EveStudioDocs

Deploy to Vercel

Deploying ships your agent to production, where it runs as Vercel Functions with durable Workflow runs, and its schedules actually fire. Studio's Deploy tab drives the whole thing (build, deploy, environment, and the logs) across three sub-tabs: Deploy & Logs, Environment, and Sandbox.

Before you deploy

Deploying requires the agent to be connected to Vercel. That's the same link that lets the model run: it associates the agent's folder with a Vercel project and pulls the AI Gateway credential. If you haven't done it yet, the header button reads Link & deploy and opens the Environment sub-tab, where the same sign-in and connect banner as Chat lives; see Connect to Vercel. Without a linked project there's nothing to deploy to. Once linked, the header's metadata line reads Vercel · linked, and after the first deploy deployed 14d ago <host>.

Build and Deploy

Open an agent and select Deploy → Deploy & Logs. Two buttons drive the pipeline:

  • Build: runs eve build locally so you can catch errors before shipping.

  • Deploy: runs Eve's own deploy command in its non-interactive form:

    eve deploy --non-interactive --yes

    eve deploy installs dependencies, runs vercel deploy --prod, and pulls the project environment afterward. --yes confirms the production deployment up front; --non-interactive means it never opens a picker or a browser, which is why the project has to be linked first.

Both stream into the Command output console, with a running… badge and a Cancel button while a step runs, and a result badge (deploy ok or deploy exit N) when it finishes; the output includes the deployment URL. Below it, the Dev server logs pane shows the local eve dev server's output live, which is where you look when the local agent misbehaves.

The header's Deploy button runs the same command.

Environment

Deploy → Environment ("Environment & secrets") manages the agent's configuration:

  • Local env files: a toggle between .env and .env.local (gitignored, auto-reloaded by eve dev). Values are masked by default, so a screen share doesn't leak keys; tick Show values to reveal them, which is also what enables editing and Save.
  • Vercel: a card badged linked · <project> or not linked. When linked it offers List remote env (vercel env ls), Pull env → .env.local (vercel env pull, which is how the AI Gateway token lands locally), and Add production env var, which encrypts a single variable into the chosen target (production, preview, or development). When not linked, the sign-in and connect banner appears here, and the tab keeps polling until the link lands.
  • Secrets: keys like API tokens live here as environment variables, never in code. Studio writes some for you and pushes them to Vercel where it can: the Arcana key and workspace, Buzz credentials from the guided wizard, and variables declared by registry items you install (those are added to .env.local for you to fill in).

Sandbox

Deploy → Sandbox shows the isolated runtime the agent uses for file and command work. With no authored sandbox it reads Default sandbox: the framework default, a /workspace bash environment on the best available backend (Vercel Sandbox once deployed). Add sandbox.ts (Vercel backend) writes an explicit agent/sandbox.ts you can edit to pick a backend, seed files, or lock down the network. An authored sandbox is shown with its source.

Chat against the deployed agent

Once deployed, you don't need to leave Studio to try it. In the Chat tab, the Target toggle switches between Local (the eve dev server on your machine) and Deployed (the production agent you just shipped). Flip it to Deployed, click Detect to fill the URL, then Test; when it reads Reachable ✓. Deployed chat is ready., your messages hit the live Vercel deployment, streaming the same turns, tool calls, and approvals as local chat. Session controls (Cancel, Compact, Clear, Reset) work against the deployed agent too.

Protected deployments

If your Vercel project has Deployment Protection enabled, the production URL is gated and Studio can't reach it by default; Test reports "Blocked by Vercel Deployment Protection". In Vercel → Project → Settings → Deployment Protection, generate a Protection Bypass for Automation secret and paste it into the Deployed target bar in Chat, then Save. Studio sends it as the x-vercel-protection-bypass header on every request, alongside the OIDC token from .env.local for the agent's own route auth. The Buzz bridge uses the same secret.

Where to go next