Dragonfly

The API is the full Redis command set — Dragonfly implements the Redis wire protocol and supports the Redis command surface relevant to most production use cases. Teams can use any Redis client library (redis-py, ioredis, Jedis) to connect to Dragonfly without modification. The operational API follows Redis conventions for monitoring, replication, and persistence configuration.

Reliability for self-hosted Dragonfly is the team's operational responsibility. Dragonfly's multi-threaded design provides better single-instance reliability under high concurrency than single-threaded Redis. Dragonfly Cloud provides the managed option with vendor-handled availability. Teams considering Dragonfly for production caching should evaluate the maturity of specific Redis features they depend on, as Dragonfly's compatibility surface, while broad, may have gaps in advanced Redis features.

Dragonfly is a Redis-compatible in-memory data store engineered for significantly higher throughput and memory efficiency than Redis, targeting the same use cases (caching, session storage, rate limiting, pub/sub) with a drop-in replacement API. For agents using Redis client libraries, Dragonfly requires no code changes — the same commands work against a Dragonfly server. The performance improvement is most significant for high-concurrency workloads where Redis's single-threaded architecture becomes a bottleneck.

Authentication uses Redis AUTH command for password-based authentication. Dragonfly supports the same ACL user management as Redis 6+ for fine-grained command and key access control. Teams migrating from Redis should audit their ACL configuration compatibility with Dragonfly's implementation.

Documentation covers the installation, configuration, and Redis compatibility notes. The Redis compatibility documentation is the most important reference for teams migrating — it identifies commands and behaviors that differ from Redis. Teams evaluating Dragonfly versus KeyDB or Valkey (Redis fork) should compare the performance benchmarks for their specific workload patterns.