Code Agent

A comprehensive project for exploring AI code agents and their capabilities, featuring step-by-step lessons on building and deploying intelligent agents.

Description

This repository contains a series of Jupyter notebooks that demonstrate how to build, monitor, and deploy AI code agents. Each step builds upon the previous one, gradually introducing more complex concepts and capabilities.

Project Structure

The project is organized into multiple steps, each focusing on different aspects of AI code agents:

• Step2: Building a simple code agent with basic tools
• Step4: Monitoring and evaluating agent performance
• Step5: Building a deep research-like agent with web search capabilities

Step-by-Step Guide

Step 2: Building a Simple Code Agent

The Step2 directory contains a notebook (Lesson2.ipynb) that demonstrates how to build a simple code agent with the following components:

• Environment Setup: Importing libraries and setting up authentication
• Data Preparation: Creating sample datasets for ice cream suppliers
• Tool Creation: Defining specialized tools for calculations
• Model Setup: Initializing a language model using Hugging Face API
• Agent Configuration: Creating and configuring a CodeAgent
• Agent Execution: Running the agent with specific queries
• Complex Problem Solving: Handling multi-factor analyses

Step 4: Monitoring and Evaluating Your Agent

The Step4 directory contains a notebook (Lesson4.ipynb) that focuses on:

• Setting up tracing: Configuring OpenTelemetry for monitoring agent behavior
• Tracing agent runs: Capturing detailed information about agent execution
• Building evaluation systems: Creating metrics to assess agent performance
• Analyzing tool usage: Understanding when and how agents use different tools

Step 5: Building a Deep Research Agent

The Step5 directory contains a notebook (Lesson5.ipynb) that demonstrates:

• Web search integration: Using Tavily API to search the web
• Multi-agent systems: Creating specialized agents for different tasks
• Data visualization: Generating maps and charts based on research findings
• Quality control: Validating agent outputs against requirements

Getting Started

Prerequisites

• Python 3.13 (recommended, but 3.11+ should work)
• Virtual environment tool (venv, conda, etc.)
• Jupyter Notebook or JupyterLab

Environment Setup

1. Clone the repository:

   git clone https://github.com/Rupeebw/coding-agent.git
   cd coding-agent

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

API Keys Setup

Create a .env file in the root directory with the following API keys:

HF_API_KEY=your_huggingface_api_key
OPENAI_API_KEY=your_openai_api_key
TAVILY_API_KEY=your_tavily_api_key
DLAI_LOCAL_URL=http://localhost:{port}/

Running the Notebooks

1. Start Jupyter:

   jupyter notebook

2. Navigate to the desired step directory (e.g., Step2, Step4, or Step5)

3. Open the corresponding notebook (e.g., Lesson2.ipynb, Lesson4.ipynb, or Lesson5.ipynb)

4. Run the cells sequentially

Dependencies and Troubleshooting

Critical Dependencies

• smolagents==1.13.0: Core library for building code agents
• openai==1.66.3: Required for OpenAI models
• tavily-python==0.5.1: Required for web search functionality
• nbformat>=4.2.0: Required for notebook processing
• plotly==6.0.1: Required for data visualization
• pandas==2.2.3 and numpy==2.2.3: Required for data manipulation

Common Issues and Solutions

1. OpenTelemetry Connection Errors:

   • If you encounter connection errors with OpenTelemetry in Step4, modify the code to use a console exporter instead of Phoenix.
   • Alternatively, ensure a collector is running on the specified port.

2. API Key Issues:

   • Ensure all required API keys are correctly set in your .env file
   • Check that the .env file is in the root directory
   • Verify that load_dotenv() is called before accessing environment variables

3. Model Selection:

   • The notebooks use different models (Hugging Face models, OpenAI models)
   • If you prefer using OpenAI models, ensure you have the correct API key and modify the model initialization code

4. Memory Issues:

   • Some models require significant memory
   • If you encounter memory errors, try using a smaller model or reducing batch sizes

5. Visualization Issues:

   • If maps don't display correctly in Step5, ensure plotly and its dependencies are correctly installed
   • Try running pip install plotly --upgrade if needed

Model Preferences

The default configuration uses:

• OpenAIServerModel with gpt-4.1-mini for code generation
• This can be changed to other models based on your preference and API access

License

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