CapgeminiDCX AI Playground Hackathon 2026

Guides

Deployment guide

Where and how to deploy an app that runs server-side AI calls.

A typical AI app needs server-side code so provider keys and privileged tools stay off the client. Static-only hosting can serve a frontend, but the AI backend must run somewhere else.

The default path. Deploy from v0, or connect a Git repo so non-production branches create previews and the configured production branch creates production deployments.

  • Deploy from v0 or a supported template.
  • Set env vars (like AI_GATEWAY_API_KEY) in Project → Settings → Environment Variables.
  • Watch for the free-tier limits below.

Heads up on the AI key: the AI SDK reads AI_GATEWAY_API_KEY automatically. On Vercel, deployments can also authenticate with zero setup via the built-in VERCEL_OIDC_TOKEN — but for local development you'll still need to set AI_GATEWAY_API_KEY in .env.local.

Free (Hobby) limits that can bite:

LimitValue
Function duration (Fluid Compute, now the default)300s (5 min) default and max on Hobby
Function duration (older projects without Fluid Compute)10s default, 60s max
Deployment rate limits60 per 5 min, 100 per hour, and 100 per day
Build time per deployment45 min
Git organisation repos on HobbyCannot be connected — use an approved personal repo or a Pro/Enterprise team

Note: Fluid Compute is enabled by default for projects created after April 2025, which is why the function timeout is now 5 minutes (not the old 10s/60s). Only legacy projects that opted out still hit the shorter limit.

Need Vercel Enterprise? We can support that for the hackathon — talk to the organisers or ask in the 🆘 Help channel.

Won't work: GitHub Pages

GitHub Pages serves static files. It cannot run an AI SDK backend or protect a server-side key. It is only viable if your backend is hosted elsewhere; for this hackathon, use the recommended full-stack path.

Golden rules

  • Back up your work as you go. Use the current Git controls in v0 to connect the intended GitHub repo, or export/download the code periodically. Don't rely on a single browser tab.
  • Never commit .env.local or keys — set them in the platform's dashboard.
  • Test the deployed URL in a private window before you call it done.

Learn more

Official docs for the recommended stack, if you want to go deeper or get unstuck:

Stuck on a deploy? See the FAQ, or ask in the 🆘 Help channel.