docstra 0.1.0

An LLM-powered CLI tool and Python package for generating repository-level documentation from code.

Docstra: LLM-Powered Software Documentation Generator & Chatbot

Docstra is an AI-powered tool designed to automatically generate in-depth, contextual documentation for code repositories. Using LangChain and ChromaDB, Docstra can analyze and link together various elements in a codebase, creating searchable, organized, and meaningful documentation that enhances the developer experience.

This project is a component of a master’s thesis at the University of Oslo, exploring the potential of Large Language Models (LLMs) in improving software documentation processes.

Key Features

• Automated Documentation: Scans and indexes code repositories to generate rich documentation with logical links between files, functions, classes, and modules.
• Multi-Platform Accessibility: Available as a local CLI and a FastAPI server, allowing integration into other applications.
• VSCode and IntelliJ Plugin Support: In development to enable seamless access within popular IDEs.
• Electron-based GUI Companion App: A cross-platform desktop application to view and interact with generated documentation.
• LLM-Powered Chatbot: Provides an interactive way to query the codebase and explore documentation.

Getting Started

Prerequisites

• Python 3.8+
• Poetry for dependency management. Install Poetry
• OpenAI API Key for access to LLM features (set up as an environment variable)

Installation

Clone this repository and install dependencies via Poetry.

git clone https://github.uio.no/docstra/docstra.git
cd docstra
poetry install

Environment Variables and Project Files

Docstra is designed to integrate smoothly into your existing project environment. When run from within an active repository, Docstra creates a dedicated .docstra folder in the root directory to store all generated embeddings and essential project files, keeping your documentation assets organized and easily accessible. This .docstra folder also houses a .env file for environment variables, including the required OpenAI API key. If the user has not yet added an API key, Docstra will prompt for one when the tool is run; the key can also be provided as a command-line argument. Once entered, the key is securely stored in .docstra/.env, allowing for streamlined future usage without additional prompts.

The .env file in the .docstra folder is automatically loaded when Docstra starts, making it the primary location for environment variables like API keys. This ensures your sensitive credentials are kept separate from your code and configuration.

Running Docstra

Docstra offers both CLI and FastAPI server options.

Command Line Interface (CLI)

Run Docstra directly from the command line:

poetry run docstra --help

Example of generating documentation:

poetry run docstra generate-docs path/to/your/codebase

FastAPI Server

To access Docstra as an API service, start the FastAPI server:

poetry run uvicorn docstra.server:app --reload

Access the API documentation at http://127.0.0.1:8000/docs (Swagger UI).

Usage

Docstra can process entire repositories or selected folders and files. It will extract structured metadata for each code component, allowing cross-referenced links for easy navigation and searchability.

Querying Documentation

The CLI offers a query command, allowing for specific questions about the codebase:

poetry run docstra query "How does authentication work?"

With the FastAPI server, you can send POST requests to /query with your question in the payload for direct querying from other applications.

Plugin and GUI Support

• VSCode and IntelliJ Plugins (in development): Access Docstra documentation within the editor for a seamless development experience.
• Electron-based Desktop App (coming soon): A graphical interface for non-technical users to explore codebase documentation.

Roadmap

• Integrate interactive LLM-based chatbot for conversational documentation
• Expand plugin support for major IDEs
• Develop Electron-based desktop application

License

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

Contact

For questions or feedback, please reach out to the project team at the University of Oslo. This project is an academic exploration into LLM-based developer tools aimed at improving the efficiency and accessibility of code documentation.

Project Repository: https://github.uio.no/docstra/docstra

This README provides a quick start guide to get up and running with Docstra, along with a look at the upcoming features and development goals of the project. Docstra is continually evolving to meet the needs of developers, researchers, and educators in the field of software engineering.