This project features a DSPy-based agent that can run commands and interact with users. It uses OpenRouter models for processing requests.

1. Ensure you are in the project root directory: /home/tom/git/agent/

2. Run the setup script

   ./setup_env.sh

   This creates a Python virtual environment in .venv, installs the requirements and generates an .env.example file. Copy .env.example to .env and fill in your API keys.

3. (Manual setup) If you prefer to do things manually, create and activate the virtual environment and install dependencies yourself:

   python3.11 -m venv .venv
   source .venv/bin/activate
   uv pip install -r requirements.txt

4. Set up OpenRouter API key:

   • Create a free account at OpenRouter
   • Add your API key to environment variables:

     echo "export OPENROUTER_API_KEY='your_api_key'" >> ~/.bashrc
     source ~/.bashrc

5. Run the SimpleDSPy agent:

   python agent_simpledspy.py

The agent's performance depends on quality training data stored in .simpledspy/modules/. To improve results:

1. Review logged interactions:

   less .simpledspy/modules/*/logged.jsonl

2. Add good samples to training data: Copy well-performing examples from logged.jsonl to corresponding training.jsonl files

3. Correct problematic samples: Edit training.jsonl to fix incorrect responses

4. Add new training examples: Create new entries in training.jsonl for desired behaviors

• The main agent script is agent_simpledspy.py
• Training data is stored in .simpledspy/modules/
• Use --lm flag to switch models: flash, r1, or dv3

  python agent_simpledspy.py --lm r1