SmartMemory

Give your LLM structured memory | Transform conversations into verified knowledge graphs

An MCP server that teaches AI assistants business rules through natural dialogue

[!CAUTION] Proof of Concept Only: This project is an experimental implementation of a Neuro-Symbolic architecture. It is designed to demonstrate how LLMs can interact with knowledge graphs for rule learning. It is NOT intended for production or professional use. Use it for research, experimentation, and learning purposes only.

🚀 Quick Start

New user? → 5-Minute Quick Start Guide

Having issues? → Troubleshooting Guide

Need to configure? → Configuration Reference

Want to understand how it works? → Neuro-Symbolic Architecture | Technical Architecture

Looking for specific docs? → 📚 Documentation Index

🎯 What is SmartMemory?

SmartMemory enables your favorite LLM (Claude, Gemini, etc.) to remember facts, learn business rules, and deduce new information.

You can use it in two main ways:

1. 💬 Conversational Mode (The "Brain")

• For: Individuals using LLM clients (Claude Desktop, etc.).
• Goal: Have your assistant remember facts and learn logic naturally as you chat.
• How: Configure it as an MCP server.
• 👉 Go to Setup

2. 🏗️ Supervision Mode (The "Factory")

• For: Teams, developers, or heavy users.
• Goal: Extract thousands of rules from documents (PDFs) and visualize the knowledge graph.
• How: Deploy the full Dashboard via Docker.
• 👉 Go to Setup

💬 Mode 1: Conversational Setup (MCP)

This mode gives your LLM "long-term memory" and logical deduction capabilities.

Option A: Install via Docker (Recommended) 🐳

Best for: Everyone! No Python installation required.

The SmartMemory Docker image is available on GitHub Container Registry.

Simply add to your MCP client configuration:

For Claude Desktop, edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "smart-memory": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "ghcr.io/mauriceisrael/smart-memory:latest"]
    }
  }
}

For Gemini (Cline), edit ~/.cline/mcp_settings.json:

{
  "mcpServers": {
    "smart-memory": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "ghcr.io/mauriceisrael/smart-memory:latest"]
    }
  }
}

Restart your client and you're done! ✅

Option B: Local Server (Private) 🔒

Best for: Developers & Privacy-conscious users who want to run from source.

Installation Steps (Local)

1. Clone & Install

   git clone https://github.com/MauriceIsrael/SmartMemory
   cd SmartMemory
   python3 -m venv venv
   source venv/bin/activate
   pip install -e .
   

2. Connect to Claude Desktop Edit your configuration file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

   {
     "mcpServers": {
       "smartmemory": {
         "command": "/absolute/path/to/SmartMemory/venv/bin/python",
         "args": ["-m", "smart_memory.server"]
       }
     }
   }
   

   (Replace /absolute/path/... with your actual path)

3. Chat! Restart Claude and try:

   "I know Bob. He goes to work by car. Can he vote?"

   See Interactive Demo below for what to expect.

🏗️ Mode 2: Supervision Setup (Docker)

This mode runs the Web Dashboard and API server. Ideally suited for:

• Visualizing the Knowledge Graph.
• Extracting rules from documents (PDFs).
• Hosting a shared memory server for a team.

Quick Start (Docker)

You don't need Python installed. Just Docker.

1. Run the container

   For Dashboard mode (web interface):

   For Ollama (local):

   docker run -p 8080:8080 \
     -e LLM_PROVIDER=ollama \
     -e LLM_MODEL=llama3 \
     -e LLM_BASE_URL=http://172.17.0.1:11434 \
     -v $(pwd)/brain:/app/data \
     ghcr.io/mauriceisrael/smart-memory:latest dashboard
   

   For OpenAI:

   docker run -p 8080:8080 \
     -e LLM_PROVIDER=openai \
     -e LLM_MODEL=gpt-4 \
     -e LLM_API_KEY=your-api-key \
     -v $(pwd)/brain:/app/data \
     ghcr.io/mauriceisrael/smart-memory:latest dashboard
   

   (Note: Add dashboard at the end to start web server. Without it, starts MCP mode)

   (The -v volume persists your knowledge graph and rules)

2. Open the Dashboard Go to http://localhost:8080

LLM Configuration

SmartMemory uses an LLM to extract business rules from documents. Configure it in two ways:

Option 1: Via Dashboard (Local Development)

1. Go to Admin page
2. Select your provider (Ollama, OpenAI, Anthropic, Google)
3. Enter your configuration (API key or Ollama URL)
4. Test connection
5. Save

Option 2: Via Environment Variables (Docker)

Already shown above! Pass -e LLM_PROVIDER=... when starting Docker.

Supported Providers:

• Ollama (local, free): llama3, qwen2.5-coder, mistral
• OpenAI: gpt-4, gpt-3.5-turbo
• Anthropic: claude-3-5-sonnet
• Google: gemini-1.5-pro

→ Full Configuration Guide

Advanced Deployment

We support deploying to Google Cloud Run, GitHub, etc. 👉 Read the Deployment Guide (DEPLOY.md) for full instructions on hosting it online.

Extracting Rules from Documents

Once your dashboard is running (or via CLI):

1. Upload a PDF (e.g., "Company_Policy.pdf").
2. Configure LLM: The server needs an LLM API key (OpenAI, Anthropic, or local Ollama) to read the doc.
3. Review: The system will propose rules. You can approve them in bulk in the dashboard.

💬 Interactive Demo: From Facts to Rules

Here is what happens when you use Conversational Mode:

> I know Bob
LLM: ✦ I've processed your statement "I know Bob".

> He goes at work by car
LLM: ✦ I've noted that Bob goes to work by car.

> Can Bob vote?
LLM: ✦ I can't answer yet. But since he drives, he is likely an adult.
     May I add a rule: "Drivers are adults"?

> yes
LLM: ✨ Rule 'drivers_are_adults' added.
     Now, may I add a rule: "Adults can vote"?

> yes
LLM: ✨ Rule 'adults_can_vote' added.
     ✦ Therefore, yes, Bob can vote.

🛠️ Technical Stack

• Backend: Python 3.11+, RDFLib, FastAPI
• Frontend: SvelteKit, TypeScript, TailwindCSS
• Reasoning: Neuro-Symbolic (LLM + SPARQL/OWL)
• Protocol: Model Context Protocol (MCP)

📜 License

MIT License - see LICENSE