A deep dive into the engineering principles behind the world's first AI-powered codebase visualization tool.
CodeAutopsy doesn't just read code; it performs a structural analysis. Our engine fetches the repository via the GitHub API and uses a heuristic parser to identify the 'vital organs'—entry points, routers, and configuration files—to provide context within LLM token limits.
We use a proprietary scoring algorithm to rank files. Files with high 'Fan-in' (referenced by many other files) or specific naming conventions (e.g., App.tsx, index.js, main.py) are prioritized for AI analysis to ensure 80% architectural clarity from 20% of the code.
For every analysis, the tool maps dependencies and constructs a system prompt for Gemini 1.5 Flash. This allows for rapid-fire responses regarding execution flow and logic bottlenecks.
Your code is never stored on our servers. We fetch the contents, process them in memory for the AI context, and then discard them. Only the resulting JSON analysis is cached in Supabase for performance.
To get the most out of CodeAutopsy, ensure your repository has a clear README.md and standard naming conventions. Our AI uses these signals to better understand your architectural intent.