A Redis drop-in for AI: RESP3 compatibility, semantic brain
Crowkis speaks RESP3, so redis-py, ioredis, and Lettuce connect unmodified. Adoption is a port change, not a rewrite, and the semantic commands sit right beside the familiar ones.
Every new infrastructure product is tempted to invent its own API. The cost hides in the ecosystem: every language needs a client, every team needs to learn new error semantics, every debugging session starts from zero tooling. Crowkis declined that tax by speaking RESP3.
Existing Redis clients, redis-py, ioredis, Lettuce, connect to Crowkis unmodified, and the full string, hash, list, set, sorted-set, stream, and pub/sub surface works as you'd expect, 40+ commands. Adoption drops to changing a port.
The semantic commands, CSET, CGET, CSIM and family, ride the same protocol as ordinary commands, negotiated over the same HELLO handshake with constant-time auth at the door. Redis-compatibility never meant inheriting Redis's exact-match matching, the brain behind the protocol is entirely Crowkis's. For teams with different tastes, gRPC, REST, and MCP front the same engine.
Inventing a protocol is a decade of ecosystem work cosplaying as a design choice. We inherited twenty years of clients instead.