Your AI built a working calculator. It runs in the chat preview and nowhere else. Here is how to get that interactive tool onto a real shareable link without touching a deploy command.
You asked an AI to build a mortgage calculator. Or a pricing estimator, a tip splitter, a unit converter, a ROI widget for a sales deck. It did. The numbers update as you drag the slider, the result animates, the styling is clean. It works.
It works in exactly one place: the preview pane next to the chat.
That is the gap nobody warns you about. The tool is real, but it lives inside the AI's window. You cannot text the preview to a client. You cannot drop it into an email. The moment you close the tab, the working tool is gone, and the only thing you can still hand over is a screenshot of something that used to be interactive — a dead picture of a live thing.
This post is about closing that gap with no code. Not "learn just enough deployment." No deployment at all.
The chat preview is a sandbox. It exists to show you the AI's work mid-conversation — it was never meant to be an address you give to other people. There is no URL you can hand out, and even where a tool offers a share button, it usually points back into that tool's own environment, behind its branding, sometimes behind its login. Your client lands somewhere that is about the builder, not about your calculator.
So the instinct is to assume the next step is technical. Export the project, install something, run a deploy command, connect a repo. For a genuine web app, that is the right path. For a calculator, it is wildly more machinery than the job needs.
A calculator is a particular kind of artifact: everything it does happens in the visitor's browser. The inputs, the math, the formatting — all of it is JavaScript running locally when someone opens the page. There is nothing to install on a server, because the server does nothing but hand over the file. That one property is what makes the no-code path possible. A document would just sit there; a calculator runs — and it runs on the viewer's machine, not yours.
The single decision that determines whether this is easy: get the tool as one self-contained HTML file.
Self-contained means the HTML, the CSS, and the JavaScript all live inside a single .html document, with no external dependencies it has to fetch at runtime. No build step. No npm install. No folder of files that reference each other and break the moment one goes missing.
Most AI builders can produce this if you ask plainly:
Put everything — HTML, CSS, and JavaScript — in one self-contained
.htmlfile. No external CDN links, no frameworks that need a build step, no separate files.
This phrasing travels across Claude artifacts, ChatGPT canvas, v0, Bolt, Cursor, and Lovable. If the tool already split things across several files or pulled in a framework, tell it to inline everything into one file before you export. The cleaner the export, the more boring the rest of this becomes — and boring is the goal.
Then download or copy that file. One file. That is your tool, fully alive, sitting on your desktop.
Here is the part that replaces deployment.
A private-link host takes that single HTML file and gives you back a working URL. You drag the file in; you get a link. The link opens your calculator live — sliders, math, animations, all of it — for anyone you send it to. They click. It runs in their browser. They need no account, no install, nothing.
There is no terminal in this story. No git. No deploy command that fails on the third try with an error you have to paste back into the AI to decode. The interactive tool that only existed in your chat preview now has a real address you can text, email, or paste into a deck.
Have the HTML file already? Upload it and your calculator is live at an unguessable link in seconds — no account needed for the people you send it to.
By default that link is private: the URL is unguessable and search engines do not index it. Right for an internal pricing tool or a client-only estimator. If instead you want the calculator to be a public tool — something Google and AI answer engines can find — you flip a per-document searchable setting. The default protects the tools you build for one person; the searchable setting is opt-in, per document, for the ones meant to be found. The same upload covers an interactive calculator, a quiz, or any vibe-coded tool that runs in the browser.
Not every "tool" an AI builds is static, and pretending otherwise sets you up to be confused later. So here is the boundary, stated plainly.
A private link fits when your tool computes and forgets. The visitor types numbers, sees a result, closes the tab, and nothing was saved. A mortgage calculator, a tip splitter, a conversion tool, a quote estimator that recalculates on screen — these never needed a server, so a single HTML file carries them perfectly.
A link is the wrong shape when your tool has to remember. If it saves submissions to a database, has user accounts and logins, sends an email when someone hits submit, or calls a paid API behind a secret key — that is not a static tool, it is an app. The remembering happens on a server, and a static file has no server. For that, deploy the backend to Vercel, Netlify, or Cloudflare with a real pipeline. That is genuinely their job, and a good one.
The tell is one question: if you reopen the link tomorrow on a different computer, does it need to know anything from before? If no, it is static — a link is enough. If yes, it needs a backend.
Most AI-built calculators land firmly on the static side. They are math and a face, and math and a face fit in one file.
The shift is small to describe and large in practice. Before, the tool was a thing you demoed live, sharing your screen, hoping the preview did not reload mid-sentence. After, it is a thing that exists whether or not you are in the room.
You can put the estimator link in a proposal and let the client play with their own numbers, on their own time. You can hand a converter to a coworker without a call. You can keep a permanent link to a tool you use weekly — one that does not expire — instead of rebuilding it in a fresh chat every time the old preview goes cold.
The AI did the hard part: it turned an idea into a working interactive tool in minutes. The only thing standing between that tool and the people who should use it was an address. Now it has one, and you never touched a line of code to give it one.
Ask for one HTML file. Drop it on a link. Send the link. That is the whole workflow.
v0 and Lovable build React/Next projects, not single files — so "export" means a folder or a public deploy, neither of which is a quiet private link. Here's how to get a clean build and share the static slice privately.
A self-contained HTML file means the images travel inside it, not as links. The prompt wording that makes an AI base64-encode assets on the first pass — plus the size budget, what to inline, and what to leave alone.
Most AI-built HTML looks right on screen and breaks the moment someone hits Ctrl+P. The exact prompt clauses, CSS hooks, and a ninety-second verification pass that make a page export to PDF cleanly the first time.
miinideck turns a single HTML file into an unguessable link with optional password and expiry. Default-private, never indexed.