WZ Reader

Open PDF, Korean HWP, Word, PowerPoint, Excel, e-mail, images and Markdown — present, read aloud, annotate, OCR and convert. 100% in your browser. No upload.

Introduce : https://whyzoo.com/WzPDF/ · Demo : https://whyzoo-lab.github.io/WZ-PDF/ · Download : Releases

Sample documents made for these screenshots.

A presentation-focused viewer, editor, and distribution platform for the documents that actually land in your inbox.

Called WZ PDF until 1.25.0 — it outgrew the name. Installed copies update in place, and the program file is still WZ PDF.exe.

WZ Reader is a fast, fully client-side document tool. Open a PDF, a Korean .hwp / .hwpx, a Word document, a PowerPoint deck, a spreadsheet, a saved e-mail (.eml), a Markdown file or an image — read it, find in it, present it fullscreen, have it read aloud, mark it up, run OCR, and save it as a PDF with real, selectable text. All in the browser or as a Windows desktop app. No backend, no upload: your file never leaves your machine.

Built with React, Konva, pdfjs, pdf-lib and Electron, plus a Rust→WASM HWP engine (@rhwp/core), on-device OCR and on-device speech. Ships with a Model Context Protocol server so AI agents (e.g. Claude) can read every format it opens and convert documents to PDF.

---

Why WZ Reader

• 🔒 Private by design — everything runs in the browser/desktop. Nothing is uploaded.
• 📄 PDF and HWP/HWPX — open Korean word-processor files exactly like a PDF, with native selectable text (no OCR needed).
• 📊 Word, PowerPoint and Excel without Office — .docx, .pptx, .xlsx / .xlsm / .xls / .ods / .csv / .tsv / .psv open with their layout, tables, charts and pictures, with zoom, single / two-page / all-pages layouts, a page list, find, print and fullscreen. Every row of a 200,000-row sheet scrolls — nothing is cut short with "open it in Excel". Save any of them as a PDF with selectable text.
• 🎤 Speaker notes, on screen and read aloud — a deck's presenter script shows under each slide; in the slideshow it runs along the bottom as captions (C or the corner button), and read-aloud reads the script instead of the slide, turning the slides as it goes. Shift+F5 starts the show from the slide in view.
• 🎬 A deck with notes becomes a narrated video — one MP4: the slides, the script read aloud in a natural voice, and the script as a subtitle track that any player can switch on and off.
• 🔊 Reads documents aloud, on your machine — press the speaker and the document is read to you, Korean included, with ten voices and a speed control; the sentence being read is highlighted, and in fullscreen the pages turn with it. It runs entirely offline once set up: nothing about what you are reading leaves the computer. (Desktop app; the voice model is a one-time 383 MB download you are asked about first.)
• 🔁 HWP → PDF that you can actually copy from — the converted file carries a real text layer, so text stays selectable and searchable in any reader. Not a picture of a document.
• 🔐 Password-protected PDFs — opens them (asking for the password), and saves them locked (AES-256) or unlocked: the padlock on the left rail decides what the next save does.
• ✍️ Stamps that remember — upload your seal once: its white paper is removed, its size is kept, and it stays in My stamps. Stamp page after page with one click each, or copy it and paste it onto every selected page at once.
• ↩️ Undo, and nothing lost by accident — Ctrl+Z / Ctrl+Y take back stamps, signatures and page edits; opening another document or closing the window with unsaved changes asks first.

Download (Windows)

From Releases:

File
WZ_Reader_Setup_<version>.exe	Installer (recommended) — shortcuts, file associations, the console converters, automatic updates.
WZ_Reader_<version>.exe	Portable single file — runs without installing, but unpacks itself on every start (~5 s) and does not update itself.

Embed the viewer in your site

Show a document inline on any page — no download, no plugin:

<iframe
  src="https://whyzoo-lab.github.io/WZ-PDF/app.html?embed=1&url=DOCUMENT_URL"
  style="width:100%; height:80vh; border:0;"
  title="Document preview"
  allowfullscreen></iframe>

• url — the document to display (URL-encode it): PDF, HWP, Word, PowerPoint, Excel, Markdown, mail or an image. embed=1 hides the editing chrome for a clean read-only viewer; drop it to open the full app.
• CORS: the web build fetches the file in the browser, so host the document on the same origin (or a CORS-enabled URL). The desktop app is not affected by CORS.
• Try it live on the demo page, which embeds this viewer with a sample document and shows a copy-ready snippet for your deployment.

Private mode — only the documents your server designates

embed=1 is a convenience: whoever sees the page can edit the address and open anything. For a viewer that shows only what you chose, view only, host the web build yourself and put a private.json next to app.html. Every build carries an example beside app.html — private.example.json — that works as soon as it is copied to that name. A live example runs on the demo page.

{
  "documents": {
    "rfp":    "docs/RFP-2026.pdf",
    "notice": { "url": "/files/get?id=17", "name": "notice.hwp" }
  },
  "print": false
}

<iframe src="https://your-site/viewer/app.html?doc=rfp" style="width:100%; height:80vh; border:0;" allowfullscreen></iframe>

• The viewer then opens only those documents (?doc= picks one; with a single document it may be left out). A file picked or dropped, ?url=, a mail attachment — all refused.
• Zoom, fit, rotation, page list, two-page and grid views, fullscreen, find and OCR work. Editing, saving and opening other documents do not — their keyboard shortcuts included — and printing is off unless "print": true (printing is also "save as PDF").
• Paths are resolved against private.json, so relative ones stay on your server. name supplies the file name (with its extension) when the URL has none.
• It is a property of the deployment, not of the link: there is no URL switch to remove. A private.json that exists but cannot be read keeps the viewer closed, never open.
• Redeploying must leave private.json in place: deploy.example.bat keeps it when it clears the server, and never uploads a local one. If you deploy another way, exclude it the same way.
• It is not access control. The browser downloads the document to show it, so who may see it at all is your server's decision (sign-in); private mode keeps the viewer from being a way to open, change or save anything else.

---

Features

View & present

• View modes — single page, two-page spread (a wide page among narrow ones gets a row of its own), a grid overview that fills the window, and a presentation fullscreen mode (USB-clicker friendly, touchpad swipe, click-to-advance, Home/End jumps). F5 presents from the start, Shift+F5 from the page in view.
• Presenter tools (fullscreen, ZoomIt-style) — pen (P), highlighter (H), rectangle (R), arrow (A), laser pointer (L), and spotlight zoom (Z); color 1–5, width [ ], undo Ctrl+Z, erase E. Two-step ESC.
• On-screen markup (any mode) — yellow highlighter (1) and red rectangle (2) for quick emphasis; never written to the file, cleared with ESC.
• Sharp rendering — pages rasterize to match zoom × display density, so small slides and HiDPI screens stay crisp; fit width in one click when you want the text bigger.

Word, PowerPoint & spreadsheets

• Word (.docx) — pages as Word lays them out, with headers, footers and bullets; a document Word never paginated is split into real pages.
• PowerPoint (.pptx) — slides with their tables, charts and pictures, checked against PowerPoint's own rendering. Speaker notes under each slide, captions in the slideshow, hidden slides skipped.
• Narrated video — a deck with notes saves as one MP4 (H.264 + AAC) with the script as a switchable subtitle track. Desktop app.
• Spreadsheets (.xlsx .xlsm .xls .ods .csv .tsv .psv) — column widths, fonts, fills, borders, number formats, merged cells, hidden sheets, pictures (a logo, a company seal), every row of a very large sheet. Korean CSV in CP949 opens correctly.
• Save as PDF — with selectable text, at each page's own size: a deck at slide size, a sheet laid out as Excel would print it.

Text, search & OCR

• Select & copy text — real selectable text layer over PDF pages, and native selectable text for HWP/HWPX (no OCR required). In Editor mode, double-click to edit text via an inline overlay.
• Search — Ctrl+F find across the document (works on OCR text too).
• OCR — recognize text in scanned/image pages, 100% on-device and offline (Korean + English, PaddleOCR PP-OCRv5). R for this page, RR for the whole document. Saving the PDF keeps the recognized text in it.
• Region OCR → clipboard — hold Ctrl and drag to highlight any area; on release it's OCR'd and the text is copied to your clipboard.
• Read aloud (S) — speaks the document from the page you are on, a sentence at a time, using Supertonic 3 on-device. Ten voices, adjustable speed, previous / next sentence (Alt+← / Alt+→), and no audio or text ever sent anywhere. The voice model downloads once (383 MB) after you agree to it; everything after that works offline. Note that the model's weights are under the OpenRAIL-M licence rather than MIT — see THIRD_PARTY_NOTICES.md.
• Follow along while it reads — the sentence being spoken is highlighted and scrolls into view in every format; in a presentation the page or slide turns with it. Scanned documents work too: run OCR first and the recognized text is what gets read.
• Screen readers — named controls, headings per page, and announcements for OCR and reading, so a scanned document can be reached by ear.

HWP / HWPX (Korean documents)

• View .hwp (binary) and .hwpx (XML) files through the same pipeline as PDF — annotations, OCR, print, and export all work unchanged.
• HWP → PDF — saving as PDF writes the pages with an invisible, selectable text layer, doubling as a converter.
• HWP ⇄ HWPX — hwp2hwpx and hwpx2hwp convert between the two formats from the command line. A measured round trip keeps the page count and every character of the text layer, and 120 documents convert in about 2 seconds.

Annotate, manage & export

• Annotate (Editor mode) — stamps, hand-drawn signatures, text edits and full-page watermarks, written into the saved PDF. Korean text via an embedded Noto Sans KR subset.
• My stamps — uploaded stamps are cleaned up (white background removed, margins trimmed), kept with the size you last gave them, and stay armed for stamp after stamp. Ctrl+C / Ctrl+V pastes one onto the page under the pointer, or onto every page selected in the page list.
• Undo / redo — Ctrl+Z, Ctrl+Y for annotations and page edits; an amber dot marks unsaved changes, and you are asked before they could be lost.
• Page management — reorder (drag), insert blank / insert from another PDF, delete, with thumbnails. Annotations follow their pages automatically. Right-click → save the selected pages as a new PDF.
• Passwords — open encrypted PDFs; lock (AES-256) or unlock on save.
• Export — annotated PDF (Ctrl+S), booklet PDF (two pages per sheet), self-contained HTML viewer, page images (ZIP), and a standalone Viewer EXE (desktop build; the first export downloads its template once).
• Print — in-app WYSIWYG print preview; every page composited with annotations, aspect ratio preserved.

App & platform

• Open from anywhere — file picker, drag-and-drop, OS file association (desktop), recent documents on the start screen, or Open from URL.
• Automatic updates — the installed app downloads a new version in the background (verified against its published hash) and installs it when you restart; it can be turned off on the start screen.
• i18n — UI and help auto-switch to Korean on Korean OS/browser locale, English everywhere else.
• Responsive — works on desktop, tablet, and phone.

Quick start (web)

npm install
npm run dev:vite      # Vite dev server only (browser)

Open http://localhost:5173. (The root is the landing/demo page; the app itself is at /app.html.)

Desktop app (Electron)

npm run dev           # compile Electron main + Vite + launch Electron

Build

npm run build         # production web build → dist/
npm run build:exe     # Windows portable + NSIS installer → release/

build:exe regenerates the app icon from public/icon.svg, copies the HWP WASM runtime, and runs two electron-builder passes (portable first, then NSIS). The OCR and HWP binary assets are gitignored and regenerated at build time (npm run setup:ocr needs Python + pyyaml; npm run setup:hwp runs automatically). See CLAUDE.md for architecture and build details.

Testing & quality

npm test              # Vitest (watch)
npm run test:run      # Vitest (single run)
npm run lint          # ESLint

MCP server (optional)

mcp/ contains a Model Context Protocol server with fifteen tools for MCP-capable AI clients: read the text of any format the app opens (doc_get_text, doc_info — PDF, HWP, Word, PowerPoint with speaker notes, spreadsheets by rows, Markdown, mail, pictures — with OCR for scanned pages and, optionally, the text inside pictures in any of them), convert any of them to PDF exactly as the app saves it (doc_to_pdf, hwp_to_pdf), and work on PDFs (info, text, search, watermark, stamp, text overlay, split, merge, delete / reorder / insert pages). The installer ships it ready to run on the app's own binary — no Node.js install needed. It runs over stdio or HTTP (Streamable HTTP), writes only .pdf files and never overwrites one unless asked. See mcp/README.md.

Deployment

• GitHub Pages / static host — the CI workflow builds and deploys the web app (and the demo landing) to Pages on every push to main. Any static host works: serve the npm run build output in dist/.
• Self-host over SSH — deploy.example.bat is a template for deploying to an nginx-served host. Copy it to deploy.bat (gitignored) and fill in your server, user, and paths.

Tagging v*.*.* triggers a GitHub Release that builds and publishes the Windows installer + portable exe, and the latest.yml that installed copies update from.

Tech stack

Area	Tooling
UI	React 19, Tailwind CSS 4, Konva / react-konva
PDF	pdfjs-dist (render/text), @cantoo/pdf-lib + fontkit (save, encryption)
HWP / HWPX	@rhwp/core (Rust → WebAssembly)
Office	docx-preview (Word), @aiden0z/pptx-renderer (PowerPoint), hucre (spreadsheets)
Mail / Markdown	own MIME parser, marked, DOMPurify
OCR	onnxruntime-web + PaddleOCR PP-OCRv5 (offline, on-device)
Speech	Supertonic 3 + onnxruntime-node (offline, on-device)
Video	WebCodecs + mediabunny (MP4)
Desktop	Electron, electron-builder, electron-updater
Build/Test	Vite, TypeScript, Vitest, ESLint
Agents	@modelcontextprotocol/sdk

License

MIT. Third-party components and their licenses (Apache-2.0, MPL-2.0, SIL OFL, MIT, etc.) are listed in THIRD_PARTY_NOTICES.md.