CodeAutopsy

Codebase intelligence platform. Point it at any GitHub repository — public or private — and get a complete architectural analysis in seconds.

---

What it does

Understanding a large unfamiliar codebase takes hours. Reviewing a PR without knowing its blast radius is a gamble. CodeAutopsy solves both.

Six analysis modules run on every repository:

Module	What it does
Blueprint Map	Interactive dependency graph with 3 visualization modes — dependency flow, folder structure, codebase weight
Risk Radar	Finds articulation points, bridge edges, circular dependencies, and orphaned code
Diagnostic Engine	CFG analysis for unreachable code, missing error handling, and infinite loops
PR Impact	Blast radius prediction before a change ships
Arch Insights	Ranks files by architectural influence using PageRank and betweenness centrality
Read Docs	Instant documentation and onboarding guide for any repo

---

This is not an AI wrapper

Most "AI code analysis" tools dump a repository into an LLM and hope for the best.

CodeAutopsy runs four deterministic layers before any AI call is made:

Layer 1 — Repository Parser Fetches the full file tree in a single GitHub Git Trees API call. Filters noise intelligently — ignores node_modules, dist, lock files, binaries. Detects entry points via manifest parsing.

Layer 2 — Static Analysis Engine This is where the real work happens. Regex parsing extracts import/require statements across the codebase. A dependency adjacency list is constructed with local path resolution. Fan-in scores are computed to identify core utility modules. Files are ranked by architectural significance using Fan-in + Depth + Role weighting.

The graph analysis uses:

• Tarjan's algorithm — articulation point detection
• Brandes' algorithm — betweenness centrality scoring
• PageRank — architectural hub identification
• Force-directed layout — interactive dependency visualization

No AI is involved in this layer. The structural risks are real because the math is real.

Layer 3 — Structured AI Analysis Only the top-ranked files reach the LLM — preventing "lost in the middle" degradation. Output is forced into a strict typed JSON schema for consistent rendering. PR diffs are pre-scanned locally for hardcoded secrets before anything reaches the AI. If secrets are found, risk assessments are physically overridden to HIGH RISK regardless of what the LLM returns.

Layer 4 — Infrastructure Supabase PostgreSQL caching keyed by repo_url + commit_sha + analysis_version. Upstash Redis rate limiting (3 analyses/day free, 10 authenticated). @supabase/ssr for stateless auth and private repo token injection.

---

Private repo support

Authenticate with GitHub and run CodeAutopsy against your actual production codebase. Not just public demos.

---

GitHub Action

Make your repository self-documenting. The CodeAutopsy Action automatically generates and injects a live architecture map into your README on every push.

1. Add the map beacon to your README:

<!-- CODEAUTOPSY_MAP_START -->
<!-- CODEAUTOPSY_MAP_END -->

2. Add the workflow:

name: Update Architecture Map

on:
  push:
    branches: [main, master]

jobs:
  update-map:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: CodeAutopsy Architecture Sync
        uses: Sidhant0707/codeautopsy-action@v1.0.0
        with:
          github_token: ${{ secrets.GITHUB_TOKEN }}

---

Tech stack

Layer	Technology
Frontend	Next.js (App Router), TailwindCSS, Framer Motion
Graph Visualization	React Flow, D3.js
AI Analysis	Cerebras — Llama 3.3 70B
Repo Chat	Groq — Llama 3.1 8B
Data Fetching	GitHub REST API, Git Trees API
Database & Auth	Supabase (PostgreSQL)
Rate Limiting	Upstash Redis
Diagrams	Mermaid.js
Deployment	Vercel

---

Running locally

git clone https://github.com/Sidhant0707/codeautopsy
cd codeautopsy
npm install

Create .env.local:

GITHUB_TOKEN=your_github_token
GITHUB_FALLBACK_TOKEN=your_fallback_token
CEREBRAS_API_KEY=your_cerebras_api_key
NEXT_PUBLIC_SUPABASE_URL=your_supabase_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key
SUPABASE_SERVICE_KEY=your_supabase_service_key
UPSTASH_REDIS_REST_URL=your_upstash_url
UPSTASH_REDIS_REST_TOKEN=your_upstash_token

npm run dev

---

Contributing

The areas that would benefit most from contribution right now:

• Additional language support for the static analysis engine
• More graph algorithm implementations
• New LLM provider integrations
• UI improvements to the visualization layers

Fork the repo, create a feature branch, open a PR. See CONTRIBUTING.md for architecture guidelines.

---

License

MIT — see LICENSE for details.

---

Built by Sidhant Kumar — CS student at GLBITM, Greater Noida.
LinkedIn · Live demo