Welcome to Pickrack: Why I Built 46 Free Tools That Don't Want Your Email

Every week I run into the same loop. I need a small tool — merge two PDFs, strip metadata from an image, decode a JWT — and the search results all look identical: "free", "online", "easy". I click. The site loads, I drop the file in, I click the button, and a popup tells me I need to sign up to download the result. Or there's a watermark. Or a daily limit of two files.

That entire pattern is a tax on people who just need to get something done.

Pickrack is my answer to that tax. It's a single site with 51 tools across six categories — PDF, image, AI, dev, text, and calc — that I personally use, and now you can too. No signup. No daily quota. No watermark. Most tools run entirely in your browser, so your files don't even leave your device.

This post is a longer "about" — what's on the site, why each design choice was made, and what's coming next.

The 51 tools, briefly

PDF (16 tools)

The category that got me started. PDF is everywhere and everyone needs to do something to one occasionally. I built merge, split, rotate, watermark, JPG↔PDF, compress, unlock, password-protect, Word↔PDF, PPTX↔PDF, EPUB↔PDF, and PDF→Markdown.

Six of these are 100% browser-side using pdf-lib: merge, split, rotate, watermark, JPG→PDF, PDF→JPG. Your file is read into browser memory, processed in JavaScript, downloaded back. The server never sees it.

The other ten need a server because they shell out to native binaries that have no browser equivalent: qpdf for unlock and AES-256 protection, Ghostscript for compression, LibreOffice headless for Word/PPTX, Calibre for EPUB, and a pdftoppm + pptxgenjs pipeline for PDF→PPTX (image-per-slide, the same approach Smallpdf and iLovePDF use). Server-side files are written to a temp directory and deleted immediately after the response. No DB, no logging, no storage.

Image (5 tools)

Image resizer, compressor, format converter (PNG↔JPG↔WebP↔AVIF), HEIC→JPG, and AI background remover. All five run in your browser. The background remover uses @imgly/background-removal, an ONNX runtime WebAssembly model — the first conversion downloads about 30MB of model weights, then everything is cached and instant. HEIC conversion uses heic2any. Resizer/compressor/converter use the standard Canvas API.

I keep recommending these to friends with iPhones who panic when their HEIC photos won't open on Windows.

AI (3 tools)

Summarizer, translator (20 languages), and grammar checker. These three need a server because the models are too large for the browser. They call Anthropic's Claude Haiku 4.5 API on the backend. Only your text is sent — never a file, never any identifying data. Per-IP rate limit is 10 requests per 24 hours, generous enough for occasional use, tight enough to keep my Anthropic bill predictable.

If you self-host Pickrack from the GitHub repo, set your own ANTHROPIC_API_KEY and the rate limit can be adjusted or removed.

Dev (6 tools)

The category I use most myself: JSON formatter, Base64 encoder/decoder, JWT decoder, hash generator (MD5/SHA-1/SHA-256/SHA-512), UUID generator (1-1000 bulk), and regex tester. All six are browser-only — Web Crypto API for hashes, native Intl for things like collation, a tiny bit of regex testing logic.

These tools were originally for me. I'd open a separate site every time I needed to decode a JWT. Now they all live in one place, on a domain I trust, with no ads above the fold for AI tools and dev tools (the user-base is technical and ad-blockers are universal).

Text (4 tools)

Word counter, case converter (9 variants), lorem ipsum generator, slug generator. The slug generator handles Vietnamese diacritics correctly (NFD normalize plus the special-case đ → d rule, which most generic slugifiers miss). Standard JavaScript, browser-only.

Calc (3 tools)

Password generator, QR code generator (URL/text/Wi-Fi/vCard/email, PNG and SVG output), and age calculator. Browser-only. The QR generator uses qrcode.js for the encoding. The password generator uses crypto.getRandomValues from the Web Crypto API — never Math.random, which is not cryptographically secure.

Why "free" with no quota?

The pattern of free-with-quota is a sales funnel, not a free tool. The real product is the upsell. Once you accept that frame, every "free" tool site becomes adversarial: how can they nag you, watermark you, throttle you, until you cave?

I don't have an upsell. There is no Pickrack Pro. The site is funded by:

1. Display ads on text-content pages (blog posts, hub pages). Not on the AI tool pages, where ads would degrade the experience.
2. Affiliate links on review posts, when I genuinely use a paid tool myself. Always disclosed at the top of any post that has them.
3. Self-host is encouraged. The whole site is MIT-licensed open source. If you want to run it yourself, the README tells you how.

That's the entire business model. There's no Series A pressure, no investor demanding 10× growth, no roadmap to convert free users into paying ones. The site exists because tools should be free.

What "privacy-first" actually means here

The phrase is overused, so it's worth being concrete. Pickrack's privacy claims are:

• 31 of 51 tools run entirely in your browser. Your file stays on your device, full stop.
• 10 PDF tools that need server processing (compress, unlock, protect, Word/PPTX/EPUB conversion, Markdown extract, OCR-related): files are written to /tmp, processed, response streamed back, and the temp directory deleted in a finally block. There is no database. There is no cloud storage bucket. The Linux process holding the file dies, and the file is gone.
• 3 AI tools (summarize, translate, grammar): only your text is sent, only to Anthropic's API. Anthropic's data policy: prompts to the API are not used for training and are deleted within 30 days. I don't add my own logging on top.
• The code is public. You can read lib/libreoffice.ts, lib/anthropic.ts, app/api/pdf/compress/route.ts, and verify every claim above.

That last point matters. "Privacy-first" with closed source is a marketing claim. Privacy-first with open source is a falsifiable statement. I want anyone with skepticism to be able to check.

What's coming next

A few directions I'm exploring, roughly in priority order:

1. More AI tools: chat-with-PDF (locally extracted text → Claude), AI YouTube summarizer (via official transcripts), document translator with file upload (preserves DOCX/PDF formatting). All gated by Anthropic cost, all rate-limited per IP.
2. Long-tail tools people keep asking for: image cropper, image upscaler (Real-ESRGAN WASM if I can get the model under 50MB), color palette extractor, currency converter, BMI calculator, tip calculator. These are easy wins and the SEO is favorable.
3. Search inside the tool catalog. With 51 tools today and growing, finding the right one is going to be the next UX problem.
4. Self-host quickstart: a Dockerfile and docker-compose.yml that boots Pickrack with all the system binaries (LibreOffice, Calibre, qpdf, Ghostscript, Poppler) baked in. PR welcome.
5. Localization. The site is English-first today. Vietnamese first, then Spanish or another major language depending on which countries the analytics show the most traffic from.

How to follow along

• GitHub: github.com/pickrack/pickrack — issues and discussions enabled, MIT license
• Twitter/X: @pickrackdev — release notes and behind-the-scenes
• Email: [email protected] — for anything substantive that doesn't fit a tweet

Bookmark the homepage. Star the repo if you find it useful. Open an issue if a tool has a bug or you wish a tool existed. That's how this gets better.

— Minh