Run Locally

Prerequisites: Node.js 22 and the Vercel CLI

1. Install dependencies: npm install
2. Link the Vercel project and pull its environment variables: vercel link, then vercel env pull .env.local
3. Run the app and the API together: vercel dev

npm run dev runs only the Vite front end. The editor and preview work, but Auto-Fix needs the /api/generate function, which runs under vercel dev or on Vercel.

Tests

npm test runs the Vitest suite. It includes a production build with a fake GEMINI_API_KEY in the environment, and it fails if that value or the name GEMINI_API_KEY appears in any file in dist/.

Deploying

The app deploys to Vercel as a Vite static site plus one Node.js function. vercel.json sets:

• Build command: npm run build
• Output directory: dist
• Function: api/generate.ts, with a 120 second maximum duration

The browser never sees the Gemini key. Auto-Fix posts { code, errorMessage } to /api/generate, and the function calls Gemini (gemini-3.1-pro-preview) with the key held on the server. Every call goes through spend-guard. Each visitor IP gets 30 requests per hour, and calls stop for the rest of the UTC day once spend reaches a daily cap.

Set these environment variables in the Vercel project:

Variable	Purpose
GEMINI_API_KEY	Gemini API key. Server only. Never give it a VITE_ prefix, because Vite puts VITE_ variables in the public bundle.
UPSTASH_REDIS_REST_URL	Upstash Redis REST URL, used for rate limits and spend counters. The Vercel Upstash integration sets KV_REST_API_URL instead, which also works.
UPSTASH_REDIS_REST_TOKEN	Upstash Redis REST token. KV_REST_API_TOKEN also works.
DAILY_CAP_USD	Daily Gemini spend limit for this app, in US dollars, for example 2.
TOTAL_DAILY_CAP_USD	Daily spend limit shared by every app that uses the same Redis database.

If the Gemini key or either cap is missing, /api/generate returns 503 and Auto-Fix shows "Auto-Fix is not available right now. Try again later." If Redis cannot be reached, the function refuses the call instead of spending without a limit.

Spend is counted from the token counts Gemini returns, at $2.00 per million input tokens and $12.00 per million output tokens. Thinking tokens bill as output. Update PRICES in server/fixMermaid.ts when Google changes its pricing.