Ross AIDocs

Deployment

Deploy Ross AI to Vercel with environment variables and path-based documentation.

Deployment

Ross AI is designed for Vercel serverless functions.

Deploy the app

  1. Push the repo to GitHub
  2. Import the project in Vercel (framework preset: Next.js)
  3. Deploy — no build command changes needed

Environment variables

In Vercel → Project → Settings → Environment Variables, add:

VariableEnvironmentsNotes
OPENAI_API_KEYProduction, PreviewYour OpenAI secret key
NEXT_PUBLIC_APP_URLProductione.g. https://ross-ai-nine.vercel.app
NEXT_PUBLIC_DOCS_URLOptionalOnly if using a custom docs subdomain (see below)

If OPENAI_API_KEY is only set for Preview, production will return 503. Enable it for Production and redeploy.

Documentation URL

Production docs are served at:

https://ross-ai-nine.vercel.app/docs

No extra Vercel domain setup is required. Navbar and footer "Docs" links point to /docs on the same host.

Why not docs.ross-ai-nine.vercel.app?

Vercel does not support subdomains of the default *.vercel.app hostname (no DNS or SSL for docs.project-name.vercel.app). Attempting to open that URL will fail with a connection error.

If you need a separate docs hostname later:

  1. Buy or use a custom domain (e.g. ross-ai.com)
  2. Add docs.ross-ai.com in Vercel → Domains with a CNAME to Vercel
  3. Set NEXT_PUBLIC_DOCS_URL=https://docs.ross-ai.com and re-enable subdomain middleware in the repo

Production limits

LimitValueWhy
Max PDF size4 MBVercel serverless request body limit (~4.5 MB)
Contract text sent to OpenAI~15,000 charsCost control
Function timeout60 smaxDuration in the API route
Rate limit10 requests / 15 min / IPAbuse protection

PDF parsing on Vercel

Production uses unpdf (pure JavaScript) instead of pdf-parse, which relies on native bindings that fail in serverless Linux environments.

Redeploy checklist

  • OPENAI_API_KEY enabled for Production
  • Latest commit deployed (includes unpdf fix)
  • Docs reachable at /docs on your production URL
  • Test with a PDF under 4MB