Skip to content

DEPLOYMENT.md

Latest commit

 

History

History
58 lines (39 loc) · 2.56 KB

DEPLOYMENT.md

File metadata and controls

58 lines (39 loc) · 2.56 KB
Raw

Deployment Guide

PrivacyAnalyzer deploys to Vercel with a Neon Postgres database (via the Vercel Marketplace) for analysis caching and history.

Prerequisites

  • Vercel account
  • Neon Postgres database (provision from the Vercel Marketplace: Storage → Create → Neon, or bring your own Neon project)
  • OpenRouter API key(s) — get one
  • Firecrawl API key (optional, improves scraping + screenshots) — get one

Environment variables

Set these in Vercel → Project → Settings → Environment Variables (and in .env.local for local dev):

Variable Required Purpose
DATABASE_URL ✅ (for caching/history) Neon Postgres pooled connection string. POSTGRES_URL is also accepted.
OPENROUTER_API ✅ Primary OpenRouter API key.
OPENROUTER_API_1, OPENROUTER_API_2 ⬜ Additional keys for rotation / higher throughput.
FIRECRAWL_API_KEY ⬜ Enables Firecrawl scraping + homepage screenshots.
CRON_SECRET ⬜ If set, protects /api/cron/refresh-keys (sent as Authorization: Bearer <secret>).

When you add Neon from the Vercel Marketplace, DATABASE_URL / POSTGRES_URL are injected automatically.

If DATABASE_URL is absent, the app still runs — analyses are computed on every request and the history / results-by-domain pages report that the database is unavailable.

Deploy

  1. Import the GitHub repo into Vercel (Add New → Project).
  2. Add the environment variables above.
  3. Deploy. Vercel auto-detects Next.js; no extra build config is needed.

The database schema is created automatically on first use (idempotent CREATE TABLE IF NOT EXISTS), so there is no manual migration step.

CLI alternative:

npm i -g vercel
vercel        # preview
vercel --prod # production   (also: npm run deploy)

AI models

Analysis uses OpenRouter with automatic fallback, in order (ANALYSIS_MODELS in src/app/api/analyze/route.ts):

  1. openrouter/free — default (auto-routed free model)
  2. nvidia/nemotron-3-ultra-550b-a55b:free
  3. openai/gpt-oss-20b:free

If one model errors or is rate-limited, the next is tried (with API-key rotation across the configured keys).

Runtime notes

  • API routes run on the Node.js runtime (export const runtime = 'nodejs') for full scraping support.
  • Scraping order: Firecrawl (if key set) → Crawlee/Playwright → plain fetch.
  • The analysis cache keys on (domain, SHA-256(content)) with a 30-day freshness window, so unchanged policies skip the AI call.