ubiquity and speed, one platform where caching, vectors, and agent memory all live, and a roadmap the returned creator has made credible again. Whether that is enough to slow Valkey is the open question.

Redis puts three figures front and center: more than 10,000 customers, a claim that 43% of enterprise AI-agent stacks touch Redis somewhere, and vector search that scales to a billion vectors on Redis 8. Its clearest customer proof point is Relevance AI, which says Redis cut vector search times by 99.5%, from two seconds to ten milliseconds, after testing Qdrant, OpenSearch, and its own build.

Developers rarely argue with the speed, and Redis is so common that adopting its newer features barely costs anything to try. The complaints are consistent too: enterprise and managed pricing feels steep for small teams, and clustering and security config have a learning curve.

The deepest wound is self-inflicted. The 2024 license change sent a chunk of the community to Valkey, and the goodwill did not come back overnight. The 2025 return to open source was, in part, an apology written in a license file.

Redis has raised about $355M and carries a $2B valuation. The headline round was a $100M Series F in 2020, co-led by Bain Capital Ventures and TCV, with Francisco Partners, Goldman Sachs Growth, Viola Ventures, and Dell Technologies Capital along for the ride. Several current job descriptions describe a company on its way to IPO, so the next chapter may be a public one.

Redis has a genuine second act in front of it. Being the fast context and memory layer for AI agents plays directly to what it is already good at, and the numbers it cites suggest real traction. The catch is Valkey, which took a slice of the open-source base and the trust that came with it. The bet for the next few years, possibly as a public company, is that speed plus a healed community plus an AI story is a stronger hand than a free fork alone.

Redis calls itself engineering-led, and the values match: Be Real, Be Unreasonable, and the two that read like a mission statement, simplicity is everything and speed is our superpower. The core database is still hand-tuned C, C++, and Rust, while the newer AI services run on Go, Java, and Python.

The cultural plot twist is Salvatore Sanfilippo, the creator, who left in 2020 and came back in late 2024 to help steer the codebase and mend fences with the open-source world. A lot of the deep systems work now runs out of a Sofia, Bulgaria hub, alongside teams in the US, UK, and beyond.

US roles disclose base bands running roughly $91K to $310K depending on function and seniority, with a smaller set of EUR-denominated engineering roles out of Bulgaria. Base is only part of it: most roles add equity grants, US postings note 401(k), unlimited time off, and discretionary bonuses or stock options, and sales roles carry a commission plan on top.

The commercial entity is Redis Ltd, Israeli-founded and now headquartered in San Francisco. The story that actually matters here is licensing: Redis moved off the permissive BSD license to a source-available model in 2024, drew a fork and a fair amount of anger, then re-added the OSI-approved AGPLv3 as an option with Redis 8 in 2025. If you build on Redis, which license version you are on is a decision, not a footnote.

The last two years of Redis headlines split into two threads. One is the license drama: the 2024 shift to source-available, the Valkey fork backed by AWS, Google, and Oracle, and the 2025 return to open source. The other is the AI push: the creator's return producing a native vector data type, billion-vector search on Redis 8, and the Iris platform for agent context. The links below trace both.