Hello world! We're Guuey.

Give your service an agent mode with generative UI, an agent on your landing page in one script tag and agent mode in your app with the SDK. Your customers ask for what they need, and your agent responds with an interface tailored to the task. If you are a founder whose landing page is already up and already collecting the same questions every day, the first thing to hire from Guuey is a rep for that page, and this post is about that rep.

Guuey is in public beta preview. You are seeing the sneak peek, and it stays a sneak peek while we keep building in the open. We will call it “beta” when it has earned the word. Everything below is deployed and working right now, down to the open-source packages behind the tooling. Today is the day we launch that preview. Anyone can sign up, and the rep is the 10-minute way to find out what the platform does.

Here, we’ll share short walkthroughs of building agents, connecting them to your service, and what your customers can do with them. This post is the rep on your landing page. The next post is Agent Mode, the same agent working inside your product, and the one after that is a tour of the Trimly demo.

Just ask

Customers skip the maze of dashboards and just ask, and a bespoke interface comes back for every request.

Here is the belief underneath that sentence. Every screen in a SaaS product is a guess, made months in advance, about what somebody will need today. The guesses pile up into a map the customer has to memorize, and the map was never the product. When a person can simply say what they want, the interface should assemble itself around the request: a slot picker when they are booking, a table when they are comparing, a form when something is missing. To us generative UI is not a feature on a checklist but the position that an interface should be the shape of the answer, not the furniture of the product.

That belief reaches your landing page as the widget, one script tag: paste it once and a Guuey rep is answering on your page, a chat surface where answers arrive as working interfaces rather than paragraphs. It is sandboxed inside an iframe served from Guuey’s own origin, blind to your page’s cookies, storage, and scripts, and dark until you allowlist your domain; the loader URL is a permanent contract, versioned behind, never in front. The other door, agent mode in your app with the SDK, is the next post’s subject; watch its walkthrough if you would rather not wait.

The video at the top of this post is the widget door, shown as an animated walkthrough. A landing page, a mock of our own drawn for the film, has just been published, and the film asks the question every landing page earns: “Where’s the agent?” The answer is an agent drafted from the site itself, one script tag pasted into the page, and a visitor who asks what it costs and gets back a card drawn for that question, with a button to start the trial. Nobody coded the screens it shows. The agent described an intent, and the interface figured out the buttons. The rest of this post is that walk in words: the interview, the install, the allowlist, and what a landing page looks like when it answers back.

Hire a rep

Every landing page gets the same questions, in the chat bubble or in the inbox, and most of them arrive after hours and go unanswered. “So what do you actually do?” “Does it work with our stack?” A Guuey rep takes those on your page itself, in your voice, and when a surface beats a sentence it draws one: a pricing view, a comparison, a hand-off into signup. An interface that fails validation never renders, so a button it draws does what it says.

Hiring one is three moves, and there is no code until the last. First, brief it the way you would brief a new hire: what the product does, who buys it, and the voice you use. Anyone who has written a job description can write this. Second, drop in what a new hire would read in week one, your docs, your pricing, your FAQ; the onboarding is drag and drop. Third, copy one script tag into your page. The rep appears in the corner, on your domain, and starts taking questions. The interview happens on platform.guuey.com, and 10 minutes is the whole install, with no rebuild, no SDK, and no redeploy of your site.

It answers from the brief and the materials you gave it, and from nothing else. It does not invent features, and when the brief runs out it says it does not know and offers a human rather than a guess. Visitors talk to it without signing up, because guest mode is on by default, and every conversation lands in your console, where a week of real questions tends to teach a founder more than a month of pageviews. A rep gets a runtime of its own on Guuey, with nobody else’s traffic beside it, and your materials stay yours, put to work answering your visitors.

One script tag is where the rep starts, not where it stops. It runs on the full platform from its first day, so it can grow into whatever the platform can do: recognize the people who sign in, pick up tools from MCP servers (calendars, payments, your own API), run on our model keys or on yours, and, when you are ready, step inside your product as Agent Mode. Guuey’s own landing page runs one: the bubble in its corner is a rep, and asking it about pricing or your stack is the shortest product tour we have.

Under the hood

Hosting is the second half of Guuey, and the rep sits on it. Each deployed agent gets a runtime of its own: an isolated pod, awake from deploy to undeploy, so the first message of the day answers as fast as the hundredth, and every one of those pods runs under gVisor. Replies stream to the visitor over Server-Sent Events, and no relay server sits in between. History persists on its own. The whole agent is one file, guuey.json, which names the system prompt, the model, and the MCP servers it may call; the rep you briefed in the console is that file, written for you. Agents that keep the default MCP server get generative UI without writing anything; that server speaks ggui, an open protocol and a separate project from Guuey, and the two compose without requiring each other.

No-code or CLI, your call

The interview is the no-code door, and for a landing-page rep it is the whole job. If you have developers who would rather work from the terminal, the rep can live in your repository as a file: version it, review it, ship it with the rest of your code. The CLI ships 3 framework runners (claude-agent-sdk, openai-agents-sdk, and google-adk) and a scaffold, and everything it installs is open source on npm under the @guuey scope:

npx @guuey/create-agentic-app my-agent
npx guuey login
npx guuey deploy

Nothing on your page changes either way; the script tag is the same one.

The honest print

A few things we would rather print than blur. A rep knows only what you gave it, so a thin brief makes a thin rep; give it the documents you would hand a person. Durable state (the hosted key-value store plus a per-user home directory that survives restarts) is rolling out per environment, and the APIs are stable, so anything written against them keeps working as the rollout reaches you. MCP servers that need an account sign-in are CLI-only for now. And a rep on your landing page answers questions; it does not act inside your product on a signed-in user’s behalf. That is Agent Mode, and it is the next post.

We print this section because a status chip that lies is worse than no chip at all.

Come poke it

The shortest way to judge any of this is to hire one. Go to guuey.com/hire-a-rep, brief the rep, paste the tag, and read what your visitors asked by tomorrow morning. Every new app starts with a 7-day free trial, and nobody needs an invite. If you would rather build from scratch, describe the agent you want at guuey.com and it deploys as you finish.

A launch bonus is running. A credit top-up made before October 8, 2026 mints 10% extra credit, and the extra covers usage, never plan fees. The bonus shows up on your own billing surface once the top-up lands: Billing → Credits in the console, or guuey billing in the terminal.

Explore Guuey or dive into the docs. Glad you’re here. We’re just getting started.