Ship your first hosted agent on Guuey

From an empty console to a live, streaming agent on a runtime of its own: the platform onboarding walkthrough, plus the CLI path if you'd rather type.

Read as .md
The Trimly demo's Select a Service menu, prices and Book Now buttons, drawn by the agent on the canvas mid-conversation.
An agent reply that is an interface: the Trimly demo drawing a services menu mid-chat, the generative UI this walkthrough turns on by default. See it live

You bring the instructions and Guuey brings the pod and the stream. This walkthrough takes you from an empty console to a deployed agent your users can talk to, with no SDK to import and no code to write unless you want some.

An earlier version of this walkthrough, published in August 2026, used a different builder. This one follows the platform console’s onboarding wizard.

What you’ll build

A hosted agent that stays up. It lives on one isolated pod that is yours alone, pushes each reply to the browser as a live stream from that pod, and keeps its own conversation history without you wiring a database. Generative UI is on unless you turn it off, so the agent answers with a working screen where a paragraph would have been. The screens come from the default MCP server, which speaks ggui, an open protocol that is a separate project from Guuey; you can swap in any MCP server you like.

Prerequisites

A Guuey account, which starts from guuey.com/agent-mode: click Start free trial and sign in at platform.guuey.com. Your first app gets a 7-day free trial. Beyond that you need a browser. That is the whole toolchain.

Create your first agent

  1. Sign in at platform.guuey.com. With zero apps, the dashboard sends you straight to onboarding, headed “Ship your first agent”.
  2. Name your agent. It is one field, with a suggested name already filled in; edit it or keep it, then click “Create your agent”. That opens the wizard.
  3. Which agent framework? Pick a framework row. Each one lists its current models, and whatever you pick runs here.
  4. Who does it serve? Check guests or signed-in users; you can check both. The difference is whether the agent knows who it is talking to.
  5. Build the guest brief. This is the heart of your agent. Paste your website into the “Your website” field and click “Draft from your site”: Guuey reads the site and drafts the instructions, and the site is allow-listed for you from the start. Or start from one of the starter chips (a rep for your website, a landing page rep, a trip planner and more) and edit from there. The instructions textarea is yours either way; everything on this step is editable and optional.
  6. If you checked signed-in users, build the auth mode next.
  7. Where will it meet users? Pick the channels, then finish. Guuey creates the app and deploys it in the same step, and lands you on the app’s Overview.

Watch it go live

The Overview’s status cell speaks the deploy’s own words: queued, then building, then deploying, then live. When it reads live, click “Test in playground”. The playground is wired to the real endpoint, the same one your users will reach, so what you see there is what they get.

Do not paste the widget’s script tag into your site before the status reads live and you have talked to the agent in the playground once. The loader stays dark for domains you have not allow-listed, by design, and a brief you have not tested is what your first visitor would meet.

If your workspace’s trial slot is already taken by another app, a plan card comes before the Overview; the same welcome fires once the plan is active.

Prefer code?

The wizard is the no-code door. If you would rather type, there is a second one: a code-mode agent scaffolded on whichever of the 3 framework runners you already know (the Claude Agent SDK, Google’s ADK, or the OpenAI Agents SDK), tested on your laptop, then shipped by the CLI:

npx @guuey/create-agentic-app@latest my-agent
cd my-agent
pnpm install
pnpm dev      # run locally with hot reload

npx guuey login
npx guuey deploy  # hosted on Guuey

The deploy prints the app it created, writes the App ID back into guuey.json, and walks through deploying and building in the terminal. After it, guuey test "hello" sends a test message and prints the response, and guuey logs --follow live-tails your runtime logs. The full CLI walk, including secrets and local serving, is in the CLI docs.

What just happened

When the status read live, Guuey started your agent on a pod that belongs to it alone, inside its own Kubernetes namespace, which is a harder wall than a shared container ever offers. That pod does not sleep between conversations, so the first question of the morning gets the same speed as the fiftieth. Your test chat travelled straight from the pod to your browser with no relay between them, and every turn was written down for you. The agent itself is still just the one file you shaped in the wizard, and guuey dev, guuey deploy, and the hosted runtime all read it the same way.

Where to go next

Ship the prompt. The interface figures out the buttons.