RAMen
RAMen is a fast in-memory data store like Redis, but built for AI: drop-in Redis protocol, native vector search, semantic caching, and a built-in MCP server for…
RAMen is a fast in-memory data store, like Redis, but built for AI apps and AI agents.
It does three things:
- It works as a drop-in cache. It speaks the same protocol as Redis, so your existing Redis client and code keep working with no changes.
- It can store and search vectors, and it can cache AI answers by meaning. So if two prompts mean the same thing, RAMen can return the saved answer instead of calling the model again. That saves you money.
- It has a built-in MCP server. That means AI agents (like Claude) can read, write, search, and remember data in RAMen directly as a tool. No extra glue code needed.
Redis was built for app servers. RAMen is built for AI agents.
Is this a Redis alternative?
Yes, for a lot of common use cases. If you use Redis (or Valkey) as a cache or a simple key value store, you can point your app at RAMen instead and it will just work, because RAMen speaks the Redis protocol (RESP2).
Where RAMen is different:
- It has vector search and a semantic cache built in. With Redis you would need an extra module or extra code for that.
- It has an MCP server built in, so AI agents can use it as a tool out of the box.
- It is one small single file (a single binary) with no extra dependencies. Easy to download and run.
What RAMen does NOT do yet (so you know what you are getting):
- No clustering, no replication, no failover.
- No append-only-file durability (it saves snapshots to disk instead).
- Not tuned to beat Redis or Valkey on raw speed.
So: great as a Redis-style cache and an AI data layer for one machine. Not yet a replacement for a big production Redis cluster.
Quick start (the simple version)
You need Go installed (version 1.25 or newer). Then run one of these.
Option 1: install with Go
go install github.com/Rohit-Dnath/RAMen/cmd/ramen@latest
ramen
Option 2: run with Docker
docker run -p 6379:6379 -p 8080:8080 ghcr.io/rohit-dnath/ramen:latest
Option 3: build from source
git clone https://github.com/Rohit-Dnath/RAMen
cd RAMen
go run ./cmd/ramen
When it starts you will see:
ramen: listening on [::]:6379
ramen: dashboard on http://localhost:8080
That is it. RAMen is now running on port 6379, and a live dashboard is at http://localhost:8080.
How to use it
Talk to it like Redis
If you have redis-cli installed, just connect to port 6379:
redis-cli -p 6379
> SET hello world
OK
> GET hello
"world"
> EXPIRE hello 60
(integer) 1
If you have an app that already uses Redis, point it at RAMen. Usually you only change one setting:
export REDIS_URL=redis://localhost:6379
Basic commands
These work just like Redis:
SET user:1 "Rohit" store a value
GET user:1 read it back
INCR visits count something
HSET person name Rohit store fields under one key
RPUSH queue job1 job2 a list
SADD tags go ai a set
ZADD board 100 alice a ranked list
SUBSCRIBE news listen for messages
PUBLISH news "hello" send a message
Full list with examples: docs/commands.md.
Vector search
Store vectors and find the closest ones. Your app gives RAMen the numbers, no API key needed.
VSET docs d1 0.1 0.2 0.9 META "intro page"
VSET docs d2 0.9 0.1 0.0 META "pricing page"
VSEARCH docs 0.1 0.2 0.85 TOPK 1 WITHSCORES
Semantic cache (save money on AI calls)
Cache an AI answer once. Next time a similar question comes in, get the saved answer back instead of paying for another model call.
SCACHE.SET "What is the capital of France?" "Paris"
SCACHE.GET "whats the capital of france" THRESHOLD 0.9
> "Paris"
This needs an embeddings provider. It works with a free local Ollama or with OpenAI. See Configuration.
Agent memory
Let an AI agent remember things across turns:
From the project's README.