From a tiny script to renamed-to - AI that turns scan_123.pdf into real file names

A few months ago I wrote a 50-line Python to fix a boring problem. I run a company and a household, so the PDFs never stop. Invoices, medical docs, contracts, insurance, car papers, apartment docs, school stuff for the kids. Filenames like scan_123.pdf, BRW4CEBBD1CF5C1_001492.pdf, and document.pdf made everything painful to find.

The script ran OCR, sent the text to a Mistral model, and produced a clean name like 2025-08-12_Acme_Invoice_1842.pdf. I used it, forgot it, then kept coming back every month. That was the signal. If I reopen the same script again and again, it wants to be a product. So I turned it into a web app: renamed.to.

The idea stayed simple. Upload a PDF, extract text, let the model propose a predictable filename. I keep it conservative. No clever names. If the document is unclear, it falls back to a safe pattern instead of guessing.

I chose this stack for speed and stability. Mistral handles the “what is this document” step well out of the box. OCR model is very good, but real scans still have errors. I run a light pass through Gemini to fix common OCR mistakes before naming. The goal is a consistent filename without rules, templates, or fiddly settings.

Small tools stick when they remove a recurring annoyance. File naming is about trust more than smarts. I want predictable names and fast feedback. If processing feels slow, it feels broken. That's why I focused on consistenc, speed and customisability.

Next I want this to live where your files live. A desktop watcher that renames on drop. Right-click actions in Google Drive and Dropbox. The API is live in v1 and a CLI for batch flows is on the way. If one of these saves you real time, tell me which one should ship first and why.

It has already found its first users, and I hope more people will find this idea useful

Try it: https://renamed.to

If it helps, tell me what worked. If it did not, tell me what blocked you. I will shave off the sharp edges next.