RAGit

RAGit is an open-source project that demonstrates how to build a Retrieval-Augmented Generation (RAG) system for understanding and exploring codebases.
Using LangChain, local embeddings, and free LLMs (CodeLlama), this tool can:

• Parse a repository and extract functions, classes, and docstrings.
• Embed and index code snippets for semantic search.
• Generate commit and PR messages based on code changes.
• Answer natural-language questions like:
  • "What does function foo do?"
  • "Which file contains the class MyParser?"
  • "Is there any function that connects to the database?"

---

Key Features

• Fully Free & Local – Uses HuggingFace for embeddings and Ollama for local LLM inference.
• Vector Search – Powered by ChromaDB or FAISS.
• Automatic Commit & PR Messages – Generates helpful summaries of staged changes.
• Extensible – Add custom parsing, retrievers, or UI layers.

---

Roadmap

1. Basic RAG pipeline (parser → embeddings → retriever → LLM).
2. Commit/PR generation from code diffs. ✅
3. Conversational mode (multi-turn Q&A).
4. UI for interactive repo queries.
5. GitHub Action for automated PR summaries.

---

Tech Stack

• LangChain for RAG orchestration.
• HuggingFace Sentence-Transformers for free embeddings.
• ChromaDB for vector storage.
• Ollama for local LLMs.
• Git and diff parsing for commit/PR analysis.
• Python for all components.

---

Repository Structure

RAGit/
│
├── README.md
├── requirements.txt
├── src/
│   ├── __init__.py
│   ├── code_parser.py         # Extract functions/classes from repo
│   ├── embedder.py            # Chunk & embed code with sentence-transformers
│   ├── vectorstore.py         # Chroma/FAISS storage and retrieval
│   ├── rag_pipeline.py        # RAG chain (Retriever + LLM)
│   ├── pr_writer/             # Generate commit/PR messages from diffs
│   │   ├── diff_reader.py     # Extract and parse git diffs
│   │   ├── pr_formatter.py    # Prompt and format commit/PR summaries
│   │   └── utils.py           # Helper functions
│   └── main.py                # Entry point: query the codebase
├── examples/
│   └── demo_repo/             # A small repo for testing
└── tests/
    ├── test_parser.py
    ├── test_embedding.py
    ├── test_rag_pipeline.py
    └── test_pr_writer.py