iLoathePDF

PDF and image tools that don't upload your files anywhere — merge, split, compress, organize and convert, running entirely on your own machine.

Try the web app → — no install, runs in your browser tab.

Why this exists

The tools that do this well (iLovePDF, SmallPDF, and friends) are websites, so every file you feed them gets uploaded to someone else's server. That's fine for a holiday photo. It's not fine for an Aadhaar card, a PAN card, a signed contract, or anything else you'd rather not hand to a stranger's backend. iLoathePDF does the same jobs with the same quality of interface, without the upload — as a Windows desktop app with zero networking code, and as a website that processes files client-side and never sends them anywhere.

Two ways to run it

	Desktop app	Web app
Install	Windows installer	None — open the link
Engine	Native Rust + Python + Ghostscript	pdf-lib / pdf.js, in your browser
File size limits	Disk-bound	~200 MB (browser memory)
Batch / whole folders	Yes	No
Networking code	None at all — cannot phone home	None — verify in DevTools → Network
Compress quality	Full (Ghostscript)	Preview (rasterizes pages, no selectable text yet)
HEIC → JPG	Yes	Preview badge (decoder not wired up yet)
PDF to Excel	Yes	No — no client-side library extracts tables well enough

Both share the same rule: your files never leave the machine they're opened on. Pick whichever fits — there's no reason to pick only one.

The thirteen tools

Tool	What it does
Merge PDF	Combine PDFs in the order you choose, with page ranges per file
Split PDF	Cut into ranges, chop every N pages, extract or delete a selection
Organize pages	Reorder, rotate and drop pages on a page canvas
Sign & Fill	Draw or upload a signature, then add text, dates and initials on the page
Protect & Unlock PDF	Add or remove a password that's required to open the file
Watermark, Page Numbers & Stamp	Add a repeating watermark, sequential page numbers, or a fixed stamp to every page
Redact PDF	Black out sensitive text, photos or signatures — visually or for good
OCR → Searchable PDF	Add an invisible, searchable text layer to a scanned PDF
PDF to Excel	Extract tables into a real, editable spreadsheet — desktop only
Compress PDF	Lossless, Balanced or Strong, with the size trade-off shown
PDF to images	Render pages to PNG or JPG at a chosen DPI
Images to PDF	Scans and photos into one PDF, one image per page
Convert images	PNG, JPG and WebP any direction — and HEIC off an iPhone

On the web app, Compress and the HEIC direction of Convert Images carry a visible Preview badge: real files go in and real processing happens, but the output isn't yet full quality — see the table above. PDF to Excel is desktop only — no client-side library extracts tables well enough to ship on the web, so it shows there as a disabled card pointing to the download. Every other tool is fully live in both the desktop app and the browser.

Recipes

Four fixed, curated multi-step chains over the tools above — drop a file, pick a recipe, get the final result, with no per-step configuration: Merge → Compress, Scan to Searchable PDF (Images to PDF → OCR → Compress), Flatten PDF (PDF to Images → Images to PDF), and Merge → Extract Tables (desktop only, since it ends in PDF to Excel). Recipes aren't a general pipeline builder and aren't counted among the thirteen tools above — they're compositions of existing tools, reusing each one's own default settings.

How it's built

Desktop:
  React UI  ──tauri invoke──▶  Rust core  ──JSON lines over stdio──▶  Python engine
     ▲                            │                                        │
     └───── job://progress ───────┘◀────── progress / result events ───────┘

Web:
  React UI  ──File API──▶  in-memory bytes  ──pdf-lib / pdf.js──▶  Blob  ──▶  download

• Desktop — React 19, TypeScript, Tailwind v4, Motion, self-hosted fonts, shipped via Tauri 2 with a custom title bar and NSIS installer. The engine is a long-lived Python process (pikepdf, Pillow, pillow-heif, img2pdf, Ghostscript) speaking newline-delimited JSON, frozen with PyInstaller.
• Web — React 19, TypeScript, Tailwind v4, React Router, deployed to GitHub Pages. Every tool page follows the same shell: drop a file, set options, run, get a result — and a live request counter proves the site makes no network calls after the page loads.

Design decisions and the reasoning behind them live in each app's own SPECS.md; the write-up of how the web app itself was designed and built is in docs/superpowers/specs and docs/superpowers/plans.

Repo layout

iLoathePDF/
├── apps/
│   ├── desktop/   the Windows app — see apps/desktop/README.md
│   └── web/       the website — see apps/web
└── docs/          design specs and implementation plans

Getting set up

npm install   # installs both apps/desktop and apps/web (npm workspaces)

Then see each app's own README for how to run and build it:

• apps/desktop/README.md — requires Node 22, Python 3.11, Rust with the MSVC toolchain, and Ghostscript.

• apps/web — npm run dev --workspace=apps/web and open the printed local URL. The OCR tool needs its tesseract.js worker, core (WASM) and English language files vendored locally under apps/web/public/tesseract/ (gitignored, not committed — the app never fetches them from a CDN at runtime, matching its no-network guarantee). After npm install, copy them in:

  mkdir -p apps/web/public/tesseract/core apps/web/public/tesseract/lang
  cp node_modules/tesseract.js/dist/worker.min.js apps/web/public/tesseract/
  cp node_modules/tesseract.js-core/tesseract-core-simd-lstm.{js,wasm,wasm.js} node_modules/tesseract.js-core/tesseract-core-relaxedsimd-lstm.{js,wasm,wasm.js} node_modules/tesseract.js-core/tesseract-core-lstm.{js,wasm,wasm.js} apps/web/public/tesseract/core/
  gzip -k -c vendor/tesseract/tessdata/eng.traineddata > apps/web/public/tesseract/lang/eng.traineddata.gz
  

  All three LSTM builds are needed: corePath is a directory, and tesseract.js feature-detects which one to load at runtime (only the -lstm variants, since this tool always uses OEM LSTM_ONLY) — relaxed-SIMD for a browser that supports it (current Chrome), plain SIMD for one that doesn't, and the plain tesseract-core-lstm.* build (no SIMD at all) for a browser with no WASM SIMD support whatsoever (older Safari/Firefox, or SIMD disabled by policy). Missing any one of the three 404s for exactly the browsers that would have picked it. Keep tesseract.js-core's explicit version in apps/web/package.json in sync with whatever tesseract.js itself depends on (check node_modules/tesseract.js/package.json's own tesseract.js-core range) so this isn't vendoring a different, possibly incomplete build than the one npm actually resolves.

  (The last line reuses the same eng.traineddata the desktop app vendors — see apps/desktop/HANDOVER.md. If that file isn't present yet, source eng.traineddata from the Tesseract project's own tessdata release instead and gzip it the same way.)

Contributing

main is built through pull requests, and CI has to pass:

git switch -c my-change
git push -u origin my-change
gh pr create --fill

Direct pushes to main are blocked by a local hook rather than GitHub's branch protection (which needs a paid plan or a public repo to configure — this repo is public, but the hook is kept as a belt-and-suspenders):

git config core.hooksPath .githooks

License

Code in this repository is MIT licensed.

That covers the code iLoathePDF's own authors wrote — it does not relicense the third-party pieces the desktop app bundles into its installer, which keep their own terms regardless of this repo's license:

• Ghostscript — AGPL-3.0. Distributing a build that includes it carries AGPL obligations. See apps/desktop/SPECS.md before redistributing a compiled installer.
• pikepdf — MPL-2.0
• img2pdf — LGPL-3.0

The web app's dependencies (pdf-lib, pdf.js, React, etc.) are all permissively licensed (MIT/Apache-2.0) and impose no extra obligations.