Cache Pot

Cache-Pot is a single-binary, Redis-compatible in-memory data store rebuilt for how AI applications actually use a cache. It speaks the Redis wire protocol out of the box, so every client and library already in your stack keeps working with zero code changes, while adding vector search, semantic response caching, and a native MCP endpoint on top.

Unlike Redis, which treats vectors and agent tooling as an afterthought bolted on through modules, Cache-Pot ships all three as first-class citizens in one self-contained binary.

Developers can drop it into an existing stack in minutes:

go install github.com/subh05sus/cache-pot/cmd/cache-pot@latestcache-pot# listening on [::]:6379, dashboard on :8080

Point any Redis client or redis-cli at it and it just works. Then reach for vector search, semantic caching, or agent memory whenever you actually need them.

Features

• Full RESP2 compatibility: strings, hashes, lists, sets, sorted sets, pub/sub, transactions, TTLs
• Vector search built in with VSET / VSEARCH / VDEL / VCARD / VDIM
• Semantic response caching via SCACHE.SET / SCACHE.GET, cache by meaning instead of exact key
• Native MCP server, wire Claude or any MCP-compatible agent in with zero adapter code
• Web dashboard for monitoring, no external services or config file needed
• Single self-contained binary: cache, vector store, semantic cache, MCP bridge, dashboard, all in one

Built using:

• Go
• RESP2 protocol
• Custom vector indexing
• MCP SDK

It's not just Redis with extra flags.

It's the cache that was actually built for how AI apps use memory.