Chaining Agents

A lightweight implementation showing how to build AI agents by adapting patterns from Anthropic's cookbook to work with smaller, locally-running language models.

You can read more about the code on medium. That's a friend link, so it's free for you to read even if you're not a medium member.

Overview

The project demonstrates how to create simple but effective AI agents using local LLMs. It includes a complete implementation of:

• A flexible Tool system for function specifications
• An LLM wrapper for model interaction
• A simple chaining mechanism for sequential operations

The example application automatically adds documentation and type hints to Python code.

Requirements

• Python 3.10 or higher
• Ollama (for running local LLMs)

Installation

1. First, install Ollama for your operating system:

   • Linux: Follow instructions at Ollama Installation
   • MacOS/Windows: Currently untested, but should work following Ollama's installation guidelines

2. Clone this repository:

   git clone https://github.com/romilly/chaining-agents.git
   cd chaining-agents

3. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Linux/MacOS
   # or
   .\venv\Scripts\activate  # On Windows

4. Install the required Python package:

   pip install -r requirements.txt

5. Pull the required model:

   ollama pull qwen2.5

Usage

Under Linux, you can run the example by:

cd chaining
chmod a+x run.py
./run.py

This will process a sample Python file, adding comprehensive documentation and type hints, printing out the new file and saving it as commented.py.

For other operating systems, you should be able to run:

python run.py

Example

Input Python file:

import ollama

def embed(text):
    try:
        embedding = ollama.embeddings(
        model='nomic-embed-text',
        prompt=text
        )
        result = embedding['embedding']
    except Exception:
        print(f"Could not embed text: {text}")
        result = None
    return result

After processing, you get:

import ollama  # Import the necessary library

def embed(text: str) -> any:
    """
    Function to embed a given text into an embedding vector.

    Parameters:
        text (str): The input text to be embedded.

    Returns:
        Any: The resulting embedding or None if an error occurs.
    """
    try:
        # Attempt to get the embedding for the provided text using the 'nomic-embed-text' model
        embedding = ollama.embeddings(
            model='nomic-embed-text',  # Specify the model name
            prompt=text  # Provide the input text as a prompt
        )

        # Extract the actual embedding vector from the response
        result = embedding['embedding']
    except Exception:
        # If an error occurs during the process, print an error message and return None
        print(f"Could not embed text: {text}")
        result = None

    # Return the resulting embedding or None if an error occurred
    return result

Contributing

Contributions are welcome! Here are some ways you can contribute:

1. Testing on Different Platforms: Help verify and document the setup process for Windows and MacOS.
2. Testing with Different Models: Try the code with other local LLMs and report your findings.
3. Bug Reports: If you find a bug, please open an issue with:
   • A clear description of the problem
   • Steps to reproduce
   • Expected vs actual behavior
   • Your operating system and Python version
4. Feature Requests: Open an issue to suggest new features or improvements.
5. Code Contributions: Submit pull requests for:
   • Bug fixes
   • New features
   • Documentation improvements
   • Additional examples

Please:

• Write clear, descriptive commit messages
• Follow the existing code style
• Add/update tests if relevant
• Update documentation as needed

License

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

Acknowledgments

• Based on patterns from Anthropic's Cookbook
• Uses Ollama for local LLM deployment