Last month, I watched someone at a café upload their national ID(Aadhaar) to an online PDF tool.
They needed to crop it, compress it, and send it somewhere. So they did what everyone does — Googled "free PDF tool," uploaded the file, downloaded the result, closed the tab.
Their government ID — name, address, date of birth, ID number — had just passed through a server they knew nothing about. Stored for how long? Accessed by whom? Under which country's data laws? They had no idea. Neither did the tool.
That moment stuck with me. Because I realised: this isn't rare. This is what hundreds of millions of people do every single day.
The documents we trust to "free" online tools
Think about what you've uploaded to online tools in the last year:
- Bank statements (to convert to Excel or CSV)
- Salary slips and tax documents (to compress or convert)
- Passport and government ID scans (to crop, resize, or convert)
- Product photos for e-commerce (to remove backgrounds)
- Client contracts and agreements (to merge or sign)
- Medical reports and prescriptions (to share with doctors)
Every one of those files, on most popular tools, travels to a server you don't control. iLovePDF, Smallpdf, and nearly every free tool you've used works this way — your file goes to their cloud, gets processed, and is "deleted after a short time." Their words, not a guarantee.
For casual documents, maybe that's fine. For everything listed above? It shouldn't be acceptable.
What 2025 browsers can actually do
Here's what changed: modern browsers are no longer just document viewers. With WebAssembly, ONNX Runtime Web, and GPU acceleration, a browser tab in 2025 can run the same AI models and processing pipelines that cloud servers ran three years ago — locally, on your device, with no network request.
That means background removal, PDF processing, document conversion, image editing — all of it can happen entirely inside your browser. Your file never leaves your device. There's nothing to breach on my end because I never receive your file in the first place.
This is architecturally different from tools that promise to "delete files after 1 hour." I can't delete what I never had.
What I'm building
pdfandimagetools.com is a browser-native productivity platform. Every tool processes files locally. Nothing is uploaded. No account required. No limits. No watermarks.
PDF tools — merge, split, compress, and convert documents. Convert bank statements to Excel or CSV. Extract text from scanned documents. All without your statement ever touching a server.
Image tools — remove backgrounds using BiRefNet (a state-of-the-art MIT-licensed AI model running entirely in your browser). Convert between image formats. Compress photos. Replace backgrounds with solid colours, gradients, or your own images.
Identity document tools — resize and crop passport photos to exact specifications. Convert ID scans to the right format and file size for government portals and visa applications. Your passport, your national ID, your documents — processed locally and gone the moment you close the tab.
The honest tradeoff
Browser-side AI has one real cost: the first load. BiRefNet's lite model is around 150MB — a real download on first use.
I handle this with progressive loading. The page and UI render instantly. A progress bar shows the model loading in the background. After that first visit, the model saves directly to your own device — not to any server — so every return visit loads in under two seconds. It works like any app on your phone that remembers its data between sessions. Yours, on your device, under your control.
For PDF tools, there's no meaningful tradeoff. Browser-based PDF processing is often faster than cloud tools because there's no upload and download round trip.
The stack
Next.js 15, TypeScript, Tailwind CSS, Hugging Face Transformers.js, ONNX Runtime Web, pdf-lib, and WebAssembly. Inference runs in Web Workers so the UI never freezes. WebGPU acceleration is used where the browser supports it.
Why this matters
Every tool that currently uploads your files to someone else's server should have a free, browser-native alternative that doesn't. That's the project.
If you've ever uploaded a sensitive document to one of these tools and had that brief moment of — wait, where did that actually go? — this is being built for you.
The site is launching soon at pdfandimagetools.com — get notified and be the first to know when it's live.
And if you're interested in following the build — the technical decisions, the mistakes, the progress — follow me here on Dev.to. I'll be writing about the architecture, the challenges, and everything I learn along the way.
If privacy-first tooling is something you care about, I'd love your thoughts in the comments below.
Building in public. No VC funding. No data collection. No compromises on privacy.