Skip to main content
Cache-Pot is an in-memory data store that starts where Redis left off and keeps going in the direction AI applications actually need. It speaks the Redis wire protocol (RESP2) on the network, so every Redis client library and the application code around it works without modification. On top of that familiar foundation it adds a built-in vector store, a semantic cache that deduplicates model calls by meaning, persistent agent memory, and a native Model Context Protocol endpoint — all bundled into a single binary you can download and run in seconds.

​
What makes Cache-Pot different

​
1. Redis-compatible by design

Cache-Pot communicates over RESP2, the same protocol Redis clients have used for years. Strings, hashes, lists, sets, sorted sets, pub/sub, transactions, TTLs, SCAN-family iterators — the everyday command surface you rely on is all there. Pointing an existing application at Cache-Pot is usually a one-line change: 
export REDIS_URL=redis://localhost:6379
No library swaps, no code changes, no migration scripts.

​
2. AI-native out of the box

Redis grew up serving application servers. Cache-Pot is aimed at AI agents and the applications that run them. Three capabilities that would require external modules or custom glue on Redis ship as first-class commands:
  • Vector store — insert embeddings with VSET, query by cosine nearest-neighbour with VSEARCH. Your application supplies the float values; Cache-Pot handles the index.
  • Semantic cache — store a model response once with SCACHE.SET. When a similar-enough prompt arrives later, SCACHE.GET returns the stored answer instead of making a fresh (and billable) model call.
  • Agent memory — give an AI agent a persistent scratchpad with REMEMBER and RECALL, scoped to a session namespace.

​
3. Zero dependencies

The entire stack — RESP2 server, vector index, semantic cache, MCP bridge, and a full web management dashboard — compiles into a single binary. There is nothing else to install, no configuration file required to start, and no external service to keep running alongside it.

​
How Cache-Pot compares to Redis

FeatureCache-PotRedis
RESP2 protocol✅✅
Single binary✅❌
Vector search built-in✅Needs module
Semantic cache✅❌
MCP server✅❌
Clustering / replication❌ (roadmap)✅
LicenseBSD-3-ClauseAGPL / RSAL
Cache-Pot is the right fit when you need a capable Redis-style cache on a single machine and want AI features that work without bolting on extra infrastructure. For large production clusters with replication and failover, Redis or Valkey remain the mature choice — and Cache-Pot’s roadmap includes replication as a future milestone.

​
Core capabilities

Data types

Strings, hashes, lists, sets, sorted sets — the full Redis data model with identical command syntax.

Semantic cache

Cache model responses by meaning and cut your inference bill every time a similar prompt hits the cache.

Vector store

Store and search high-dimensional vectors with cosine similarity — no separate vector database required.

Agent memory

Give AI agents a named, persistent memory namespace they can read and write across turns.
Cache-Pot does not require authentication by default. Pass --auth <password> (or set CACHEPOT_AUTH) whenever you expose Cache-Pot on a network that isn’t fully trusted. See the configuration reference for the full list of flags and their environment-variable equivalents.