AI Codebase Explorer

ai-codebase-explorer ingests a GitHub repository and lets you ask questions about the codebase from a single interactive command.

What It Does

Pipeline:

1. Clone or update repository.
2. Parse code files into chunks.
3. Build a dependency graph from Python + JS/TS imports.
4. Embed chunks and persist them in SQLite.
5. Retrieve relevant chunks for a user question.
6. Ask the LLM to answer strictly from retrieved evidence.

This supports questions like:

• "Where is authentication implemented?"
• "How does the API talk to the database?"
• "Which files control user login?"

Quick Start

python -m pip install -r requirements.txt

Set GEMINI_API_KEY in .env.

copy .env.example .env

Run:

python app.py

The app will prompt for:

• repo URL/path
• repo name
• questions in a loop (type exit to quit)

Implementation Notes

• Embeddings are stored in SQLite as JSON arrays for portability.
• Vector retrieval currently performs in-process cosine similarity.
• Dependency graph extraction is implemented for:
  • Python (import, from ... import ...)
  • JavaScript/TypeScript (import, export ... from, require)
• Answers are prompted to be source-grounded and include evidence references.