KiviDB: a Redis-compatible in-memory database with built-in vector search

KiviDB is a multi-threaded alternative to Redis and Valkey written in Rust. It speaks RESP2 and RESP3, so redis-cli and existing client libraries work unchanged. It listens on port 6380 by default, so it can run alongside Redis on 6379, and it persists data to snapshots or an optional append-only file.

Search is built in. FT.CREATE and FT.SEARCH cover full-text (BM25) search, tag and numeric filters, and vector search using FLAT or HNSW indexes on hash fields. Vectors are stored as raw FLOAT32 little-endian bytes, and native snapshots can keep the HNSW graph across restarts.

The project's own benchmarks, run in one October session on AWS Graviton3 hardware, report 21.2 million pipelined GETs per second against 9.1 million for Dragonfly 2.0.2 and 2.4 million for Redis 8.10. The README says these figures are one controlled capture and should be measured on your own workload. Lua scripting and TLS are optional Cargo features, and the server is licensed under AGPL-3.0.

Developers who already use Redis for caching or agent session state can try KiviDB by changing the port, since the README says standard clients work unchanged. Anyone relying on the FT.* search commands should check them against the project's COMMANDS.md before switching. The AGPL-3.0 terms apply to modifications offered as a network service, not to applications that merely connect to KiviDB.