The dictation I wanted

I have written this paragraph four times. The first three were on the keyboard. This one is voice — one hold of the hotkey, one sentence at a time, with the wrist budget I would otherwise have spent typing “the” and “and” five hundred times before lunch.

That is the whole pitch for the product I have been building. The keystrokes a developer pays for in a day are not all code. Half of them — the commit messages, the PR descriptions, the design docs, the replies to the people the code is for — are prose. The keyboard is the right tool for the code. It is not the right tool for the prose around the code.

The dictation that did not exist

I started by trying everything that already existed. The built-in macOS dictation is fast and free and built for the system text fields. It is not built for the rougher edges of my day — the terminal pane I have been ssh’d into for an hour, the inline GitHub PR comment box that lives inside the editor, the Notion page that reflows when I switch windows. It works some of the time and disappears at the wrong ones.

The cloud-first tools are the other shape. They are accurate and polished. They also send the audio of my working day — names of clients, half-written code reviews, half-formed thoughts about my employer — through a server I do not control. The accuracy is real; the cost is a privacy footprint I do not want to defend to anyone who asks.

The middle is empty. A native macOS dictation app that processes audio on the device, types into whichever field I am already in, and stays small enough that I can read every line of code that runs on my Mac. The middle, it turns out, is exactly the shape of the product I would pay for. So I built it.

The constraints

I wrote down the constraints before I wrote any code. The discipline of that step ended up being half the product. Five rules:

1. Audio never leaves the Mac in the default configuration. If somebody wants to opt into a frontier cloud model with their own API key, that is their call and their bill. The default is local.
2. The dictated text appears in the focused field. No separate transcription window, no clipboard step, no “switch to the dictation app and paste”.
3. There is a free tier and the free tier is the full app. Not a trial, not a limit per week, not “ten minutes per session”. A dictation tool with a quota is a tool I have to think about; I want one I forget about.
4. The custom dictionary is local. The list of words I correct most often — service names, client names, technical jargon — sits on my Mac in a file I can edit. It is too specific a signal to hand to anyone.
5. Donations route through PayPal, and the license arrives by email from a real human within two business days. I have shipped enough indie software to know that an automated checkout pipeline is a first-month distraction. The manual loop holds up; the automated one can come later.

Every product decision since then has fallen out of those five rules. A lot of features that other dictation apps ship — meeting transcription, server-side dictionary sync, an account model with a profile and a forum — fail at least one of the constraints. They are not in the v1 product.

What that produced

What I use every day looks like this. I hold the Option key with my left thumb, I talk into a desk microphone, I let go, and the paragraph appears in whatever I had focused. The custom dictionary knows that “next js” means “Next.js” and that the names of three of my clients have diacritics. It runs on every transcription regardless of which app or text field I am in.

Some of my day is faster than it was. The commit messages are longer — the body explains the why now, where before it usually did not. The PR descriptions have context the diff cannot carry. The AI prompts I write inside the editor are longer and more specific because the cost of writing them has stopped being a typing budget. The wrist-load at the end of a long day is materially different from a keyboard-only day.

Some of my day is not faster. Identifiers, shell commands, brackets and semicolons — none of that is dictation work. The keyboard is the right tool for the language the computer parses; the voice is the right tool for the language the people around me read. I use both.

What I will not promise

I will not promise that Voiacast is the best dictation tool. The honest answer is that the best dictation tool depends on what you need to do and where your trade-offs live. For a quiet desk, a Mac with Apple Silicon, and a daily flow that mixes code and prose, Voiacast is the tool I use. For someone who wants meeting transcription with diarisation, Voiacast is not yet that tool. For someone who lives in Windows, there is no Voiacast — and there might not be for a long time.

What I will promise is that the five rules above hold. Audio never leaves the Mac in the default configuration. The text lands in the focused field. The free tier is the full app. The dictionary is local. The license comes from a human in two business days. If any of those drift, we change the marketing before we change the product.

That is the dictation I wanted. It is now the dictation I use. If that maps to what you wanted, the download page is one hold-Option-Space away.