yrvelez / docuchat

Yet another document chat application, but with locally stored embeddings

DocuChat: A Document Chat Application with Local Embeddings

DocuChat is a chat application that allows users to communicate with a document. The application uses OpenAI embeddings that are stored locally, eliminating the need for an external data store (e.g., Pinecone).

Technology

DocuChat is built using HTML, PHP, and JavaScript. The front-end of the application is built using HTML/JS. The back-end of the application is built using PHP. The local embeddings are stored as JSON output inside the browser.

Requirements

To use DocuChat, you need to have an OpenAI API Key. You should edit the config.php file to include your API Key before using the application.

Usage

Upload a PDF file (<25 pages) to start a conversation. Enter your message in the input field and click "Send". The application will use OpenAI embeddings to find relevant chunks of the PDF and display a summary after incorporating the most relevant chunks into the context.

Performance

The application is optimized for documents up to 50 pages in length (25 page limit can be removed). Longer documents may result in decreased performance.

Installation

Clone this repository. Edit the config.php file to include your OpenAI API Key. Install and configure a web server with PHP and MySQL. Upload the files to your web server. Open the application in a web browser.

Credits

DocuChat was developed by Yamil Ricardo Velez. It uses OpenAI's API for text embeddings and chat functionality.