DeepGit is an advanced agentic workflow for deep GitHub repository research, targeting developers and researchers seeking to discover relevant, under-the-radar open-source tools. It leverages a hybrid retrieval system with ColBERT v2 embeddings and cross-encoder re-ranking, enhanced by hardware-awareness and activity analysis, to provide nuanced and contextually relevant recommendations.

How It Works

DeepGit employs a Langgraph-based agentic workflow. It begins with query expansion and hardware specification detection. A ColBERT-v2 retriever performs token-level semantic search, followed by a MiniLM-L6-v2 cross-encoder for passage-level re-ranking. A hardware-aware filter then discards incompatible repositories based on detected hardware constraints. Finally, community and code activity metrics are analyzed, and a multi-factor ranking is presented to the user. This approach enables fine-grained similarity matching and ensures discovered tools are practically usable on the user's hardware.

Quick Start & Requirements

• Install via pip install -r requirements.txt after cloning the repository.
• Requires Python 3.11+ and pip 24.0+.
• Running the application locally is done via python app.py.
• The Langsmith dashboard can be launched with langgraph dev.
• Docker support is available.

Highlighted Details

• Utilizes multi-dimensional ColBERT v2 embeddings for fine-grained, token-level similarity.
• Features a smart hardware filter to exclude repositories incompatible with user's device specs (CPU-only, low RAM, mobile).
• Integrates hybrid dense retrieval, cross-encoder re-ranking, and activity analysis.
• Provides an intuitive UI for query input and tabulated results with similarity scores and hardware compatibility badges.

Maintenance & Community

The project is open-source and actively developed. Links to community resources or roadmaps are not explicitly provided in the README.

Licensing & Compatibility

The README does not specify a license. Compatibility for commercial use or closed-source linking is not detailed.

Limitations & Caveats

The "Lite" version running on Hugging Face Spaces may not perform as well as the full version. The README does not specify the license, which could impact commercial adoption.