The world's most aggressive browser-native PDF watermark removal engine.
No server. No rasterization. No compromises. 100% free, forever.

---

🛡️ Privacy & Stability Policy

Important

DocPurge AI processes your PDFs entirely inside your browser.
Your files are never uploaded to any server — not even ours (we don't have one).
The Gemini API key you enter is stored as a plain JavaScript variable in the current tab session only.
It is never written to localStorage, cookies, IndexedDB, or any remote endpoint.
Closing or refreshing the tab permanently clears it.

This is not a marketing claim. This is the architecture.

---

⚡ What Is This?

DocPurge AI is a zero-backend PDF utility suite that runs entirely in your browser using pdf-lib and pdf.js. It surgically removes watermarks at the byte level — no page-to-image conversion, no quality loss, no rasterization. Your vectors stay crisp. Your fonts stay embedded. Your DPI stays untouched.

Phase 1 — Auto-Detect    →  Scan all content streams, XObjects & OCG layers
Phase 2 — Native Remove  →  Delete watermark objects at raw PDF byte level
Phase 3 — Hint Mode      →  User provides text hint → targeted stream search
Phase 4 — Gemini AI      →  AI identifies exact watermark string from page text
Phase 5 — Export         →  Clean PDF assembled & downloaded — never stored

---

🗂️ The Arsenal

🔥 Watermark Purge Flagship Tool — LIVE ⚡ Zero Rasterization — Operates on raw PDF stream operators (Tj, TJ, Do, XObjects). Vectors, fonts, and DPI 100% preserved. 🔍 5-Phase Engine — Auto-detect → Native remove → Hint mode → Gemini AI → Clean export. Every scenario covered. 🧠 Gemini AI Assist — Sends only extracted page text (not your PDF) to Gemini 3.5 Flash / 3.1 Pro via your own API key. 🐘 No Size Limit — Handles 200 MB+ PDFs via direct ArrayBuffer reads. Web Workers keep the UI alive. 🔒 Session-Only Key — Your Gemini API key lives only in a JS variable. Gone on tab close. 📥 Instant Download — Output assembled in-memory and downloaded directly. Never stored.

📦 PDF Compressor Coming Soon 🗜️ Lossless + Lossy — Dual compression modes with configurable quality targets. 🖼️ Smart Image Downsampling — JPEG re-encoding with DPI-aware quality settings. 📉 Up to 80% Reduction — Proven on real-world document types. ⚡ Text Layers Untouched — Only image streams are compressed. 🔗 PDF Merger & Splitter Coming Soon 🔗 Merge N PDFs — Drag to reorder before merging. ✂️ Split to Pages — Custom page range extraction. 🖱️ Visual Page Organiser — Drag-and-drop page thumbnails.

🕵️ Metadata Scrubber Coming Soon 🧹 XMP Metadata Wipe — Author, creation date, GPS, software fingerprints — gone. 🔍 Forensic-Grade — Removes fields most tools miss (custom schema namespaces). 🔒 Zero Fingerprints — Leave no trace of origin software or author.

🤖 Gemini AI Model Lineup 2026 Production Stack Model Best For Speed gemini-3.5-flash Most watermarks · Default ⚡ Fastest gemini-3.1-pro Complex marks · 2M context 🎯 Highest accuracy gemini-3.1-flash-lite Simple hints · Budget 💰 Cheapest

---

🚀 Getting Started

Step 1 — Open the App

https://official-arvind.github.io/docpurge-ai/

No installation. No download. Works in Chrome, Edge, and Firefox on HTTPS.

Step 2 — Drop Your PDF

Drag and drop any PDF into the upload zone — or click to browse.
No size limit enforced. 200 MB, 500 MB — the engine handles it with Web Workers.

Step 3 — Watch the 5-Phase Engine Run

Auto-detection kicks in immediately. If a watermark is found, it's natively removed and your clean PDF is ready for download.

Step 4 — Stuck? Use Gemini AI Assist

1. Type a hint →  CONFIDENTIAL  or  DRAFT  in the Hint field
2. Enter your Gemini API key (free at aistudio.google.com)
3. Select model → gemini-3.5-flash (recommended)
4. Hit "Identify with Gemini" → AI returns the exact watermark string
5. Hit "Purge" → download your clean file

---

🗂️ Project Structure

/
├── index.html                     ← Branded splash + redirect
├── 404.html                       ← Custom GitHub Pages 404
├── .nojekyll                      ← Skip Jekyll — serve raw HTML
│
├── /home/
│   └── home.html                  ← Dashboard — hero terminal, tools, FAQ
│
├── /watermark-purge/
│   ├── watermark.html             ← Full tool interface
│   └── watermark-logic.js        ← 5-phase purge engine (ES Module, 41 KB)
│
├── /assets/
│   ├── /css/tailwind-output.css   ← Full custom design system (23 KB)
│   ├── /js/core-ui.js             ← Shared UI — nav, FAQ accordion, terminal sim
│   └── /img/                      ← AI-generated brand assets
│
└── /test/
    ├── test-watermark.js          ← Node.js test suite (no browser needed)
    └── package.json               ← pdf-lib dependency

---

🧪 Node.js Test Suite

Test watermark removal logic against real PDFs — no browser required:

cd test
npm install

# Auto-detect and remove from all PDFs in ./samples/
node test-watermark.js

# Target a specific watermark text
node test-watermark.js --text "CONFIDENTIAL"

# Single file
node test-watermark.js --file ./samples/my-doc.pdf --text "DRAFT"

# Dry run — no output files written
node test-watermark.js --dry-run

# Strict mode — exits non-zero if nothing removed (CI-friendly)
node test-watermark.js --strict

Output files written to test/output/ with _PURGED suffix. Styled table printed to console with per-file status, detection count, removal count, and timing.

---

🛠 Tech Stack

Layer	Technology	Notes
PDF Manipulation	pdf-lib v1.17.1	Native object-level surgery — no rasterization
Text Extraction	pdf.js v4.3	ES Module import via cdnjs
AI Engine	Gemini API (REST)	3.5 Flash · 3.1 Pro · 3.1 Flash-Lite
Hosting	GitHub Pages	Static, no server, HTTPS — ES modules work perfectly
Backend	None	🚫
File Uploads	None	🚫
Analytics / Tracking	None	🚫

---

👤 Credits

Built by Arvind · Jigar Corp
AI brand assets generated with Google Gemini
MIT License — Free forever.

---