LangChain AI Code Assistant

This project is an AI-powered code assistant that uses the LangChain framework and the Ollama language model to generate code and answer questions about provided code. The assistant can read code files from the project directory using the code_reader tool and generate code based on user prompts.

Features

• Analyzes code and answers questions about provided code
• Generates code based on user prompts
• Uses the code_reader tool to read contents of code files in the project directory
• Uses the file_lister tool to list files in a specified directory (defaults to project)
• Utilizes the LangChain framework and Ollama language model for code generation and analysis
• Provides a Python REPL tool for looking up Python documentation and examples
• Parses the generated code into a JSON object with code, description, and filename
• Saves the generated code to a file in the output directory
• Classifies the agent's response as either "code" or "conversation" using a response classifier

Installation

1. Clone the repository:

git clone https://github.com/yourusername/langchain-ai-code-assistant.git

2. Install the required dependencies:

pip install -r requirements.txt

3. Set up the necessary environment variables (e.g., API keys) in a .env file.

Usage

1. Place the code files you want the assistant to analyze in the project directory.

2. Run the main.py script:

python main.py

3. Enter a prompt for the AI code assistant. The assistant will generate code, provide a description, and save the code to a file in the output directory, or provide a conversational response.

4. To quit the program, enter 'q' when prompted for input.

Project Structure

• main.py: The main script that initializes the LangChain agent, tools, and prompts, and handles user input and output.
• agent_tools.py: Defines the CodeReader and FileLister tools that read code files and list files in a directory.
• prompts.py: Contains the prompts used by the LangChain agent, code parser, and response classifier.
• project/: Directory containing code files for the assistant to analyze.
• output/: Directory where the generated code files are saved.

Dependencies

• LangChain
• langchain-community
• dotenv
• ast

Contributing

Contributions are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request.

License

This project is licensed under the MIT License.

Acknowledgements

• LangChain for providing the framework for building language model-powered applications.
• Ollama for the language models used in this project.