Finding things inside photos, scans and voice notes
Some files are easy to find. A Word document called Q3 budget final.docx will come up in any search for "budget". The trouble starts with files that have no words a normal search can see.
These are the files people lose most often. Here are the common ones, and what Khoj does with each.
Screenshots
A screenshot is a picture. Your computer stores it as pixels, and the file name is only a date and a time. But a screenshot usually exists because it showed something you wanted to keep: a payment confirmation, a booking number, an error message, a piece of a chat.
Khoj reads the text in each screenshot and also looks at what the picture shows. Type "UPI payment Sai Motors" or "the error about the printer" and the screenshot comes up.
Scanned documents
A scanned PDF looks like a document, but inside it is a stack of photos of pages. Ordinary search sees nothing. Khoj reads each scanned page as a picture and pulls out the text, so a search for a GST number, a policy number or a name on a form can find it.
Photos of paper and whiteboards
We all take photos of things so we do not have to write them down: the bill at the clinic, the notice at school, the plan on the office whiteboard. Khoj treats these the same way as screenshots. It reads the writing and understands the scene, so "whiteboard from Monday" finds the photo even when the photo is called IMG_4471.HEIC.
Voice notes and recordings
A voice note can hold a date, a price or an address that you never wrote anywhere else. Khoj listens to recordings in short parts. When you search, it can point you to the part where the words were said, for example "at 2:14, service is due at ten thousand km".
Videos
For videos, Khoj looks at one frame from each shot and listens to the sound. A search for "the slide with the sprint plan" can find the meeting recording and the moment the slide was on screen.
You do not need the exact words
Khoj matches by meaning. "Rent agreement" can find a lease that says "tenancy contract". "Car service" can find a voice note about "the garage visit".
All of this reading happens on your own computer. Nothing is uploaded, and it works without internet. If you want to see the idea in action, try the sample search on the home page.