Bind one shortcut, say what you mean, let go. The words land at your cursor in whatever app you were already in: cleaned up, punctuated, and spelled the way your team writes them.
so um basically checkout retries twice on a declined card, uh it reproduces on staging with the test card ending 0002, and it blocks the friday release
All of it inside one request, for one action's worth of credits. There's no separate voice meter
Checkout retries twice on a declined card. It reproduces on staging with the test card ending 0002, and it blocks the Friday release.
With the palette open, the same key fills its idea bar instead of typing into your document. The transcript is filled in, never fired: you read it back, then press Enter and Khint resolves a plan out of the agents and tools you already have.
Dictation retargets to the idea bar the moment the palette is open, so a sentence you say is a sentence you can still edit. Planning costs no credits, the plan names every step before it runs, and a run worth keeping can be saved as a real workflow.
Raw speech isn't writing. Between the key and your cursor Khint drops the hesitations, punctuates the sentence, spells your own nouns the way you do, and refuses the takes that hold nothing.
Hesitations out, punctuation in, inside the same request. What lands reads like something you wrote rather than like a transcript, and it costs one action's worth of your monthly credits, cleanup included. If the cleanup can't run, you still get the raw take rather than nothing.
One gesture on every key that listens. A short press latches the mic open; hold it longer and it ends the moment you let go.
Generic dictation guesses at your nouns. Khint primes the recognizer with your own agent and workflow names, so ticket keys and product terms come out the way you mean them.
A silent take is refused rather than sent. A speech model answers silence with its most frequent filler, and that reaches you as your own words, so the level itself is the gate. A take you want back is thrown away by holding the key past the deadline, before anything leaves the machine.
There's no always-on mode to switch off, because there's no always-on mode. The key you press is the whole permission.
Security and privacyBind the dictation key, point an agent at it, or give a workflow its own. Each one takes about a minute to set up.
Bind one key in Settings and every text field on your Mac becomes dictatable.
Set an agent's input to Voice and its own shortcut becomes the microphone.
See what an agent readsA whole routine can begin from a sentence, and end in the tools your team reads.
See workflowsYour voice is one way in. These three are what it can set off.
Set a dictation shortcut in Khint, press it, speak, and press it again. The transcript is typed at your cursor in whatever app you were in: Mail, Slack, Jira, a code editor, anywhere you can type. Prefer holding the key instead? That works too, and it ends when you let go. No recorder window, nothing to paste, and no model to download first.
Yes. When you dictate plain text, hesitations and filler words are dropped and punctuation is fixed in the same pass, so what lands reads like something you wrote. If the cleanup can't run, you still get the raw transcript rather than nothing. Dictations that feed an agent, a workflow, or the idea bar skip the cleanup: the agent reads through hesitations itself, and skipping it keeps the run fast.
That's the part most dictation tools miss. Khint primes the recognizer with the names of your agents and workflows, so the proper nouns that matter in your work come out spelled the way you mean them: ticket keys like AUTH-112, product terms, project names.
Yes. Set an agent's input to Voice and its own shortcut starts listening: press, describe, press again, and the agent runs on your words, pasting its result back as usual. A workflow can start from your voice the same way, and it shows your transcript as the run starts, so you see what it heard before the steps land anywhere. With the palette already open, the same gesture fills its idea bar instead, and stops there until you press Enter.
One dictation costs one action's worth of your monthly credit balance, cleanup included: it runs inside the same request. There's no separate voice meter and no per-minute pricing. There's also nothing to download and nothing to warm up, so it's just as fast on an ordinary work laptop as on a new Mac.
The microphone opens on a press and closes on a press: your OS recording indicator is lit for exactly that span, never outside a dictation you started, and Khint keeps an indicator on screen the whole time. A silent take is refused rather than sent. The clip is then transcribed on Khint's servers by a speech provider with zero data retention: audio and transcripts aren't stored and not used for training, and the provider is named plainly on our privacy page. Khint keeps no server-side copy of the audio.
Free with 300 credits a month. No credit card. macOS 13+ or Windows 10 22H2+.
Create issue