EveStudioDocs

Connect to Vercel

Before an agent can reply, it needs a model credential. Studio walks you through getting one from Vercel — no keys to copy, no CLI to install.

Why it's needed

An Eve agent's default model routes through the Vercel AI Gateway. The Gateway is what actually calls the model provider, so the agent needs a credential proving it's allowed to use it. Studio obtains a VERCEL_OIDC_TOKEN — a short-lived token tied to a linked Vercel project — and writes it to the agent's .env.local.

Until that credential exists, the model can't run. In Chat, an agent with no credential shows a Connect to Vercel banner instead of replying. Clicking it starts the flow below.

The flow

  1. Sign in (if you aren't already). Studio asks for your email, then sends a verification link to your inbox. Open the link; the app polls until it's confirmed and continues automatically — nothing to paste back.
  2. Pick a team. If your Vercel account belongs to more than one team, Studio asks which one to use. With a single team it skips this step.
  3. Link the project and pull the token. Studio links the agent's folder to a Vercel project (vercel link) and pulls the environment, which includes the AI Gateway token (vercel env pull → VERCEL_OIDC_TOKEN in .env.local).

A console shows a live "working…" spinner and an elapsed timer through the linking and pull. When it's done, the Connect banner disappears and the agent can respond.

The app runs the Vercel CLI via npx vercel@latest, so there's nothing to install globally — the same way it provisions the Node runtime and Eve.

Alternative: an API key

VERCEL_OIDC_TOKEN is scoped to a linked project and refreshes as needed, which is why Studio prefers it. If you'd rather use a standalone credential — for example an agent you don't want tied to a specific project — you can instead set AI_GATEWAY_API_KEY in the agent's environment. Either credential satisfies the Gateway; you don't need both. You manage environment variables from the Deploy tab's Environment pane.

Next

  • Chat — talk to the agent now that the model can run.
  • Deploy to Vercel — ship the linked project to production.
  • The workspace — where the Connect banner and Environment pane live.