KindaCan Grok Bot build Wispr Flow?
It gets most of the way, but you stay in the loop · and here's exactly where.
- Price
- $15/mo
- You'd save
- $180/yr
- Who runs it
- You do
- Category
- AI assistants
Same shape as Granola, same seam: the microphone and the keyboard are on your machine, and a bot in the cloud is on neither. Dictation that types into whatever app has focus needs a local process with accessibility permissions only you can grant - so the bot writes it, you install it. The pieces are genuinely commodity now: whisper.cpp is excellent, a hotkey-to-text loop is small, and the result is private in a way a cloud dictation product can't be. What it won't have is the part Wispr actually sells - the model that turns rambling into clean prose as you go, tuned per app, on every platform at once. Also check the free tier first: 2,000 words a week covers a lot of people.
The prompt
Written for a bot that already has a browser, a terminal and your logins · paste it in whole.
The microphone and the keyboard are on my machine, not yours - so build this in two halves and be honest about the seam. Half one, runs on MY computer (you write it, I install it): - A tray app with one global hotkey: hold to talk, release to type the words into whatever field has focus. - Transcribe locally with whisper.cpp, small model for speed, and type the result in. Under two seconds from release to text or it will not get used. - A second hotkey for 'tidy this': send the transcript through an LLM (key from the env) to strip the umms and make sentences, then type that instead. - No cloud, no accounts, no telemetry. Audio never leaves the machine except that opt-in tidy call. - Ship me a README that says exactly which microphone and accessibility permissions to grant, and where to click. Half two, yours: - Keep the repo, iterate when I report friction, and test each build before you hand it over. - After a week, ask me for the transcripts of ten dictations I marked as bad and tune: model size, vocabulary hints, the tidy prompt. Before any of this: check my Wispr usage. If I'm under the free tier's 2,000 words a week, tell me to stop paying and change nothing else.
What you lose
- the cleanup model · theirs rewrites your rambling into prose as you speak, yours transcribes
- tone that adapts per app · an email and a Slack message come out the same
- your phone · the local build is desktop-only unless you build it twice
- the free tier, 2,000 words a week, which may already be all you need
The numbers
Pro is $15/user/mo billed monthly ($12 annual). The free tier includes 2,000 dictated words a week on desktop, which covers light use entirely.
wisprflow.ai/pricingChecked against the live page on 2026-08-24