Launching today
HotKV
Redis-compatible store that uses every core. AI built in.
0 followers
Redis-compatible store that uses every core. AI built in.
0 followers
A multi-threaded, Redis-compatible in-memory data store written in C++20. Existing clients connect unchanged. Vector search, a semantic cache for LLM response and agent memory run inside the server. Benchmarks with Redis, Valkey and Dragonfly are public.
Hello Product Hunt.
HotKV is a Redis-compatible in-memory data store written from scratch in C++20. It runs commands on every core of the machine at once instead of one. Same protocol, same clients, nothing to rewrite.
What is different
Vector search (HNSW), a semantic cache for LLM responses and agent memory live inside the server, not as modules or a second service. In a benchmark on four Claude models, HotKV in front of the API cut calls, tokens and cost by about 70% on 3,000 support requests. The cases where cached answers were worse were measured and published as well.
Performance
On a 160 vCPU Azure instance HotKV had the highest throughput on 11 of 12 commands against Redis 8.10.2, Valkey 9.1.2 and Dragonfly 2.0.2. The harness, the raw data and the one command it did not win are all on GitHub.
Try it
HotKV Cloud is in free early access, no card required. HotKV is bootstrapped, so instances are provisioned in batches. Write "Product Hunt" in the use case field to be in the first batch.
Questions about the engine, the benchmarks or the trade-offs are welcome below. Every one gets an answer.