Your hardware is enough

The laptop you already own can transcribe an hour of audio with a modern speech model. It can also upscale a photo, separate a voice from background noise, translate speech, read the text off a scan, and run a language model you can have a real conversation with. The models and the runtimes are open, and browsers can now use your GPU. Even so, most AI tools on the web still want your file on their server and stop after three free minutes unless you pay.

That made sense when this kind of work needed a data center. For a lot of everyday tasks it no longer does, because the computer on your desk is enough. So when a service caps your minutes and charges a monthly fee for one of these tasks, you are paying rent on software that could run on your own machine. We built YourDevice to run these tasks on your machine instead.

Running on your own hardware also means everyone gets the same tools. A newsroom and a student both pay nothing, and anyone with a browser can use them, whether or not they can afford another subscription.

How it works

Every tool here works the same way. The model downloads to your browser once and then runs on your machine, on your GPU if you have one or on your CPU through WebAssembly if you don't. There is no upload step, because there is no server doing the work, and you don't need an account. Your files are never uploaded — everything runs on your device.

Because of that, our costs are the same whether you transcribe one voice memo or ten years of podcasts. We have no reason to meter you, so the tools are unlimited.

It also means your privacy does not depend on trusting us. We never ask for your files and never receive them, so we have nothing to leak, sell, hand over to a court, or train a model on. This is how the tools are built, so there is no setting we could quietly switch off in next week's update. You can check it yourself: open the network tab in your browser's developer tools, or turn off wifi after a tool has loaded and see that it still works.

Open source and free

The engine behind the tools is open source under the MIT License, and the models are open too, starting with Whisper. You can read the code, or fork it and build something better. The tools are free and will stay free. There are no ads today. If we ever add them, they will be unobtrusive and tracking-free. You can support the work on Ko-fi if you want to. None of this limits the tools or gives anyone access to your files.

What comes next

We started with transcription and have since added image upscaling, noise removal, OCR, PDF tools and, lately, a private chat with a real language model. Each of these is usually sold as a metered online service. Our file converters need no model at all. Your browser can already decode a video or read an AVIF file, so a converter site that has you upload and wait is charging you for a decoder you already have.

As hardware gets faster and open models get better, more tasks will run well on your own device, and we will keep adding tools for them. New tools will work the same way as the ones here now.