CodeWise is a Flask-based web application that allows you to ingest, index, and chat with your code repositories using large language models (LLMs) such as GPT-OSS (via Hugging Face or Ollama). It provides a modern UI for exploring code, saving chat history, and leveraging retrieval-augmented generation (RAG) for code understanding.

• Repository Ingestion: Add public or private GitHub repositories for analysis.
• Code Indexing: Automatically chunk, embed, and index code for semantic search.
• Chat with Code: Ask questions about your codebase and get context-aware answers from an LLM.
• Saved Chats: Save and organize chat responses as markdown files for future reference.
• Modern UI: File explorer, chat interface, and saved chat management.
• Supports Multiple LLM Backends: Use Hugging Face API or run models locally with Ollama.

• Python 3.12 or newer
• Git (for repo ingestion)
• Ollama (optional, for local LLM inference)

Install Python dependencies:

pip install -r requirements.txt

Create a .env file in the project root for sensitive configuration:

HF_API_KEY=your_huggingface_api_key_here

• HF_API_KEY is required if you use Hugging Face's hosted LLMs. in this version though you can hard code the key in rag_repo.py
• For Ollama, no API key is needed.

python app.py

The app will be available at http://localhost:5000.

• Enter a GitHub repository URL in the UI, or add path to the private repo.
• (Optional) Provide a GitHub Personal Access Token for private repos.
• Click "Ingest" to process and index the repository.

• Select the ingested repository from the dashboard.
• Use the chat interface to ask questions about the codebase.
• The LLM will answer using context from the indexed code.

• Save any AI response as a markdown file.
• View, search, and reuse saved chats from the sidebar.

Beauty of NaviGit is that it let's the user go totally offline, does not even need a browser to interact with it, Most of larger codespaces have a shared server that can be only accessed with cli, so NaviGit is inbuilt with its own CLI support without the need of even turning on the flask App

• Terminal will prompt you to enter the github repo url, and PAT only needed for private repos, you may select your local path too
• After successful ingestion move on to next step

• This will start building the indexes if not already built
• Once built you can directly chat with GPT OSS for your queries related to the codebase!

• Ensure HF_API_KEY is set in your .env. for this version you can hardcode it directly intot rag_repo.py, search for langchain, hard code it inside that function
• The app will use Hugging Face's GPT-OSS endpoint.

• Install and run Ollama.
• Pull a model, e.g.:

  ollama pull gpt-oss

• In rag_repo.py, comment out the Hugging Face ask_llm function and uncomment the Ollama version:

  def ask_llm(prompt: str):
      client = OpenAI(
          base_url="http://localhost:11434/v1",
          api_key="ollama",  # Any string, not used by Ollama
      )
      completion = client.chat.completions.create(
          model="llama3",  # Or your preferred model
          messages=[{"role": "user", "content": prompt}],
      )
      return completion.choices[0].message.content

• Restart the Flask app.

• WebSocket Disconnects: In development, the chat may disconnect on code changes. This is normal due to Flask's auto-reload. Run on gunicorn rather than Werkzeug
• Indexing Errors: Ensure the repository URL is correct and accessible. For private repos, use a valid GitHub token. Timeouts may occur if your Network speed is a bottleneck
• LLM Errors: Check your API key, model name, and backend availability.
• Torch Errors During installing sentence transformers, you may run into pyTorch errors, you can fix them by just installing the appropriate torch versions!

noodl/
├── app.py
├── rag_repo.py
├── requirements.txt
├── .env
├── static/
│   └── index.css
├── templates/
│   ├── index.html
│   └── workspace.html
├── indexes/
│   └── <repo>/
│       ├── repo.index
│       ├── chunks.json
│       ├── graph.pkl
│       └── saved_chats/
└── ...

This project is Open Sourced, feel free to contribute, expand and fork this project into something amazing. Review the licenses of any LLMs or datasets you use with this tool.

• Flask
• Ollama
• Hugging Face
• OpenAI Python SDK