Deploy to Vercel

Everything you've built runs the same way locally and on Vercel, which makes this last step gratifyingly boring. There's no special production rewrite of the agent, no "now port it to the cloud." You build, you deploy, you poke the live URL and watch the same dispatcher answer from a server instead of your laptop.

The one production-specific choice is where the sandbox runs, and even that's mostly handled for you. Let's ship it.

Outcome

The Spoke & Mirror dispatcher runs on Vercel, and you've confirmed it live by hitting its deployed routes.

Hands-on exercise

Pick the sandbox backend. Your agent has a sandbox (it seeds torque-specs.md into /workspace). Locally that runs on your machine; on Vercel it runs on hosted Vercel Sandbox. One definition covers both, defaultBackend() resolves to the right one per environment:

agent/sandbox/sandbox.ts

import { defineSandbox, defaultBackend } from "eve/sandbox";
 
export default defineSandbox({
  backend: defaultBackend(),
});

Pinning a backend explicitly

defaultBackend() is the portable choice: it picks the best available environment (Vercel Sandbox when hosted on Vercel, then Docker, microsandbox, or just-bash). To pin Vercel unconditionally instead, import its factory from the nested path and use it directly:

import { vercel } from "eve/sandbox/vercel";
// backend: vercel({ runtime: "node24" })

eve ships a pinned factory per backend; check node_modules/eve/docs/sandbox.mdx for the set in your installed version.

Build and deploy. npx eve build compiles the agent and writes the Vercel output; vercel deploy ships it. From the project root:

npx eve build
vercel deploy

The deployed app serves the exact same health, session, and stream routes you've been hitting since 1.3, plus the web dashboard from 4.1. If you wired up Slack in 4.2, this is the deploy that gives its webhook a real home.

Try It

Smoke-test the live agent. Health first, it's public, no auth:

curl https://<your-app>/eve/v1/health

{ "ok": true }

Then drive a real turn as a customer (the deployed routes enforce the fail-closed auth from 5.1, so send a session):

curl -X POST https://<your-app>/eve/v1/session \
  -H 'content-type: application/json' -H 'cookie: shop_session=demo-pro' \
  -d '{"message":"what does a full overhaul cost?"}'

Or, the comfortable way, point the dev TUI at the deployment and just talk to it:

npx eve dev https://<your-app>

Same dispatcher, same tools, same approval gate, now answering from production. Ask it to book the overhaul and the approval still parks the turn, exactly as it did on your laptop.

Find your runs in the dashboard

Once deployed, Vercel auto-detects eve as the framework and surfaces an Agent Runs tab under your project's Observability view. Each conversation is a trace you can open and walk, every tool call, every approval, every turn. (The tab is gated per team in the current release; if you don't see it, ask your Vercel contact to enable it.)

A 401 on every live request? That's the fail-closed lock doing its job, send a valid shop_session cookie, or for a quick poke, hit /eve/v1/health, which is always public. If eve build fails on discovery, read the diagnostics it prints (and .eve/diagnostics.json); the most common cause is a tool file that doesn't export default its defineTool.

Done-When

agent/sandbox/sandbox.ts sets a backend (defaultBackend() is fine).
eve build succeeds and vercel deploy ships the app.
curl https://<your-app>/eve/v1/health returns ok.
A real turn against the deployed URL works (TUI via eve dev <url> or curl with a session).

Solution

The only code is the sandbox backend; the rest is two commands:

agent/sandbox/sandbox.ts

import { defineSandbox, defaultBackend } from "eve/sandbox";
 
export default defineSandbox({
  backend: defaultBackend(),
});

npx eve build && vercel deploy

The dispatcher you started as a persona in 1.1 now diagnoses, quotes from a real catalog, remembers bikes, adapts per tier, asks before it spends, answers on the web and in Slack, and runs in production behind real auth. One last short stop: a look at the three things the shop didn't need, and when you would.