Should I vibe code
Hotkey → record → Whisper → paste at cursor
Hotkey, Whisper, paste at cursor. This is the vibe coding success story everyone quotes.
?
Their verdict, the price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-03.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
Local dictation is the rare case where the homemade version is competitive on day one. Keep the model local and there is no data question at all.
What actually breaks
not "if". the specific failures.
- Audio sent to a hosted API when you assumed it was local
- A hotkey that conflicts with something else and types into the wrong window
Nothing here is dangerous, which is why the verdict is a flat go — but there is one honest moment. You set it up with a hosted transcription API because it was two lines and the local model was a download away, meaning to switch later. Weeks pass. Then you dictate something you would not have typed into a web form — a medical detail, a client's name, a half-formed complaint about someone — and realise mid-sentence that it has already been sent, and that you never read the retention policy of the service you chose in a hurry.
Is that you?
the verdict is a default, not a law
- You keep the model local
- It is just you
- You would be dictating confidential material through a third-party API without checking their retention policy
If you build it anyway
the checklist, then the prompt that enforces it
- Run the model locally. whisper.cpp is fast enough on modern hardware and removes the entire data question.
- Never persist audio after transcription unless you asked for it.
- Check the focused window before pasting, so a slow transcription does not type into whatever is in front of you now.
Build me a push-to-talk dictation tool. Requirements: 1. Local model only — whisper.cpp or an equivalent that runs on my machine. Do not add a hosted API fallback without asking. 2. Audio is held in memory and discarded immediately after transcription. Never write it to disk unless I pass an explicit --save flag. 3. Global hotkey to start and stop. Show a small unmistakable indicator while recording — there must be no way to be recording and not know it. 4. Before pasting, capture which window was focused when recording started and verify it is still focused. If it changed, put the text on the clipboard and notify me instead of typing into the wrong place. 5. Add a configurable dictionary of replacements for terms it reliably gets wrong (names, jargon), applied after transcription. 6. Keep a rolling buffer of the last few transcriptions accessible via a hotkey, in case a paste goes somewhere unexpected. 7. No telemetry, no accounts, no network access at all. Make that verifiable.
That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
Barely worth considering. This is the category where the homemade version is genuinely competitive on day one, and it keeps your voice on your own machine.
Nothing to exit. Delete the binary and your workflow returns to typing.
local transcription engine
open-source macOS dictation app
Questions
Is local transcription actually good enough?
For dictation, yes. Local Whisper variants have been fast and accurate enough for real-time use for a while now, and the latency is often better than a round trip to a server.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
last reviewed 2026-08-03 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice