How the Valkey fork and Redis 8 differ in license, features and managed pricing in 2026, whether your client code cares, and how to switch an ElastiCache cluster.

In March 2024 Redis moved from the BSD license to a dual RSALv2 and SSPLv1 license, and within weeks the Linux Foundation forked the last BSD release, Redis 7.2.4, as Valkey, with AWS, Google Cloud, Oracle and others behind it. A year later Redis 8 added AGPLv3 as a third option. Two codebases now share a protocol and a history and are drifting apart in features and licensing.
For most applications the question is practical: which engine does your managed provider run, what does it cost, and does any client code change. This guide compares the two on licensing, features, compatibility and managed pricing, and shows how a switch works on ElastiCache.
Valkey's BSD license lets anyone run, modify and offer it as a service without obligations beyond attribution. Redis 8's AGPLv3 option is an OSI-approved open source license, but its network clause requires you to publish modifications to Redis if you offer a modified Redis to users over a network, and many companies restrict AGPL software by policy. RSALv2 and SSPLv1 are source-available licenses that forbid offering Redis as a managed service.
For teams that use a managed cache and never modify the server, none of this changes daily work. It matters for platform teams that self-host and patch the server, for vendors embedding it, and for legal reviews that ban AGPL.
Valkey's releases have focused on performance and operations. Valkey 8 added multithreaded I/O, and Valkey 9.0 added hash field expiration, atomic slot migration for resharding, and multiple logical databases in cluster mode. JSON, Bloom filters and vector search are available as separate Valkey modules.
Redis 8 put the former Redis Stack modules in the core distribution: JSON, the query engine with full-text and vector search, time series and probabilistic structures, plus a new vector set type. Redis compares the two on its own comparison page, which is worth reading with the author in mind.
If your application uses strings, hashes, lists, sets, sorted sets, streams and pub/sub, both engines cover it. If it uses FT.SEARCH or JSON.SET, check the module versions your provider runs on Valkey before switching.
Valkey answers the same protocol and core commands as Redis 7.2, so redis-py, go-redis, ioredis, Jedis and Lettuce connect without changes. The Valkey GLIDE client, maintained by the Valkey project, adds cluster-aware features but is optional. Commands added to Redis after the fork, and Redis 8 module commands, are the gaps to check.
// The same go-redis code runs against Redis 7.2 and Valkey.
rdb := redis.NewClient(&redis.Options{Addr: os.Getenv("CACHE_ADDR")})
if err := rdb.Set(ctx, "session:"+id, userID, 30*time.Minute).Err(); err != nil {
return err
}
AWS priced ElastiCache for Valkey 20 percent below Redis OSS for node-based clusters and 33 percent below for serverless, and supports in-place, zero-downtime upgrades from Redis OSS to Valkey with existing reservations still applying. Google Cloud offers Memorystore for Valkey alongside Memorystore for Redis, and announced Valkey 9.0 as generally available in March 2026. Redis Ltd. runs Redis Cloud with the Redis 8 feature set on all three major clouds.
The upgrade is a modify operation on the replication group. Test it on a staging cluster first, and check the parameter group family for the target version.
Best fit: Valkey for managed caches on AWS or GCP where the lower price applies and you use core data types; Redis 8 when you depend on its bundled search, JSON or vector features.
Encore brings infrastructure from code, and its cache primitive is a cluster plus typed keyspaces declared in Go. On deploy, the infrastructure docs list ElastiCache for Redis on AWS and Memorystore for Redis on GCP, in your own account, and encore run provides an in-memory Redis locally. Your code calls keyspace methods rather than a Redis client, so the application has no import, connection string or command tied to an engine or its license.
package ratelimit
import (
"context"
"time"
"encore.dev/beta/auth"
"encore.dev/beta/errs"
"encore.dev/storage/cache"
)
var cluster = cache.NewCluster("ratelimit", cache.ClusterConfig{
EvictionPolicy: cache.AllKeysLRU,
})
// Requests per user per minute; the key pattern and expiry live with the code.
var RequestsPerUser = cache.NewIntKeyspace[auth.UID](cluster, cache.KeyspaceConfig{
KeyPattern: "requests/:key",
DefaultExpiry: cache.ExpireIn(time.Minute),
})
func check(ctx context.Context, uid auth.UID) error {
n, err := RequestsPerUser.Increment(ctx, uid, 1)
if err != nil {
return err
}
if n > 100 {
return &errs.Error{Code: errs.ResourceExhausted, Message: "rate limit exceeded"}
}
return nil
}
Encore's Heroku migration guide notes that the cache primitive does not expose the full Redis or Valkey API, so applications that need search modules or arbitrary commands keep using an external Redis or Valkey store. For rate limits, sessions and lookups, the typed keyspaces cover it, and Encore provisions an IAM role for the service, or a service account on GCP.
An agent adding a cache gets a typed key pattern and expiry checked at compile time, and runs it against the local in-memory Redis before the change ships.
Best fit: teams that want caching declared in application code, provisioned in their own AWS or GCP account, without binding the code to a Redis client or engine.
Pick Valkey for managed caches when your provider offers it, for the lower ElastiCache price and the BSD license, as long as you use core data types. Pick Redis 8 when you rely on its bundled search, JSON or vector sets, or Redis Cloud's features. If you build on Encore, keep app code on typed keyspaces, and treat the engine as an infrastructure detail of the environment.
Valkey is a fork of Redis OSS 7.2.4 created in 2024 under the Linux Foundation and licensed BSD-3-Clause. Redis 8 is licensed under AGPLv3 with RSALv2 and SSPLv1 as alternatives, and bundles JSON, search, time series and probabilistic data types in the core. Both speak the same protocol for core commands.
Yes. For applications using core Redis commands up to version 7.2, the protocol and commands are the same, and common clients such as redis-py, go-redis and ioredis work against Valkey. Features added to Redis after 7.2, such as the bundled query engine, have different or no Valkey equivalents.
Yes. On ElastiCache, Valkey is priced 20 percent lower than Redis OSS for node-based clusters and 33 percent lower for serverless caches, and existing Redis OSS clusters can be upgraded to Valkey in place without downtime.
Encore's docs list ElastiCache for Redis on AWS and Memorystore for Redis on GCP for its cache primitive. Application code uses Encore's typed keyspaces rather than a Redis client, so it has no dependency on a particular engine or license.
Deploy to your cloud on AWS and Google Cloud
without managing infrastructure
Encore lets developers and agents define application infrastructure directly in code, then automatically provisions it from local development to production in your AWS or GCP account.