#1★ TOP PICK
Valkey
The Linux Foundation fork — same Redis, governance that cannot be revoked.
96
OPEN SOURCEBSD-3-ClauseSELF-HOSTLOCAL-FIRST
Valkey is the community fork of Redis taken from the last BSD-licensed release, now under Linux Foundation stewardship with AWS, Google and Oracle contributing. It is the same codebase, so the protocol, commands and clients are identical and migration is a hostname change. BSD-3-Clause, around 27k stars. The argument for it is not technical — it is that no single company can relicense it, which is precisely the event that created it. Performance has since moved ahead of Redis in several published benchmarks through multi-threaded I/O work the fork prioritised.
Strengths
- +Drop-in for Redis — same protocol, same clients, hostname change
- +BSD-3-Clause under Linux Foundation governance; no vendor can relicense it
- +Backed by AWS, Google and Oracle with real engineering behind it
- +Faster than Redis on several published benchmarks after multi-threading work
Trade-offs
- −Does not include Redis Ltd's proprietary modules — no RedisJSON or RediSearch
- −Younger brand, so some managed providers still list only Redis
- −Documentation still catching up with Redis's two decades
- −Divergence from upstream Redis will grow over time
Free and open source. Managed offerings from AWS, Google and others. #2
Memcached
The one that only caches — and does it without ceremony.
95
OPEN SOURCEBSD-3-ClauseSELF-HOSTLOCAL-FIRST
Memcached predates Redis and does one thing: a distributed in-memory key-value cache. No persistence, no data structures, no pub/sub, no clustering logic. BSD-3-Clause, and stable for so long that it is genuinely boring, which for infrastructure is a compliment. Listed because a great many Redis deployments are caching a string against a key and nothing more — and for that job Memcached uses less memory, has fewer failure modes, and gives you nothing to misconfigure. Ask what you are actually storing before you assume you need Redis.
Strengths
- +BSD-3-Clause and completely uncontroversial for two decades
- +Lower memory overhead than Redis for simple key-value caching
- +Very few moving parts, so very few ways to break it
- +Supported by every major managed cloud
Trade-offs
- −Caching only — no persistence, no data structures, no pub/sub
- −No replication or failover built in
- −Cannot serve as a queue, session store or message bus
- −Wrong tool the moment you need more than a cache
#3
Garnet
Microsoft Research's cache — MIT, and unusually fast under contention.
92
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST
Garnet is a Redis-protocol cache store out of Microsoft Research, written in C#, and its published throughput under high client concurrency is striking — the workload where Redis's single-threaded design shows most. MIT licensed, around 12k stars, used inside Microsoft before release. It is the most permissively licensed serious option here. The caution is ecosystem: a .NET server is an unusual dependency for teams that are not already there, and it is the newest of these by some distance.
Strengths
- +MIT — the most permissive licence in this comparison
- +Exceptional throughput under high concurrency
- +Microsoft Research behind it, with real production use
- +Redis protocol compatible, so clients work unchanged
Trade-offs
- −Newest option here with the shortest track record
- −A .NET runtime dependency many teams will not want
- −Smaller community than Valkey or DragonflyDB
- −Not all Redis commands and data types are implemented
#4
KeyDB
The earlier fork — multithreaded Redis, BSD, now Snap-owned.
89
OPEN SOURCEBSD-3-ClauseSELF-HOSTLOCAL-FIRST
KeyDB forked Redis before any of this to add multithreading, and was acquired by Snap in 2022. It remains BSD-3-Clause with around 12.5k stars and full Redis protocol compatibility, including active replication where multiple nodes accept writes. It is the option with the longest track record of the forks here. The honest caveat is momentum: development has slowed considerably since the acquisition, and Valkey has absorbed most of the community energy that KeyDB once held.
Strengths
- +BSD-3-Clause with a longer production history than Valkey
- +Multithreaded well before Redis addressed it
- +Active-active replication — multiple writable nodes
- +Full Redis protocol compatibility
Trade-offs
- −Development pace has slowed markedly since Snap acquired it
- −Community energy has largely moved to Valkey
- −Corporate owner rather than foundation governance
- −Fewer managed hosting options
#5
DragonflyDB
Redis-compatible, built for modern multi-core machines.
70
SOURCE-AVAILABLEBusiness Source License 1.1 (source-available, converts to Apache-2.0 in time)SELF-HOSTLOCAL-FIRST
DragonflyDB is a ground-up reimplementation of the Redis and Memcached protocols designed for the hardware people actually run now: it scales across all cores rather than being effectively single-threaded, and reports substantially higher throughput per machine with lower memory use. Around 31k stars. That means fewer, smaller instances for the same load, which is a real cost line rather than a benchmark. The licence is Business Source License 1.1 — source-available, converting to Apache-2.0 after a delay — so it is not open source today and should not be described as such.
Strengths
- +Uses every core — far higher throughput per machine than Redis
- +Lower memory footprint for the same dataset
- +Speaks both Redis and Memcached protocols
- +Fewer, smaller instances for the same workload
Trade-offs
- −Business Source License — not open source today, whatever lists say
- −Cannot be offered as a competing managed service
- −Younger codebase with a shorter production record
- −Some rarely used Redis commands are unimplemented
Free to self-host within BSL terms. Dragonfly Cloud is paid.