Visigine

GEO/SEO-Auditor für AI-Suchen (ChatGPT, Claude, Perplexity, Gemini …). Frontend: React 19 + Vite. Backend: Node 20 + Express.

Dev-Setup

# Frontend deps
npm install

# Backend deps
npm install --prefix server

# Backend secrets
cp server/.env.example server/.env
# → MISTRAL_KEY=...  in server/.env eintragen

Start

npm run dev

Startet client (Vite, Port 5173) und server (Express, Port 3001) parallel. Vite proxied /api/* automatisch an den Backend-Port.

Health-Check: curl http://localhost:3001/health → {"ok":true}.

Secrets

Der Mistral-API-Key liegt nur in server/.env als MISTRAL_KEY. Es gibt keine VITE_*-Keys im Frontend — der Browser sieht den Schlüssel nie.

Stack

• React 19, Vite 8 (kein TypeScript)
• Express 4, Node 20 native fetch
• Mistral mistral-large-latest (nur serverseitig)

Production Deployment

Backend (server/)

Required env vars:

• MISTRAL_KEY — Mistral API key (server-only, never exposed to the client).
• PORT — port to bind (default 3001).
• ALLOWED_ORIGINS — comma-separated CORS whitelist (e.g. https://www.visigine.de,https://visigine.de).
• NODE_ENV=production — disables debug mode unconditionally.
• ALLOW_PRIVATE_HOSTS=0 — keep at 0 in production (SSRF protection).

Run: npm start inside server/. The Express listener binds 0.0.0.0:$PORT so containers and reverse proxies can reach it.

Health check: GET /health → { "ok": true }.

Frontend

Static build: npm run build in the repo root → dist/. Serve via any static host. The frontend calls /api/analyze — your hosting layer must proxy /api/* to the backend (nginx / Caddy / Cloudflare Workers / etc.).

Operations

• Rate limit: 20 requests / 60 s / IP on /api/analyze.
• In-memory cache: 1 h TTL, 1000 entries (LRU). X-Cache: HIT|MISS header.
• Every request gets an X-Request-Id and a structured log line.
• ?debug=1 is ignored when NODE_ENV=production.