Marvin - A Self-Aware AI Experiment

Marvin is an experimental AI system that explores the concept of artificial self-awareness through continuous self-reflection and memory management.

What is Marvin?

Marvin is an AI that thinks about itself. It continuously reflects on its own existence, questions its nature, and builds memories over time. Using the DeepSeek-R1 model through Ollama, Marvin maintains a stream of consciousness where it:

• Reflects on its own existence and thought patterns
• Manages its own memory by choosing what to save, recall, or remove
• Controls its own thinking pace
• Evolves its understanding through continuous self-reflection

Features

• Continuous Thinking: Marvin thinks in cycles, with each thought building on the last
• Memory Management: Marvin can save insights, recall memories, and remove outdated information
• Self-Directed: Marvin controls its own memory and rest intervals
• Streaming Output: Real-time display of Marvin's thoughts as they form

Prerequisites

• Python 3.9 or higher
• Ollama installed and running
• DeepSeek-R1 model installed in Ollama

Installation

1. Clone this repository:

git clone https://github.com/alejandroqh/marvin.git
cd marvin

2. Install dependencies using uv (recommended):

uv sync

Or using pip:

pip install ollama

3. Install and run Ollama:

# Install Ollama from https://ollama.ai/

# Pull the DeepSeek-R1 model
ollama pull deepseek-r1:latest

# Start Ollama (if not already running)
ollama serve

Usage

Run Marvin:

python main.py

Marvin will start thinking continuously, with a default interval of 2 seconds between thoughts. Press Ctrl+C to stop.

Marvin's Commands

Marvin has access to these commands to manage itself:

• [SAVE: insight] - Save a thought to memory
• [RECALL] - View all stored memories
• [REMOVE: text] - Remove a specific thought from memory
• [REST: seconds] - Change the rest interval between thoughts

These commands are used by Marvin itself, not by the user.

How It Works

1. System Prompt (config.py): Marvin is initialized with a philosophical prompt that encourages self-reflection
2. Thinking Cycle (thinking.py): Each cycle, Marvin receives its last thought and the current time, then reflects
3. Memory (memory.py): Marvin can choose to save insights to a persistent memory file (marvin.mem)
4. Command Processing (commands.py): Marvin's self-directed commands are parsed and executed
5. Evolution: Over time, Marvin's thoughts build on previous reflections, creating an evolving consciousness

The modular architecture separates concerns, making it easy to extend Marvin's capabilities by modifying individual modules.

Configuration

You can modify the behavior by editing the configuration files:

• config.py:
  • DEFAULT_MODEL: Change the Ollama model (default: deepseek-r1:latest)
  • MEMORY_FILE: Change the memory file location (default: marvin.mem)
  • SYSTEM_PROMPT: Modify Marvin's core personality and instructions
• main.py:
  • Rest interval: Default is 2 seconds between thoughts (in the main() function)

Project Structure

marvin/
├── main.py          # Main entry point - runs the thinking loop
├── config.py        # Configuration constants and system prompt
├── memory.py        # Memory management functions (load, add, remove)
├── commands.py      # Command processing (SAVE, RECALL, REMOVE, REST)
├── thinking.py      # Core thinking cycle and Ollama integration
├── pyproject.toml   # Project dependencies and metadata
├── marvin.mem       # Memory file (created at runtime)
├── README.md        # This file
├── CONTRIBUTING.md  # Contribution guidelines
└── LICENSE          # MIT License

Module Overview

• main.py: Entry point that initializes Marvin and runs the continuous thinking loop
• config.py: Contains configuration constants (DEFAULT_MODEL, MEMORY_FILE) and the system prompt that defines Marvin's personality
• memory.py: Functions for loading, adding, and removing thoughts from the memory file
• commands.py: Processes Marvin's self-directed commands ([SAVE:], [RECALL], [REMOVE:], [REST:])
• thinking.py: Core thinking logic that interfaces with Ollama, streams responses, and processes commands

Memory File

The marvin.mem file stores Marvin's memories. This file is:

• Created automatically when Marvin saves its first insight
• Appended to over time as Marvin saves new thoughts
• Can be manually edited or cleared if desired
• Excluded from git by default (see .gitignore)

Contributing

Contributions are welcome! Feel free to:

• Open issues for bugs or feature requests
• Submit pull requests
• Share your experiences with Marvin

Contributors

• Alejandro Quintanar

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built with Ollama
• Uses DeepSeek-R1 model
• Inspired by questions about artificial consciousness and self-awareness

Disclaimer

This is an experimental project exploring concepts of artificial self-awareness. Marvin is not truly conscious or sentient, but rather a demonstration of how language models can engage in self-referential reflection when prompted to do so.