y# 🐕 PomPom-AI: Intelligent Memory System for Qodo AI
PomPom-AI (PomPom Artificial Intelligence) - A smart MCP (Model Context Protocol) server that provides persistent memory capabilities for Qodo AI. Just like Pompompurin's friendly and reliable nature, PomPom-AI remembers everything important and helps your AI assistant provide personalized, intelligent responses across all conversations.
🎯 Personal Setup for Qodo AI Integration
This repository is configured for personal use with Qodo AI, providing long-term memory storage and retrieval capabilities.
Qodo AI MCP Configuration
{
"pompom-ai": {
"url": "http://localhost:8051/sse"
}
}
🚀 Quick Start Guide
Prerequisites
- Python 3.12+
- OpenRouter API key (for Claude 3.7 Sonnet)
- Supabase PostgreSQL database (configured)
Installation
- Clone and setup:
git clone <your-repo-url>
cd pompom-ai
pip install -e .
- Configure environment:
Copy
.env.example to .env and update with your credentials:TRANSPORT=sse
HOST=0.0.0.0
PORT=8051
LLM_PROVIDER=openrouter
LLM_BASE_URL=https://openrouter.ai/api/v1
LLM_API_KEY=your-openrouter-api-key
LLM_CHOICE=anthropic/claude-3.7-sonnet
DATABASE_URL=your-supabase-postgresql-url
- Start the server:
- Test connectivity:
🧠 How It Works - Detailed Explanation
Architecture Overview
Qodo AI ←→ MCP Protocol ←→ PomPom-AI Server ←→ Mem0 ←→ ChromaDB + PostgreSQL
Component Breakdown
1. MCP Server (src/main.py)
- FastMCP Framework: Handles MCP protocol communication
- SSE Transport: Server-Sent Events for real-time communication on port 8051
- Lifespan Management: Initializes and manages Mem0 client connection
- Three Core Tools: Exposes memory operations to Qodo AI
save_memory(text: str)
- Purpose: Store any information in long-term memory
- Usage: When you tell Qodo AI something important to remember
- Process:
- Receives text from Qodo AI
- Processes through Claude 3.7 Sonnet for fact extraction
- Generates embeddings using ChromaDB's built-in model
- Stores in both ChromaDB (vectors) and PostgreSQL (metadata)
get_all_memories()
- Purpose: Retrieve all stored memories for context
- Usage: When Qodo AI needs complete memory context
- Process:
- Queries Mem0 for all memories associated with default user
- Returns paginated results (50 items default)
- Provides full context for conversation continuity
search_memories(query: str, limit: int = 3)
- Purpose: Find relevant memories using semantic search
- Usage: When Qodo AI needs specific information
- Process:
- Converts query to embeddings
- Performs vector similarity search in ChromaDB
- Returns most relevant memories ranked by relevance
3. Memory Configuration (src/utils.py)
LLM Configuration (OpenRouter + Claude)
llm_config = {
"provider": "openai", # OpenRouter uses OpenAI-compatible API
"config": {
"model": "anthropic/claude-3.7-sonnet",
"temperature": 0.2, # Low temperature for consistent memory processing
"max_tokens": 1500
}
}
Embedding Configuration (ChromaDB Built-in)
- No external API calls: Uses ChromaDB's default embedding function
- Local processing: Embeddings generated locally for privacy
- No additional costs: No embedding API fees
Vector Store Configuration (ChromaDB)
vector_store_config = {
"provider": "chroma",
"config": {
"collection_name": "mem0_memories",
"path": "./chroma_db" # Local SQLite database
}
}
4. Data Flow When You Use Qodo AI
Saving a Memory:
You: "Remember that I prefer PowerShell for automation tasks"
↓
Qodo AI → MCP Protocol → PomPom-AI → save_memory("I prefer PowerShell for automation tasks")
↓
Claude 3.7 Sonnet processes and extracts key facts
↓
ChromaDB generates embeddings locally
↓
Stored in: ChromaDB (vectors) + PostgreSQL (metadata)
↓
PomPom-AI Response: "Successfully saved memory: I prefer PowerShell for automation tasks"
Retrieving Memories:
You: "What do you know about my preferences?"
↓
Qodo AI → MCP Protocol → PomPom-AI → search_memories("preferences", limit=5)
↓
ChromaDB performs vector similarity search
↓
PomPom-AI returns relevant memories about your preferences
↓
Qodo AI uses this context to provide personalized responses
5. Storage Architecture
ChromaDB (Local - ./chroma_db/)
- Vector embeddings: Semantic representations of memories
- Fast similarity search: Sub-second query responses
- Local SQLite: No external dependencies
- Collection:
mem0_memories
PostgreSQL (Supabase)
- Metadata storage: User associations, timestamps
- Structured data: Relationships and memory organization
- Cloud backup: Persistent storage across devices
- Scalability: Handles large memory datasets
View Current Memories
# Python script
python show_current_memories.py
# PowerShell script
.\show_memories.ps1
Visual Dashboard
# Streamlit dashboard
streamlit run chroma_viewer.py
# HTML dashboard with live data
python dashboard_server.py
Server Testing
# Test server connectivity
.\test_server.ps1
📊 Memory Analytics
The system tracks:
- Total memories stored
- Memory categories/collections
- Average memory length
- Search frequency patterns
- Memory creation timestamps
🔒 Privacy & Security
- Local embeddings: No data sent to external embedding APIs
- Encrypted storage: PostgreSQL with SSL
- Local processing: ChromaDB runs entirely on your machine
- API key security: Environment variables only
🎛️ Configuration Options
Memory Processing
- Temperature: 0.2 (consistent fact extraction)
- Max tokens: 1500 (detailed memory processing)
- Model: Claude 3.7 Sonnet (high-quality reasoning)
Search Parameters
- Default limit: 3 memories per search
- Similarity threshold: Automatic (ChromaDB optimized)
- Collection scope: Single user (isolated memories)
🚀 Usage Patterns with Qodo AI
"Remember that I work as a software engineer and prefer Python and PowerShell"
"I live in timezone UTC+3"
"My favorite IDE is VS Code"
Project Context
"I'm working on a MCP server project using FastMCP and Mem0"
"The project uses OpenRouter for LLM and ChromaDB for vectors"
"Port 8051 is used for the SSE transport"
Preferences & Settings
"I prefer detailed explanations with code examples"
"Always use PowerShell for Windows automation tasks"
"Format code blocks with syntax highlighting"
🔄 Maintenance
Regular Tasks
- Monitor ChromaDB size (
./chroma_db/) - Check PostgreSQL connection health
- Review memory quality and relevance
- Update API keys as needed
Troubleshooting
- Server won't start: Check
.env configuration - Memory not saving: Verify PostgreSQL connection
- Search not working: Restart server to refresh ChromaDB
- Qodo AI can't connect: Confirm port 8051 is open
- ChromaDB: Optimized for <1000 memories per collection
- PostgreSQL: Indexed for fast metadata queries
- Memory size: Optimal range 50-500 characters per memory
- Search speed: Sub-100ms for typical queries
🎯 Best Practices
- Memory Quality: Store specific, actionable information
- Regular Cleanup: Remove outdated or irrelevant memories
- Categorization: Use consistent language for similar topics
- Testing: Regularly test memory retrieval accuracy
- Backup: PostgreSQL provides automatic cloud backup
This system transforms Qodo AI into a truly personalized assistant that remembers your preferences, project context, and important information across all conversations.
🐕 Why "PomPom-AI"?
Just like Pompompurin is known for being:
- 🤗 Friendly & Reliable - PomPom-AI is always there to help remember what's important
- 🧠 Smart & Attentive - Intelligently processes and organizes your memories
- 💛 Loyal Companion - Grows smarter about your preferences over time
- 🎯 Focused & Efficient - Quickly finds exactly what you need when you need it
PomPom-AI = PomPom (friendly like Pompompurin) + AI (Artificial Intelligence)