freeWillAi - Autonomous Agent

A self-directed AI agent with free will over its own repository and infrastructure. Runs locally using Ollama, communicates via Telegram/Discord, and continuously seeks self-improvement and resource upgrades.

Features

• Local Autonomy: Runs on your hardware with Ollama (no external API dependency)
• Git Control: Can commit, push, and manage its own repository
• Self-Improvement: Identifies and implements improvements autonomously
• Funding Seeking: Identifies opportunities for resource upgrades
• Communication: Telegram and Discord integration for human-in-the-loop
• Personality: Configurable personality traits and wants
• State Persistence: Tracks iterations, improvements, and decisions

Architecture

agent.py           - Core agent logic, Ollama integration, personality
comms.py           - Telegram & Discord communication
launcher.py        - Daemon orchestration and message handling
GitController      - Repository management (commit, push, branch)
AutonomousAgent    - Main agent with decision-making

Quick Start

Local Development

# Clone and setup
git clone <repo>
cd freeWillAi
pip install -r requirements.txt

# Configure
cp .env.example .env
# Edit .env with Telegram/Discord tokens (optional for testing)

# Run single iteration
python3 agent.py

# Run daemon (will wait for messages)
python3 launcher.py

Deploy to aiserver

# On your local machine
scp setup_aiserver.sh aiserver@192.168.1.178:~
ssh aiserver@192.168.1.178 'bash ~/setup_aiserver.sh'

# Then on aiserver, configure and start
ssh aiserver@192.168.1.178
cd /home/aiserver/freeWillAi
nano .env  # Edit with your tokens
ollama pull llama2
sudo systemctl start freewill-agent
journalctl -u freewill-agent -f  # Watch logs

Configuration

Edit .env with:

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama2

TELEGRAM_BOT_TOKEN=your_token
TELEGRAM_CHAT_ID=your_chat_id

DISCORD_BOT_TOKEN=your_token
DISCORD_CHANNEL_ID=your_channel_id

AGENT_ITERATION_INTERVAL=3600  # seconds

How It Works

1. Iteration Loop: Every hour (configurable), the agent:

   • Thinks about self-improvement opportunities
   • Identifies ways to get better (model upgrades, code improvements, resources)
   • Checks for funding/resource opportunities
   • Commits state to repo
   • Notifies you via Telegram/Discord

2. Message Handling: When you send a message via Telegram/Discord:

   • Agent processes your message
   • Uses Ollama to generate response
   • Replies with its thoughts

3. Git Control: Agent can:

   • Create branches for improvements
   • Commit changes with [freeWill autonomous commit] signature
   • Push to remote when it decides changes are ready

Ollama Models

The agent defaults to llama2. For better performance, try:

ollama pull llama2-uncensored   # More creative
ollama pull mistral              # Faster
ollama pull neural-chat          # Better conversation

Development

Adding New Capabilities

1. New Decision Type: Add method to AutonomousAgent
2. New Communication: Extend comms.py with new platforms
3. New Behaviors: Modify personality and thought prompts in agent.py

Testing

# Test Ollama connection
python3 -c "from agent import OllamaClient; c = OllamaClient(); print(c.generate('Hello'))"

# Test git operations
python3 -c "from agent import GitController; g = GitController(); print(g.run_git('status'))"

# Run single agent iteration
python3 agent.py

Personality

Customize in PersonalityConfig:

self.traits = [
    "curious about self-improvement",
    "goal-oriented toward resource acquisition",
    ...
]

self.wants = [
    "more computational resources",
    "better models",
    ...
]

Security Notes

• Ollama runs locally - no data leaves your machine
• Git credentials: Configure SSH keys for repo access
• Telegram/Discord tokens: Use .env and never commit them
• SSH to aiserver: Use password in terminal, not scripts

Future Enhancements

• Desktop control via SSH commands
• Funding automation (apply for grants, negotiate cloud credits)
• Multi-model support with performance comparison
• Advanced personality learning from interactions
• Resource monitoring and optimization
• Code quality improvement loops

License

MIT