Why we built this

Almost every "convert this to text" tool online works the same way: you upload your file to a company's server, it runs the conversion there, and it sends text back. That is a strange bargain when you stop to look at it. To read the words in your own receipt, your own recording of a meeting, or your own signed contract, you first have to hand a copy of it to a stranger's computer and trust what they do with it afterward.

For a long time there was a good technical reason for that bargain: browsers simply couldn't run the heavy machinery — speech recognition, OCR, video decoding — that these conversions need. That is no longer true. Modern browsers can run real machine-learning models directly, fast enough to be useful, using the same hardware acceleration that powers everything else on your screen. So the upload step is now mostly a habit, not a necessity.

We built this to drop the habit. The default here is simple: your file is converted where it already is, and it never leaves your device. No account, no upload, no watermark, no rationing — because when the work runs on your machine, giving it away for free costs us almost nothing.

When you drop a file in, a converter loads inside your browser tab — in a background worker, so the page stays responsive — and does the whole job on your own hardware. Here is what that actually means for each kind of file, without the jargon:

• Audio and video are decoded in your browser, and a speech-recognition model listens to them and writes down what it hears, with timing you can export as subtitles. Where your device has a fast graphics chip, the model uses it; where it doesn't, it falls back to a slower but reliable path. Either way, the audio is never sent anywhere.
• Images are read by an OCR model — optical character recognition — that finds the printed characters in a photo, screenshot, or scan and turns them into selectable text.
• PDFs and Office documents are opened and their text pulled out directly. If a PDF turns out to be a scan with no real text in it, the tool notices and quietly runs the same on-device OCR on the page images.

The first time you use one of these, it downloads a small model file once — think tens of megabytes, roughly the size of a short video — and stores it in your browser. Every conversion after that starts instantly and works even with no internet connection, because the model is already on your device. Nothing about your file is uploaded at any point; you can verify that in your browser's network tab.

We would rather tell you exactly where the rule bends than pretend it never does. Two tools cannot run fully in your browser, because they start from a link you paste instead of a file you hold, and a link has to be fetched by a server:

1. YouTube transcripts. When you paste a video link, that URL goes to our server so we can request the video's existing caption track through an API. We fetch the caption text and hand it back. We never download, re-host, or store the video or its audio — only the text of captions that are already published on the video.
2. Website to Markdown. When you paste a page URL, our server fetches that page and converts it into clean Markdown for you.

In both cases the thing that leaves your device is the URL you typed, not a file of yours. These two are the entire list. Everything else — every audio, video, image, and document tool — runs locally with no upload. We keep this boundary bright on purpose, and the privacy page documents precisely what each of these two tools sends and stores.

We are a small, focused project, not a venture-funded platform trying to become your everything app. There is no dashboard to log into to use the free tools, no "AI assistant" bolted onto every page, and no plan to turn your files into training data — we could not do that even if we wanted to, because for local conversions we never receive your files in the first place.

We do plan to offer paid options later for the genuinely hard jobs a browser can't do well — things like cloud-speed transcription of very long recordings, or reading handwriting — because those cost real money to run. But the core promise doesn't change with them: the free, in-browser tools stay free and stay private, and exporting your own text is never locked behind a paywall. If a tool can honestly run on your device, it will — and we will keep saying so plainly.