I made a memory system for AI\LLMs

Smerrills · August 12, 2025, 3:43am

I made a memory system. Which i’m going to use in my own AI/LLM setup. I made a memory system, and I also made it completely open source. Here’s the info about it:

Persistent AI Memory System

Community Call to Action: Have you made improvements or additions to this system? We want to include your work! Every contributor will be properly credited in the final product. Whether it’s bug fixes, new features, or documentation improvements - your contributions matter and will help shape the future of AI memory systems. Submit a pull request today!

GITHUB LINK - GitHub - savantskie/persistent-ai-memory: A persistent local memory for AI, LLMs, or Copilot in VS Code.

Recent Changes (2025-08-11)

Robust startup: No background tasks (maintenance, import, deduplication) auto-start; explicit, safe startup only
Advanced deduplication: All major database classes (conversations, AI memories, schedule, VS Code projects) now have advanced duplicate detection and migration logic
Centralized maintenance: Maintenance and deduplication are now generic and centralized, not format-specific
Parity with Friday: Open-source version matches Friday’s main repo for all core features, including migration, deduplication, and startup sequencing

A comprehensive AI memory system that provides persistent, searchable storage for AI assistants with conversation tracking, MCP tool call logging, and intelligent scheduling.

Exciting News: Desktop App Coming Soon!

We’re thrilled to announce the development of a new desktop application that will make the Persistent AI Memory System even more powerful and user-friendly!

Upcoming Desktop Features:

Universal LLM Integration:
- LM Studio - Direct API integration and conversation tracking
- Ollama - Real-time chat capture and model switching
- llama.cpp - Native support for local models
- Text Generation WebUI - Full conversation history
- KoboldCpp - Seamless integration
- More platforms coming soon!
Enhanced GUI Features:
- Real-time conversation visualization
- Advanced memory search interface
- Interactive context management
- Visual relationship mapping
- Customizable dashboard
- Dark/Light theme support
Extended Capabilities:
- Multiple MCP protocol support
- Cross-platform conversation sync
- Enhanced embedding options
- Visual memory navigation
- Bulk import/export tools
- Custom plugin support

Stay tuned for the beta release! Follow this repository for updates.

Multiple Installation Options Available: We’ve created 4 different ways to install this system - from one-command installation to manual setup - so you can get started immediately regardless of your platform or preference!

New to this from Reddit? Check out the Reddit Quick Start Guide for a super simple setup!

Quick Installation - Choose Your Method!

Option 1: One-Command Installation (Linux/macOS) - FASTEST

curl -sSL https://raw.githubusercontent.com/savantskie/persistent-ai-memory/main/install.sh | bash

Option 2: Windows One-Click Installation - EASIEST

curl -sSL https://raw.githubusercontent.com/savantskie/persistent-ai-memory/main/install.bat -o install.bat && install.bat

Option 3: Manual Installation - MOST CONTROL

git clone https://github.com/savantskie/persistent-ai-memory.git
cd persistent-ai-memory
pip install -r requirements.txt
pip install -e .

Option 4: Direct pip Installation - SIMPLEST

pip install git+https://github.com/savantskie/persistent-ai-memory.git

Health Check

After installation, verify everything is working:

python tests/test_health_check.py

Available Tools

Core Memory Tools (Available in All Environments)

These tools are available in all environments (LM Studio, VS Code, etc.):

Memory Management:
- search_memories - Search through stored memories using semantic similarity
- store_conversation - Store conversation messages
- create_memory - Create a new curated memory entry
- update_memory - Update an existing memory entry
- get_recent_context - Get recent conversation context
Schedule Management:
- create_appointment - Create calendar appointments
- create_reminder - Set reminders with priorities
System Tools:
- get_system_health - Check system status and database health
- get_tool_usage_summary - Get AI tool usage statistics
- reflect_on_tool_usage - AI self-reflection on tool patterns
- get_ai_insights - Get AI’s insights and patterns

IDE-Specific Tools

These tools are only available in specific development environments:

VS Code Tools

save_development_session - Save VS Code development context
store_project_insight - Store development insights
search_project_history - Search project development history
link_code_context - Link conversations to specific code
get_project_continuity - Get context for continuing development work

Features

Enhanced Memory System:
- SQLite-based persistent storage across all databases
- Registry-based extensible import system
- Advanced duplicate detection and migration logic in all major database classes
- Centralized, generic maintenance and deduplication routines
- Robust, explicit startup (no auto background tasks)
- Database-backed deduplication across all sources
- Incremental imports (only new messages)
- Enhanced error handling with detailed logging
- Automatic system maintenance and optimization
- AI-driven self-reflection and pattern analysis
- Cross-database relationship tracking
- Smart memory pruning and archival
Dedicated Chat Format Support:
- Independent parsers for each chat GUI
- No merged/refactored import logic
- Easy addition of new chat formats
- Format-specific metadata preservation
- Source-aware deduplication
Core Features:
- Vector Search using LM Studio embeddings
- Real-time conversation monitoring
- MCP server with tool call logging
- Advanced AI self-reflection system:
  - Usage pattern detection and analysis
  - Automated performance optimization
  - Tool effectiveness tracking
  - Learning from past interactions
  - Continuous system improvement
- Multi-platform compatibility
- Zero configuration needed
- Full feature parity with Friday’s main repo
Platform Support:
- LM Studio integration
- VS Code & GitHub Copilot
- Koboldcpp compatibility
- Ollama chat tracking
- Cross-platform (Windows/Linux/macOS)

Quick Start

Basic Usage

import asyncio
from ai_memory_core import PersistentAIMemorySystem

async def main():
    # Initialize the memory system
    memory = PersistentAIMemorySystem()
    
    # Store a memory
    await memory.store_memory("I learned about Python async programming today")
    
    # Search memories
    results = await memory.search_memories("Python programming")
    print(f"Found {len(results)} related memories")
    
    # Store conversation
    await memory.store_conversation("user", "What is async programming?")
    await memory.store_conversation("assistant", "Async programming allows...")

if __name__ == "__main__":
    asyncio.run(main())

MCP Server (for Claude Desktop, etc.)

# Run as MCP server
python ai_memory_core.py

File Monitoring

# Monitor conversation files (like ChatGPT exports)
from ai_memory_core import PersistentAIMemorySystem

memory = PersistentAIMemorySystem()
memory.start_conversation_monitoring("/path/to/conversation/files")

Architecture

The system includes 5 specialized databases with enhanced cross-source integration:

Conversations:
- Multi-source chat history with embeddings
- Registry-based extensible import system
- Independent parsers per chat format
- Database-backed deduplication
- Source tracking and sync status
- Cross-conversation relationships
- Incremental import tracking
- Comprehensive metadata per source
AI Memories:
- Long-term persistent AI memories
- Cross-source knowledge synthesis
- Relationship tracking between memories
Schedule:
- Time-based events and reminders
- Cross-platform calendar integration
- Smart scheduling with context
VS Code Projects:
- Project context and file tracking
- Development conversation tracking
- Code change history integration
- Context-aware project insights
MCP Tool Calls:
- Model Context Protocol interaction logging
- Tool usage analytics
- Self-reflection capabilities
- Performance monitoring

Configuration

The system works with zero configuration but can be customized:

memory = PersistentAIMemorySystem(
    db_path="custom_memory.db",
    embedding_service_url="http://localhost:1234/v1/embeddings"
)

System Maintenance

The system now includes automatic and centralized maintenance features:

Centralized Maintenance & Deduplication:
- Generic, registry-based maintenance routines
- Advanced duplicate detection and migration logic for all database classes
- No format-specific or auto-startup tasks; all maintenance is explicit and robust
Database Optimization:
- Automatic vacuum and reindex
- Smart memory pruning
- Performance monitoring
- Index optimization
Error Management:
- Comprehensive error logging
- Automatic recovery procedures
- Failed operation retry
- Data consistency checks
AI Self-Reflection:
- Tool usage pattern analysis
- Performance optimization suggestions
- Automated system improvements
- Usage statistics and insights

Examples

Check the examples/ directory for:

Basic memory operations
Conversation tracking
MCP server setup
Vector search demonstrations
Custom chat format integration
Deduplication system usage
Registry-based importing
Source tracking setup

Platform Integration Guides

AI Platforms

Koboldcpp Integration - Complete setup guide for Koboldcpp compatibility
LM Studio - Built-in support for embeddings and conversation capture
VS Code - MCP server integration for development workflows
SillyTavern - MCP server support with character-specific memory tools
Ollama - Compatible through file monitoring and HTTP API approaches

Integration Methods

File Monitoring - Automatic conversation capture from chat logs
HTTP API - Real-time memory access via REST endpoints
MCP Protocol - Standardized tool interface for compatible platforms

Cross-Source Memory Integration

The system now provides comprehensive cross-source memory management:

Source Tracking:
- Automatic source detection and monitoring
- Per-source metadata and sync status
- Error tracking and recovery
- Active source health monitoring
Relationship Management:
- Cross-conversation linking
- Context preservation across platforms
- Conversation continuation tracking
- Reference and fork management
Supported Sources:
- VS Code/GitHub Copilot
- ChatGPT desktop app
- Claude/Anthropic
- Character.ai
- SillyTavern (file monitoring + MCP server)
- text-generation-webui
- Ollama
- Generic text/markdown formats
- Custom source support via plugins
Sync Features:
- Real-time sync status tracking
- Source-specific metadata preservation
- Robust deduplication across sources
- Failure recovery and retry logic

Testing

Run the complete test suite:

python tests/test_health_check.py
python tests/test_memory_operations.py
python tests/test_conversation_tracking.py
python tests/test_mcp_integration.py

� API Reference

Core Methods

Memory Operations

store_memory(content, metadata=None) - Store a persistent memory
search_memories(query, limit=10) - Semantic search of memories
list_recent_memories(limit=10) - Get recent memories

Conversation Tracking

store_conversation(role, content, metadata=None) - Store conversation turn
search_conversations(query, limit=10) - Search conversation history
get_conversation_history(limit=100) - Get recent conversations

MCP Tool Calls

log_tool_call(tool_name, arguments, result, metadata=None) - Log MCP tool usage
get_tool_call_history(tool_name=None, limit=100) - Get tool usage history
reflect_on_tool_usage() - AI self-reflection on tool patterns

System Health

get_system_health() - Check system status and database health

Development

Setting up for Development

git clone https://github.com/savantskie/persistent-ai-memory.git
cd persistent-ai-memory
pip install -e ".[dev]"

Running Tests

pytest tests/

Contributing

We welcome contributions! This system is designed to be:

Modular: Easy to extend with new memory types
Platform-agnostic: Works with any AI assistant that supports MCP
Scalable: Handles large conversation histories efficiently

Roadmap

Semantic Tagging Assistant - AI-powered memory categorization
Memory Summarization - Automatic TL;DR for long conversations
Deferred Retry Queue - Resilient file import with retry logic
Memory Reflection Engine - Meta-insights from memory patterns
Export/Import Tools - Backup and migration utilities

Community Requests & Platform Support

Recently Added Platforms (Based on Reddit Community Feedback):

SillyTavern - AI character chat interface with conversation logging
Gemini CLI - Google’s Gemini command line interface support
Open WebUI - Local web-based LLM interface (multiple install locations)

Total Platform Support: 11+ Chat Platforms

LM Studio, Ollama, VS Code Copilot, ChatGPT, Claude, Perplexity, Jan AI, Open WebUI, Text Generation WebUI, SillyTavern, Gemini CLI

Upcoming Community Requests:

GraphDB Integration - Graph database support for relationship mapping (community requested)
Discord Bot Integration - Chat logging for Discord AI bots
Telegram Bot Support - Conversation tracking for Telegram bots
API Standardization - Universal chat format for easier platform integration

Have a platform request? Open an issue or submit a PR - all contributors get credited!

License

MIT License - feel free to use this in your own AI projects!

� Contributors

This project is the result of a collaborative effort between humans and AI assistants:

@yourusername - Project vision, architecture design, and testing
GitHub Copilot - Core implementation, database design, MCP server development, and tool call logging system
ChatGPT - Initial concept development, feature recommendations, and architectural guidance over 3 months of development

�🙏 Acknowledgments

This project represents a unique collaboration between human creativity and AI assistance. After 3 months of conceptual development with ChatGPT and intensive implementation with GitHub Copilot, we’ve created something that could genuinely change how AI assistants maintain memory and context.

Special thanks to:

ChatGPT for the original insight that “If this ever becomes open source? It’ll become the standard.”
GitHub Copilot for the breakthrough implementation that solved foreign key constraints and made real-time conversation capture work flawlessly
The open source community for inspiring us to share this foundational technology

Built with determination, debugged with patience, and designed for the future of AI assistance.

GITHUB LINK - GitHub - savantskie/persistent-ai-memory: A persistent local memory for AI, LLMs, or Copilot in VS Code.

If this project helps you build better AI assistants, please give it a star!