BriefGPT is a locally-hosted tool designed for document summarization and querying using Large Language Models (LLMs). It targets users who prioritize data privacy and control, offering a secure way to interact with documents via a simple GUI.

How It Works

The tool processes documents by chunking them and creating FAISS indexes for efficient similarity search. A novel re-ranking function refines retrieved results by stripping stopwords and using fuzzy matching to improve relevance over pure similarity. For summarization, documents (or YouTube transcripts) are embedded, clustered using K-means, and then summarized in a two-step process: individual chunk summarization followed by a final aggregation.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Run the application: streamlit run main.py
• Place documents in the documents folder.
• For EPUB support, ensure pandoc is installed and in your PATH.
• API key setup via test.env.
• Local LLM support requires models in the models folder and environment variable configuration.

Highlighted Details

• Supports both OpenAI API and fully local LLM execution (LlamaCpp, GPT4ALL).
• Handles PDF and TXT documents, with EPUB support via pandoc.
• YouTube transcript summarization is also supported.
• Utilizes K-means clustering for thematic document grouping.

Maintenance & Community

This project was made for fun and is open to contributions and bug reports.

Licensing & Compatibility

The repository does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is experimental, potentially buggy, and not fully optimized, especially the local LLM functionality which may be significantly slower with variable quality. Summary state persistence is a noted TODO item.