GitHub: https://github.com/RuliLG/docc
Background
Started this as my capstone project but the problem was real – I kept cloning repos and having no idea where to start. Reading docs doesn't answer specific questions, and asking ChatGPT about code it hasn't seen leads to hallucinations.
How it works
Points Claude Code CLI (or OpenCode) at your repository, gets it to analyze the actual code, then renders the response as an interactive player. Monaco editor shows the exact files and line numbers while explaining what's happening. You can enable TTS and literally listen to the walkthrough.
The AI response is structured JSON, so you get: - TL;DR summary - Relevant code blocks with specific line ranges - Explanatory markdown for each section
Tech stack
- Backend: FastAPI + Python - Frontend: React + Monaco editor - AI: Claude Code CLI / OpenCode CLI for repo analysis - TTS: ElevenLabs or OpenAI with caching - Runs in Docker
Current limitations
- It's a capstone project, so rough in places - Performance depends on repo size and AI provider - TTS caching helps but first runs can be slow
Looking for feedback on:
1. What questions do you most often ask about unfamiliar codebases? 2. Would you prefer this as a CLI tool vs web app? 3. Any AI providers you'd want supported?
The codebase is on GitHub, MIT licensed. Happy to answer questions about the implementation.