AI tools that run on your device
Cut the background out of a photo, turn a recording into text, read the text off a photo of a letter, with the model running inside your browser tab instead of on someone else’s server. You pick the file, your own device does the work, and the result downloads from the same page.
What “runs on your device” means
The traffic goes the other way round. Instead of sending your photo to a server, the page downloads the model once and the browser keeps it: about 44 MB for background removal and 40 MB for speech on a phone, up to 170 MB and 150 MB on a computer whose browser uses the GPU, and about 6 MB for text recognition in a pair of languages. From then on the work happens on your own processor, at whatever speed your device manages. Your file is never sent anywhere, because there is nowhere for it to be sent.
What that costs you instead
Here is the trade: the first run spends real megabytes of your connection, and a phone is slower than a data-centre GPU. A photo is seconds either way; a long recording is minutes, and the tab has to stay in the foreground while it works. A backgrounded tab gets throttled by the browser. Speech works on speech, not on songs, and the small model mishears names, so the transcript is a draft you can edit in place rather than a finished document. Text recognition reads print, not handwriting.
Questions
Are my photos or recordings uploaded?
No. The model is downloaded to you; your file stays in the browser tab and is never sent to a server.
Does it work offline?
Partly. After the first run the model stays in your browser, so later files need no download, and a page that is already open keeps working if the connection drops. Opening a tool page still needs a connection; for offline use, install the desktop app and run each tool once while online.
Is there a credit limit or a watermark?
Neither. There is no account, no per-image credit, and nothing is stamped on the result.
Why is it slower than an online AI service?
Because your device is doing the work rather than a rack of GPUs. That is the trade: you wait a little longer, and your file never leaves your machine.