Local by default
Your raw audio is processed on your PC. Network calls are listed plainly in the app. Cloud features, if any, are separate and clearly marked.
Local-first Windows dictation. Tap a key, speak, paste where your cursor is. Your audio never leaves your device — no cloud round-trip, no silent rewrites, no surprise model swaps.
Windows first. Mac and mobile next. Audio never leaves your device in local mode.
Most cloud dictation tools quietly run an AI cleanup pass. It softens your voice, removes the rough edges that make it sound like you, and sometimes invents words you didn't say. Heard Right ships in raw mode by default. Clean-up exists, it's clearly labelled, and it's off until you turn it on.
“Pull the latest from main, run the migration, then restart the worker — it's been hanging since the deploy yesterday.”
“Could you please pull the latest changes from the main branch, execute the database migration, and then restart the worker process? It has been experiencing issues since yesterday's deployment.”
“Pull the latest from main, run the migration, then restart the worker — it's been hanging since the deploy yesterday.”
Raw is the default. If you want polish, you opt in — and you can see what changed.
The first version stays narrow on purpose. If these three things are weak, nothing else matters.
Your raw audio is processed on your PC. Network calls are listed plainly in the app. Cloud features, if any, are separate and clearly marked.
Raw mode keeps your words. Clean-up can help when you want it, but it should never silently turn you into a different person.
If focus changes, paste fails, or an app blocks input, your transcript is still visible and recoverable. No work disappears into the void.
A hotkey-driven dictation tool that gets out of your way. Fast to start, fast to stop, predictable about where the text lands.
Default: Ctrl + Space. Speak naturally. Tap again and the transcript pastes where your cursor is.
Say "new line", "question mark", "open paren" mid-sentence and they get inserted as actual punctuation. Standard dictation, just reliable.
Names, cities, brand terms, repos, jargon. Accuracy on proper nouns jumps the moment you list them — and the list stays on your machine.
You see the words form as you talk. If paste fails, the text waits for you in the pill instead of disappearing.
Each of these tools is good at something. Here's where each one falls short for people who want their words to stay their words, on their machine.
| Heard Right | Wispr Flow | Win+H / Apple Voice | |
|---|---|---|---|
| Audio stays on your machine | Local first — cloud, if any, is a separate paid tier you opt into | Cloud only — every word leaves your device | On-device |
| Stays raw — no AI rewrite | Raw default — clean-up is opt-in and clearly labelled | AI cleanup is on — common complaint: "it rewrote me into a different person" | Raw — no AI rewriting |
| Private vocabulary that's actually private | Stays local — your terms never leave your PC | Cloud-stored — your custom words live on their servers | Limited — basic OS-level support |
| Recoverable paste | Always — transcript stays visible if paste fails | Sometimes — text can disappear if focus shifts | No — fails silently |
| Pricing | $69 lifetime founders / $99 standard | $15/month — no lifetime option | Free — bundled with the OS |
| Doesn't degrade after updates | Models are pinned per release — you choose when to upgrade | Reddit threads like "Wispr got worse this week" after silent model swaps | Stable — OS-level dictation rarely changes |
Wispr Flow is genuinely good if you don't mind cloud. Win+H is free and that matters; it also gives up on anything custom. Heard Right is the option for people who want dictation that respects their words and their privacy.
Lifetime pricing for the first Windows release. One-time purchase, no subscription, no usage caps on local mode.
First 200 customers. Lifetime updates for V1.
After founder pricing closes.
Optional. Capped by audio minutes — no surprise bills.
Not in local mode. The app still needs small network calls for licensing and updates, and any optional cloud features are separate. All network calls are shown plainly in the app.
The default installer is CPU-only. Auto mode chooses a smaller model on weaker machines and a better one when there's room.
That's what the private vocabulary is for. Add product names, repo names, cities, people — accuracy on those words jumps immediately, and the list stays on your machine.
The app keeps the transcript visible. You can copy it manually, retry paste, or cancel without losing the recording.
No. Models are pinned per release. You see the version number, and you choose when to upgrade.
Not for joining the beta list. Payment starts when the Windows beta is ready enough to charge for.
Join with email only. After that, tell us what you currently use and what it keeps getting wrong if you want a more useful beta invite.