True Parallelism
Every worker executes its own connections' commands, concurrently. There is no single command thread to saturate, so throughput grows with core count.
Fine-Grained Concurrency
The keyspace is divided so that commands on different keys run at the same time and rarely wait for each other.
Predictable Latency
A native C++ engine with no garbage collector and no stop-the-world pauses. Snapshots are written on a background thread instead of a forked process, so saving does not double memory.
Life of a Command
Every request moves through the same four stages, on the core that received it.
Parse
Reads the request, including partial reads. HotKV speaks RESP2, RESP3, and inline commands, with protocol switching through HELLO.
Dispatch
Finds the command and checks its arity, flags, and key positions before execution starts.
Execute
Runs the command against its keys. Expired keys are removed when they are touched and by a background sweep.
Respond
Sends the reply in the protocol version the client negotiated.
More Than a Cache
The same parallel engine provides 300 core commands, eight eviction policies, and the subsystems below, plus more than 170 AI and enterprise commands on licensed plans. No modules to bolt on.
Snapshots + AOF
Binary snapshots that are written atomically, plus an append-only file with configurable fsync (always, everysec, no) and BGREWRITEAOF compaction.
Lua 5.4, Sandboxed
EVAL and EVALSHA run scripts atomically in a sandboxed Lua 5.4 runtime via hotkv.call(), with redis.call() kept as a compatibility alias. Named function libraries ship via FUNCTION and FCALL.
HNSW Indexes
VSIM commands provide approximate nearest-neighbor search over HNSW indexes, with cosine, L2, and inner-product distance, including tag-filtered queries.
Sketches & Rate Limits
HyperLogLog, Bloom filters, Count-Min Sketch, and Top-K heavy-hitter tracking, plus server-side GCRA rate limiting with burst control. No client-side math required.
Prometheus-Native
A built-in HTTP metrics endpoint (off until you set metrics-port) exposes per-command call counts and latency. LATENCY HISTOGRAM and INFO latencystats report p50, p99, and p99.9, and HOTKEYS ranks your most-accessed keys.
ACLs + TLS
Per-user access control scopes commands, keys, and channels, with hashed passwords. TLS terminates in-process, for client connections and for cluster and replication links. On-Premise adds roles with LDAP and OpenID Connect single sign-on.
Put every core to work
Spin up a managed instance on HotKV Cloud, or run HotKV in your own data center with an on-premise license. Same engine, same architecture, every core.