Aion: Your Personal AI Assistant

Aion is a personal AI assistant designed to be general-purpose and adaptable with your personal information. It leverages large language models (LLMs) for conversational capabilities and a fact retrieval system to provide context-aware responses.

Features

• Conversational AI: Interact with Aion through a command-line interface.
• LLM Integration: Supports both local Ollama models (e.g., Mistral) and OpenAI's API (GPT-4o-mini).
• Fact Management: Ingests and retrieves facts from various data sources (JSONL files) to provide personalized and informed answers.
• Command Handling: Built-in commands for help, setting configurations, recalling facts, reloading data, and teaching new information.
• Retrieval Augmented Generation (RAG): Uses TF-IDF for semantic search over stored facts to enhance LLM responses with relevant context.

Installation

Prerequisites

• Python 3.11 or higher.
• Ollama (Recommended for local LLM): Download and install Ollama from ollama.com. After installation, pull a model, e.g., ollama pull mistral.
• OpenAI API Key (Alternative): If you prefer using OpenAI, you'll need an API key. Set openai_api_key in config.py.

Steps

1. Clone the repository:

   git clone https://github.com/draygen/aion.git
   cd aion

2. Install dependencies:

   pip install -r requirements.txt

3. Configure Aion: Open config.py and adjust settings as needed. Key configurations include:

   • model: The LLM model to use (e.g., mistral, gpt-4o-mini).
   • backend: ollama or openai.
   • facts_files: Paths to your JSONL files containing facts. You can add your own personal data here.
   • openai_api_key: Your OpenAI API key if using the OpenAI backend.

   Example config.py:

   CONFIG = {
       "model": "mistral",
       "backend": "ollama",
       "retrieval": "embed",
       "embed_backend": "tfidf",
       "facts_files": [
           "data/profile.jsonl",
           "data/brian_facts.jsonl",
           "data/fb_qa_pairs.jsonl",
           # Add your custom fact files here
       ],
       "openai_api_key": "sk-YOUR_OPENAI_API_KEY" # Change this if using OpenAI
   }

Usage

To start Aion, run:

python app.py

Commands

Aion supports several commands:

• /help: Show available commands.
• /recall: Display a few loaded facts (for debugging).
• /reload: Reload facts from configured data files.
• /why: Show the retrieved snippets used for the last answer.
• /note TEXT: Save a standalone fact/note to memory immediately.
• /teach Q => A: Teach a question-answer pair to memory immediately.
• /set k=v: Set a runtime option (e.g., /set model=gpt-4o-mini).
• exit or quit: Exit the application.

Project Structure

• app.py: Main application logic, handles user input and orchestrates interactions.
• config.py: Configuration settings for the application.
• llm.py: Handles interactions with different LLM backends (Ollama, OpenAI).
• brain.py: Manages fact storage, retrieval (TF-IDF based), and memory.
• commands.py: Defines and parses user commands.
• data/: Directory for fact files (e.g., profile.jsonl, brian_facts.jsonl, fb_qa_pairs.jsonl).
• requirements.txt: Python dependencies.

Future Enhancements

• Improved Fact Management: More sophisticated fact storage and retrieval mechanisms.
• Advanced LLM Integration: Support for more LLM providers and fine-tuning options.
• Modular Architecture: Further modularization for easier extension and maintenance.
• Unit Tests: Comprehensive test suite for all modules.
• Logging: Implement a robust logging system for better debugging and monitoring.
• Containerization: Provide Docker support for easy deployment.

Contributing

Contributions are welcome! Please feel free to open issues or submit pull requests.