Remembra - AI Memory Layer

Persistent memory for AI applications. Self-host in 5 minutes.

What Is This?

Remembra is a universal memory layer for LLMs. It solves the fundamental problem that every AI forgets everything between sessions.

from remembra import Memory

memory = Memory(user_id="user_123")

# Store memories
memory.store("User prefers dark mode and works at Acme Corp")

# Recall with context
result = memory.recall("What are user's preferences?")
print(result.context)
# → "User prefers dark mode. Works at Acme Corp."

MCP Server (Claude Code / Cursor)

Remembra ships with a built-in Model Context Protocol server. Any MCP-compatible AI assistant (Claude Code, Claude Desktop, Cursor, etc.) can use it as persistent memory.

Setup

pip install remembra[mcp]

Add to your Claude Code config:

claude mcp add remembra \
  -e REMEMBRA_URL=http://localhost:8787 \
  -e REMEMBRA_API_KEY=your_key \
  -- remembra-mcp

Or add manually to .mcp.json in your project:

{
  "mcpServers": {
    "remembra": {
      "command": "remembra-mcp",
      "env": {
        "REMEMBRA_URL": "http://localhost:8787",
        "REMEMBRA_API_KEY": "your_key"
      }
    }
  }
}

MCP Tools

MCP Resources

Environment Variables

Why We're Building This

The Problem

Every AI app needs memory. Developers hack together solutions using vector databases, embeddings, and custom retrieval logic. It's complex, fragmented, and everyone rebuilds the same thing.

Current Solutions Fall Short

• Mem0: Pricing jumps from $19 to $249, self-hosting is complex
• Zep: Academic, complex to deploy
• Letta: Not production-ready
• LangChain Memory: Too basic, no persistence

Our Approach

• Self-host in 5 minutes: One Docker command, everything bundled
• MCP-native: Works with Claude Code and Cursor out of the box
• Open source core: MIT license, own your data
• Built for production: Entity resolution, temporal decay, hybrid search

Core Features

Hybrid Search

Vector (semantic) + BM25 (keyword) search combined. Finds memories even when the query doesn't match exact words.

Entity Resolution

Knows that "Adam", "Adam Smith", "Mr. Smith", and "my husband" are the same person. Automatically extracts and links people, organizations, locations, and concepts.

Temporal Awareness

Memories have time context. TTL support. Ebbinghaus-inspired decay curves. Historical ("as of") queries.

Hybrid Storage

Vector (Qdrant) + Graph (relationships) + Relational (SQLite metadata) in one system.

Observability Dashboard

See what's stored, debug retrievals, visualize entity graphs.

Quick Start

1. Start the Server

docker run -d -p 8787:8787 remembra/remembra

2. Install the SDK

Python:

pip install remembra

from remembra import Memory

memory = Memory(
    base_url="http://localhost:8787",
    user_id="user_123",
    project="my_app"
)

# Store
result = memory.store("User's name is John. He's a software engineer at Google.")
print(result.extracted_facts)
# → ["John is a software engineer at Google."]

# Recall
result = memory.recall("Who is the user?")
print(result.context)
# → "John is a software engineer at Google."

# Forget
memory.forget(memory_id=result.memories[0].id)

JavaScript/TypeScript:

npm install @remembra/client

import { Remembra } from '@remembra/client';

const memory = new Remembra({
  url: 'http://localhost:8787',
  apiKey: 'rem_xxx',
});

// Store
const stored = await memory.store('Alice is the CTO of Acme Corp');
console.log(stored.extracted_facts);

// Recall
const result = await memory.recall('Who leads Acme?');
console.log(result.context);
// → "Alice is the CTO of Acme Corp."

// Entities
const entities = await memory.listEntities({ type: 'person' });

API Reference

REST Endpoints

Full API docs available at http://localhost:8787/docs (Swagger UI).

Documentation

• Architecture - Technical architecture details
• API Reference - Full API documentation
• Product Spec - Product specification

License

MIT License - Use it however you want.

Built by DolphyTech | remembra.dev