What a dictation tool has to know about your screen

When we think about privacy with dictation tools, we often stop at the microphone. Questions like “Where does my audio go?” and “Who’s listening?” are totally valid, and we’ve answered them in detail in what Voxor sends to the cloud. But there’s another, less obvious question lurking beneath the surface.

Raw speech recognition spits out a jumble of words, but turning that jumble into usable text requires some decision-making magic. Where do sentences begin and end? Which words need a capital letter? And was that “period” supposed to be punctuation or part of a variable name?

Spoiler: none of these decisions can be made just from the audio. It’s all about the context: what app you’re using, what type of field your cursor is in, and even the surrounding characters. We broke this down when discussing dictating into a terminal.

The more your dictation tool knows about your screen context, the better it can format your text. If it doesn’t know whether you’re chatting or coding, it’s going to treat everything like an email, and that can lead to some pretty confusing documents!

Why this hits home for developers

For developers, this impact is even more pronounced.

Picture this: you’ve got a code editor open, and suddenly the number of places your text could go skyrockets. You’ve got an integrated terminal expecting literal characters, a source file that’s picky about spelling, comments that need to read like plain English, and a chat panel buzzing with its own set of rules. All of this is just a keystroke away!

A top-notch dictation tool must differentiate between all these destinations, and that’s not a design flaw, it’s part of the job! But it also means the tool needs to have a good look at your workspace, more than just being “a microphone app.” So it’s reasonable to wonder exactly where that watching happens.

What Local Mode actually covers

And here’s where Voxor’s Local Mode comes into play!

When Local Mode is activated, both halves of the process happen right on your computer. The speech recognition and the formatting layer that shapes your text based on the app you’re using both run locally. Many might think this is trivial, but getting formatting right is a game changer: without it, you’d just be staring at a wall of unpunctuated words!

The context that the formatter needs is kept on your own machine, not whisked away elsewhere. This is what we mean by “private by architecture,” not just by promise. It’s a real approach that doesn’t compromise your context.

The honest boundaries

Let’s outline four crucial points, because vague privacy claims aren’t very convincing.

This post covers Local Mode on Mac and Windows. Voxor is also on iPhone and Android now, but everything in this post describes the Desktop app, not a phone.

Hardware considerations. Running recognition and a formatting model locally needs real headroom. Voxor won’t enable Local Mode on underpowered machines, and the performance depends on your computer’s capabilities. So “runs on your machine” comes with an important note: make sure your hardware can handle it!

Local Mode vs. being offline. Just because Local Mode is on doesn’t mean the whole app will work offline. Things like signing in and updates still need the network. The switch just governs where your dictation content goes, not whether the product exists.

Cloud mode isn’t lesser. Cloud mode has its own advantages, and for some tasks, it’s totally worth it. The key here is that you should know your options, and that “where does my audio go?” is just one piece of the puzzle.

What to ask, of us or of anything else

If you’re dictating something you wouldn’t feel comfortable sharing with a stranger, here are three key questions to consider:

1. Does the transcription happen on my machine, or somewhere else?
2. Does the formatting and cleanup happen on my machine, or somewhere else? (This one tends to get overlooked but is extremely important!)
3. Can I turn the network off entirely and still dictate?

Feel free to throw these questions at us! Voxor’s cloud transcription and cleanup are performed in-house, and we keep how that works up to date in our privacy policy, not in a blog post that may get outdated. Plus, you can personally verify that Local Mode is working by disconnecting from the network and continuing to dictate. As easy as that!

So if you code for a living, Voxor for developers is the place to start, and there are setup notes for Cursor and VS Code.