Personal Assistant - Multi-Memory AI Chatbot

Learning to fly in AI

A LangChain-based personal assistant with multi-memory capabilities, powered by Claude (Anthropic). This assistant remembers facts about you, tracks your todos, follows custom instructions, and stores everything locally on your machine.

Features

• Multi-Memory System

  • 📝 Instructions: System-level directives that guide the assistant's behavior
  • 📚 Facts: Factual information about you (preferences, personal info, habits, etc.)
  • ✅ Todos: Your task list with priorities and completion tracking

• TrustCall Integration: Uses Claude's structured output for reliable memory extraction

• Spy Monitoring: Comprehensive logging of all operations and memory updates

• Local Storage: All memories stored in JSON files on your machine

• Auto-Memory Extraction: Automatically extracts and updates memories from conversations

Architecture

src/
├── memory/           # Memory system (types, storage)
├── agent/            # LangChain agent and memory processor
└── monitoring/       # Spy logging system

memories/             # Local JSON storage (created on first run)
├── instructions.json
├── facts.json
└── todos.json

logs/                 # Session logs (created on first run)
└── session_*.jsonl

Setup

1. Install Dependencies

pip install -r requirements.txt

2. Configure API Key

Create a .env file in the project root:

cp .env.example .env

Edit .env and add your Anthropic API key:

ANTHROPIC_API_KEY=your_actual_api_key_here

Get your API key from: https://console.anthropic.com/

3. Run the Assistant

python main.py

Usage

Basic Chat

Just type naturally and the assistant will respond while automatically extracting and storing memories:

You: Hi! I'm John and I live in San Francisco.
Assistant: Hello John! Nice to meet you...

You: I prefer Python over JavaScript for backend work.
Assistant: Got it! I'll keep that in mind...

Commands

• /todos - View your todo list
• /facts - View all stored facts about you
• /instructions - View system instructions
• /summary - View session statistics
• /clear - Clear conversation history (keeps memories)
• /quit or /exit - Exit the program

Examples

Adding Todos:

You: Remind me to buy groceries tomorrow and finish the project report by Friday.
Assistant: I'll help you track those tasks...

Setting Preferences:

You: I prefer concise responses and always use metric units.
Assistant: Understood! I'll keep my responses concise and use metric units...

Asking Questions:

You: What do you know about my preferences?
Assistant: Based on what you've told me...

How It Works

1. Memory Extraction: After each conversation, the system uses Claude with structured output (TrustCall) to extract:

   • New facts about you
   • Todo items
   • Behavioral instructions

2. Local Storage: All memories are stored in memories/ as JSON files:

   • instructions.json - How the assistant should behave
   • facts.json - Information about you
   • todos.json - Your task list

3. Context Building: Each message includes current memories in the system prompt, so the assistant always has access to everything it knows about you

4. Monitoring: The spy() system logs all operations to logs/ for debugging and analysis

Memory Types

Instructions

• Priority-based (1-10)
• Guide assistant behavior
• Example: "Always be concise", "Use metric units"

Facts

• Categorized information about you
• Categories: preferences, personal_info, habits, work, hobbies, etc.
• Example: "User prefers dark mode", "User is a software engineer"

Todos

• Priority-based (1-5)
• Track completion status
• Optional due dates and tags
• Example: "Buy groceries [Priority 3]"

Development

Project Structure

• src/memory/ - Memory types and storage system

  • memory_types.py - Pydantic models for Instructions, Facts, Todos
  • memory_store.py - JSON-based persistence layer

• src/agent/ - Agent implementation

  • assistant.py - Main PersonalAssistant class
  • memory_processor.py - TrustCall-based memory extraction

• src/monitoring/ - Logging and monitoring

  • spy.py - Event logging system

Extending the System

Add a new memory type:

1. Define model in src/memory/memory_types.py
2. Add storage methods in src/memory/memory_store.py
3. Update MemoryUpdate model
4. Update memory processor prompt

Customize extraction: Edit the system prompt in src/agent/memory_processor.py

Requirements

• Python 3.8+
• Anthropic API key
• Dependencies listed in requirements.txt

Privacy & Security

• All data stored locally on your machine
• No data sent anywhere except to Anthropic's API for processing
• API key stored in .env (not committed to git)
• Memories stored in plain JSON (human-readable)

License

MIT License - Feel free to modify and use as you like!