Folio is a free online PDF studio: 41 tools to merge, split, compress, convert to and from Word, Excel, PowerPoint and images, OCR, sign, protect and redact, plus a full page editor that can change the text already in a PDF. There is no account and no sign-up. Files are processed on Folio's own server and deleted after 6 hours (24 hours for documents open in the editor).
The problem
Everyday PDF jobs are small: join two files, shrink one to fit an upload limit, turn a scan into searchable text, sign a form. I wanted one place that does all of them without an account or a payment step, and an editor that changes the words already on the page instead of only drawing new ones on top.
Decisions that shaped it
One tool is one Python function. Every tool is
run(inputs, options, out_dir)that returns the output files, registered with its title, category, accepted file types and an option schema. The React front end draws each tool's options from that schema, so adding a tool means one function and one registry entry, with no new page to build. Workflows reuse the same registry: the files one step produces are the inputs of the next, so "make searchable, then compress" is just two entries in a list.Use proven engines, and write the glue. PyMuPDF does parsing, rendering, editing and redaction. pikepdf (qpdf underneath) handles encryption and repair, pdf2docx keeps tables and layout when going to Word, pymupdf4llm keeps headings, lists and links when going to Markdown, and Tesseract does OCR with English, Hindi and Tamil installed. On the server, LibreOffice converts Office files and a headless Chromium prints web pages to PDF.
Edit the real text, not a layer over it. The server renders each page as an image and sends back every text span with its box, font, size and colour. When you click a span and type, the old text is removed with a redaction that leaves images alone, and the new text is written into the same box. Fonts embedded in PDFs are usually subsets that lack the letters you are adding, so Folio looks up the same family among the installed fonts, with metric-compatible free stand-ins on the server (Liberation Sans for Arial, Carlito for Calibri). Every save is a new version of the document, so undo just goes back one.
Text inside pictures gets the same treatment. Scans and screenshots have no text layer, so OCR reads each line, then Folio works out the font, weight and size by rendering installed fonts against the image and comparing. An edit redraws only from the first changed letter onward, with matching anti-aliasing, so the change blends in.
Redaction removes, it does not cover. Matching text is taken out of the file itself, with presets for email addresses, phone numbers and card numbers, and the metadata can be wiped too. A black box drawn over text that is still underneath is not a redaction.
The website says only what the server can do. The same code also runs as a local app on a PC. The website leaves out what its server can't serve: Translate PDF needs a translation model the server doesn't have, so the site lists 41 tools instead of 42, and the pages that manage a computer's own folders, AI models and saved details are switched off there and kept out of search. The privacy line, the tool count and the hours files are kept all come from one file, and the build checks the tool count against the real tool list.
Test every tool. The backend has 161 test functions covering conversions, page organising and security, editor operations, OCR quality, screenshot text editing and batch speed.
Built with
Python, FastAPI, PyMuPDF, pikepdf, Tesseract, LibreOffice, React, TypeScript, Vite, Docker
Originally published at folio.niranjand.in.
