The in-memory data store that answers the oldest question in software - "why is this so slow?" - in under a millisecond.
Open an app. Tap a feed, load a cart, check a score, ask a chatbot something. Somewhere in that half-second, a request flickers through a small red box of memory and back out before your screen finishes drawing. You will never see it. That is rather the point. Redis is the database that works hardest when no one is watching - the stagehand of the modern internet, holding the props so the show never pauses.
Most databases are built to remember. Redis is built to remember fast. It keeps data in RAM instead of on disk, which is a bit like keeping your keys in your pocket rather than in a vault across town. The result is sub-millisecond response times that turned a clever shortcut into critical infrastructure for companies you use every day.
It did not start as a company, or even as a plan. It started as one developer's irritation.
In 2009, an Italian developer named Salvatore Sanfilippo - known online as antirez - was building a real-time web analytics tool called LLOOGG. The system needed to record visitor events and read back recent activity, constantly and quickly. MySQL, the obvious choice, buckled under the pace.
So he did the thing that founds legends and breaks weekends: he wrote his own. A small server that held data in memory and spoke in simple commands. He called it Redis - REmote DIctionary Server - and open-sourced it. Developers took one look at the speed and never gave it back.
Two years later, in 2011, Ofer Bengal and Yiftach Shoolman founded a company - first Garantia Data, then Redis Labs, eventually just Redis - to build a business around the project. In 2015 they hired antirez himself to steward the open-source core.
The arc since reads less like a roadmap and more like a novel: a beloved creator, a runaway open-source hit, a billion-dollar company, a public fight over licensing, and a homecoming. We will get to the fight.
Redis is often filed under "cache," which is a little like calling a Swiss Army knife "a bottle opener." It is true, but it undersells the rest of the tools.
Sit in front of a slow database and hand back hot data instantly. The classic job, still the most common one.
Hold who-is-logged-in for millions of users without hammering your primary store on every click.
Leaderboards, live counters, location tracking, feeds - anything that has to update the instant it changes.
Pub/sub and streams to move events between services, queue work, and fan out notifications.
Store documents, run full-text and secondary-index queries - a query engine baked into the core.
Vector similarity search, semantic caching and LLM memory - the recall layer behind AI agents and RAG.
Approximate description of common use cases - exact capabilities depend on version and configuration.
The core is open source and costs nothing. The business is the open-core model: managed cloud (Redis Cloud) and self-managed enterprise software for the companies that want it run for them, or run at scale. Investors clearly liked the math.
For most of its life Redis was BSD-licensed - about as permissive as open source gets. Then came 2024, and the most-watched corporate drama in databases that year.
antirez writes Redis to speed up his analytics tool and open-sources it.
Bengal and Shoolman found the business that becomes Redis Labs, then Redis.
antirez comes on board as lead open-source maintainer.
Redis drops BSD for a dual RSALv2 / SSPL model. Cloud providers can no longer use the source freely. The community erupts.
The codebase is forked into Valkey under the Linux Foundation, backed by AWS, Google and Oracle.
The creator rejoins Redis and pushes for a path back to open source.
Redis returns to an OSI-approved open-source license, folds JSON, search, time series and vector sets into the core.
Redis shows up wherever a delay would be noticed - real-time feeds, location updates, session stores, and the recall layer behind AI features. Production users have included internet-scale names alongside a large share of the Fortune 500.
Representative production users and sectors per public reports; not an exhaustive or endorsed customer list.
Open the app again. The feed loads, the cart remembers, the chatbot recalls what you said three messages ago. None of it pauses to think, because something already did the thinking and stashed the answer in memory. That is the same red box, doing the same quiet job it did for a frustrated developer in 2009 - except now it does it for millions of apps, and increasingly for the AI sitting between you and your questions.
Redis spent its fifteenth year arguing with itself in public about licenses and then coming home to open source. The drama made headlines. The database, meanwhile, kept answering in under a millisecond - which was always the only thing it promised to do. The internet got faster, and most people never noticed who to thank.