PDF to Code

Turn a PDF or a presentation into a standalone HTML page or a drop-in React component.

---

What it takes in

Three sources, one engine. Everything becomes a PDF first, so the conversion below is the same work in all three cases — but how each one gets here differs, and the difference is a privacy one:

Source	How it arrives	Leaves your machine?
PDF	read locally	No. Never uploaded, never sent anywhere.
Google Slides	exported by Google straight to the browser tab	No. It goes from Google to your browser; this app's server is not in that path.
PowerPoint (.pptx, .ppt)	converted to PDF by LibreOffice on the server	Yes — and only this one. Off unless NEXT_PUBLIC_ENABLE_PPTX=true.

PowerPoint is the exception because browsers have no way to render it: there is no pptx.js doing what pdf.js does. Rebuilding DrawingML by hand in the browser was the alternative, and it cannot reach the fidelity this product promises. So a .pptx is sent over, converted, and deleted before the response is written — nothing is stored, queued or logged. The interface says so on screen when it happens, rather than leaving it to be assumed.

What it does

Each converted page is built from two layers:

1. A raster layer carrying vector art, images and rules — everything a text-only reconstruction would lose.
2. A real text layer of absolutely positioned markup using the document's original fonts, laid transparently over it.

The result looks identical to the PDF, and the text stays real text: selectable, searchable, translatable and editable. Turn the background off and you get clean markup instead — editable, but without the graphics.

Right-to-left scripts are handled properly. <html lang> and dir are derived from the document's own content, so a Hebrew or Arabic PDF exports as a correctly-declared right-to-left page rather than one mislabelled as English.

Running it

npm install
cp .env.example .env.local   # then fill in SESSION_SECRET
npm run dev

PDF conversion needs nothing else. The other two sources are optional and each is off until configured — see .env.example:

• PowerPoint needs NEXT_PUBLIC_ENABLE_PPTX=true and LibreOffice reachable as soffice, or SOFFICE_PATH pointing at it. It is off by default because it cannot run on serverless hosting at all — see docs/deploy-vercel.md. In production it means a container image with LibreOffice installed, given no network of its own and a memory and time limit, because it parses files this app did not write. With the flag off, .pptx is refused with a way forward and everything else works.
• Google Slides needs a Google Cloud project with the Drive and Picker APIs enabled, and the three NEXT_PUBLIC_GOOGLE_* values. The button does not render without them.
• Accounts need a Firebase project with Email/Password and Google sign-in enabled, the four NEXT_PUBLIC_FIREBASE_* values, and a service account in FIREBASE_SERVICE_ACCOUNT. Required in production; see below for how development gets by without one.

Getting in

An account is the only thing that proves anything. Set SESSION_SECRET, which signs the cookie that carries it:

node -e "console.log(crypto.randomUUID()+crypto.randomUUID())"

Firebase verifies the account once, at sign-in, in a Node route handler — firebase-admin cannot run on the Edge. What middleware checks on every request after that is an HMAC-signed cookie carrying the uid, which Web Crypto verifies in either runtime with no network call.

A shared beta code used to stand in front of this, and it is gone. Anyone who reaches the app can now create an account, so who may sign in is Firebase Authentication's decision — restrict it there, not in this codebase.

In development the sign-in screen also offers a developer button, which mints a session outright so the app runs with no Firebase project at all. Its route answers 404 whenever NODE_ENV is production, so it cannot be reached on a deployment.

In development a Sign in as developer button skips both, so a machine with no Firebase project can still run the app. It is gated on NODE_ENV inside the route handler, not merely hidden — the endpoint refuses the developer path in any production build, so it cannot be reached by calling the API directly either.

Scripts

command	what it does
npm run dev	development server
npm run build	production build (writes to .next-build, so it never collides with a running dev server)
npm test	converter unit checks over the fixture set
npm run qa	injection, scale, cancellation, pricing, language and React-render checks
npm run validate	HTML5 validation and JSX compilation of the generated output
npm run contrast	WCAG contrast audit computed from the design tokens
npm run fixtures	download the public-domain test PDFs
npm run analyze	measure text-vs-graphics composition of the fixtures
npm run bakeoff	side-by-side conversion-engine comparison harness

Layout

src/converter/   the conversion engine — pure, isomorphic, no DOM
src/lib/         intake (the three sources), pdf.js setup, job state, auth, pricing
src/lib/firebase/  accounts — the client SDK, and the admin SDK that verifies sign-ins
src/i18n/        locale registry; direction is derived, never configured
app/             Next.js App Router — login gate and dashboard
components/      shadcn/ui primitives and app chrome
scripts/         test, QA, validation and fixture tooling
messages/        one JSON file per language

src/converter/ has no dependency on React, Next or the DOM, which is why the same code runs in Node for the test suites and in the browser for the product.

Adding a language

One line in src/i18n/config.mjs and one file in messages/. Direction comes from the language subtag, so nothing else needs touching.

Notes and limits

• Scanned PDFs have no text layer. The output keeps the page images, but the text is not selectable or searchable without OCR, which this does not do.
• Pages are fixed-size. They match the original layout rather than reflowing, so narrow screens scale rather than rewrap.
• Fonts are referenced by the names embedded in the PDF with generic fallbacks. Install the real families for an exact match.
• Nothing is stored. No files, no accounts, no history — conversion activity lives in the browser tab and disappears on refresh. A PowerPoint file passes through the server to be converted and is deleted in the same request; nothing about it is written down.
• PowerPoint fidelity is LibreOffice's, since it does the rendering. Decks using fonts the server does not have will substitute them, so install the fonts you care about in the image.
• Google's export has a size limit. Drive refuses to export a deck past a fixed size; when that happens the app says so and suggests downloading it as a PDF instead.

Licence

Not yet chosen.