Persistent memory for AI assistants

An MCP server that gives AI assistants memory that survives between sessions. Local Redis, vector search, and knowledge graphs with zero configuration.

npx @n3wth/r3

Without r3

> What's my preferred stack?

I don't have any information
about your preferences.

With r3

> What's my preferred stack?

Based on our past conversations:
React + TypeScript, Tailwind,
Postgres with Drizzle ORM.

How it works

r3 runs a local Redis server with vector search. Your AI stores memories as embeddings and retrieves them by meaning, not just keywords.

Semantic searchKnowledge graphsLocal-only

See it in action

r3 storing and retrieving memories across sessions.

Antigravity CLI setup

agy mcp add r3 npx -y @n3wth/r3
agy mcp list

Start a session and use /mcp to inspect the connection.

Get started

MCP Desktop Clients

Add r3 to your MCP config file.

json
// MCP client config (e.g. .gemini/settings.json)
{
"mcpServers": {
"r3": {
"command": "npx",
"args": ["@n3wth/r3"]
}
}
}

MCP CLI Tools

Add with a single command.

bash
# MCP CLI tools
gemini mcp add r3 npx -y @n3wth/r3
gemini mcp list

Full setup guide

What you get

r3 runs entirely on your machine. Embedded Redis, vector search, and knowledge graphs with no external services.

Semantic Search

Cosine similarity ranking across 384-dimension vectors. Query by meaning, not keywords.

Knowledge Graph

Automatic entity extraction links memories into a traversable graph of relationships.

Fast local reads

Embedded Redis serves as both cache layer and vector store. Local embedding generation, no API calls.

MCP compatible

Works with any MCP-compatible client. Desktop apps, CLI tools, and custom integrations.

TypeScript SDK

Typed memory operations, search results, and configuration. Ships its own type declarations.

Fully local

Embedded Redis server, local vector store. No cloud services, no API keys required.

Your AI forgets everything between sessions.

One command adds persistent memory.

npx @n3wth/r3