Skip to content

Getting Started with Kagura AI v4.0

Universal AI Memory Platform - 10-minute setup

Kagura is a universal memory layer that connects all your AI platforms (Claude, ChatGPT, Gemini, etc.) with shared context and memory.

📋 What is Kagura v4.0?

Kagura v4.0 = MCP-native Universal Memory

  • For Claude Desktop: Local MCP server with all 31 tools
  • For ChatGPT: HTTP/SSE connector with memory access
  • For Teams: Self-hosted API with authentication
  • For Developers: REST API + Python SDK

🚀 Quick Start (Choose Your Path)

Setup time: 5 minutes

# Install Kagura
pip install kagura-ai[full]

# Auto-configure Claude Desktop
kagura mcp install

# Restart Claude Desktop
# That's it! Kagura is now available in Claude

Try it in Claude Desktop: 

"Remember: I prefer Python for backend development"
"What do you know about my preferences?"

See: MCP Setup Guide

Path 2: ChatGPT Connector User

Setup time: 10 minutes

  1. Start Kagura API: 

    # Using Docker
docker compose up -d

# Or local
pip install kagura-ai[api]
uvicorn kagura.api.server:app --port 8000

  2. Expose with ngrok (for testing): 

    ngrok http 8000
# Get URL: https://abc123.ngrok.app

  3. Configure ChatGPT:

  4. Enable Developer Mode
  5. Add Connector:
    • URL: https://abc123.ngrok.app/mcp
    • Name: Kagura Memory

See: MCP over HTTP/SSE Guide

Path 3: Self-Hosted Production

Setup time: 30 minutes

# Clone repository
git clone https://github.com/JFK/kagura-ai.git
cd kagura-ai

# Configure
cp .env.example .env
nano .env  # Set DOMAIN and POSTGRES_PASSWORD

# Deploy
docker compose -f docker-compose.prod.yml up -d

# Generate API key
docker compose -f docker-compose.prod.yml exec api \
  kagura api create-key --name "production"

# Verify
curl https://your-domain.com/api/v1/health

See: Self-Hosting Guide

🧩 Key Features

1. Universal Memory

Store memories once, access from any AI:

# Via MCP tool (Claude Desktop, ChatGPT, etc.)
memory_store(
    user_id="jfk",
    agent_name="global",
    key="coding_style",
    value="Always use type hints in Python",
    scope="persistent",
    tags='["python", "best-practices"]'
)

2. Graph Memory

Track relationships and patterns:

# Record interaction
memory_record_interaction(
    user_id="jfk",
    query="How do I write async functions?",
    response="...",
    metadata={"topic": "python", "skill_level": "intermediate"}
)

# Analyze patterns
memory_get_user_pattern(user_id="jfk")

3. Remote Access

Access your memory from anywhere:

  • ChatGPT Connector: HTTP/SSE transport
  • API Keys: Secure authentication
  • Tool Filtering: Automatic security (no file ops remotely)

4. Export/Import

Own your data completely:

# Backup
kagura memory export --output ./backup

# Restore
kagura memory import --input ./backup

📚 Next Steps

For Claude Desktop Users

  1. Complete MCP Setup
  2. Try built-in tools: kagura mcp tools
  3. Explore memory operations

For ChatGPT Users

  1. Setup HTTP/SSE Connector
  2. Generate API key: kagura api create-key
  3. Connect and test

For Self-Hosters

  1. Follow Self-Hosting Guide
  2. Configure SSL/TLS with Caddy
  3. Set up backups

For Developers

  1. REST API Reference
  2. Architecture Overview
  3. Memory Export/Import

🔍 Available Commands

# MCP Management
kagura mcp serve           # Start MCP server (Claude Desktop)
kagura mcp install         # Auto-configure Claude Desktop
kagura mcp tools           # List available tools
kagura mcp doctor          # Run diagnostics
kagura mcp connect         # Configure remote connection
kagura mcp test-remote     # Test remote API

# API Key Management
kagura api create-key      # Generate API key
kagura api list-keys       # List all keys
kagura api revoke-key      # Revoke key

# Memory Management
kagura memory export       # Export to JSONL
kagura memory import       # Import from JSONL

# System
kagura --version           # Show version

💬 Support

  • Documentation: https://kagura-ai.com/docs
  • GitHub Issues: https://github.com/JFK/kagura-ai/issues
  • Discussions: https://github.com/JFK/kagura-ai/discussions

Version: 4.0.0 Protocol: MCP (Model Context Protocol) License: Apache 2.0