Say it once,and it is already written
Between two meetings, the fastest keyboard is your voice. Hold a shortcut, say what happened, let go, and Khint types it where your cursor is. Point an agent or a workflow at that same gesture and what comes back is not a transcript: it is the drafted story, the summary, the ticket.
The notes you never write
are the ones you had no hands for
A product owner’s worst losses are not the long documents: they are the thirty seconds of clarity on the way back from a stakeholder’s desk, the decision someone drops in a corridor, the three blockers you understood perfectly at 09:12 and can only half-reconstruct at 11:00. Typing is what stands between the thought and the record, and typing needs a surface, two hands, and a window you are not currently presenting from. Speech does not. Khint’s dictation exists for that gap: one key held down, spoken words, text at the cursor. And because it plugs into the same agents and workflows as everything else in Khint, the words do not have to stay words.
From held key to finished text
Bind it once, then speak into any app, any agent, any workflow.
Bind the key once
Dictation ships switched off: Khint has no voice shortcut until you set one in Settings. That is deliberate, because a microphone that could open without you asking is not a feature. Choose a key combination you can hold with one hand, and you are done configuring.
Hold it, speak, let go
The microphone opens when the key goes down and closes when it comes back up: release is what ends the take. A moment later the text lands at your cursor, in whatever app you were already typing in, the same way every Khint action delivers its result. There is no always-on listening and no pre-roll buffer, so your operating system’s recording indicator is lit exactly as long as your finger is down.
Fillers out, punctuation in
Spoken sentences are not written sentences. A plain dictation goes through a tidy-up pass in the same request: the “um”, the “so yeah”, the false starts come out, and the punctuation you never said gets added. If that pass cannot run, you still get the raw transcript rather than nothing. One dictation, one credit.
Your project keys come out right
On clean speech, common words are rarely the problem: names are. AUTH-112, the name of the module nobody outside your team has heard of, the vendor everyone mispronounces. Khint hands the transcription a short vocabulary hint built from the agents and workflows in your active pack, so the terms you work with every day are already expected.
Point an agent at your voice
An agentcan declare that its input is your voice instead of your selection, and then its own shortcut becomes hold-to-speak: no second binding, no new gesture to remember. Hold your “Draft user story” key, describe what the stakeholder just asked for, release, and what comes back is the drafted story, not a transcript of you thinking out loud.
Or start a whole workflow with it
A workflow can be voice-triggered the same way. Your spoken words become the input to step one, and the chain runs as it always does, up to a final step that writes the result into Jira, Confluence, Linear, Notion or Slack through the integrations you connected. You speak once; a ticket exists.
Speak into the palette, then read it back
When the palette is open, the same gesture behaves differently on purpose: the transcript fills the idea bar rather than being typed into whatever sits behind the window. It is filled, not fired. You read what Khint heard, fix a word if you want, and press Enter to run it, which is the difference between dictating an instruction and having one executed unseen. See the natural-language command bar for what happens after Enter.
A Tuesday, spoken
09:12.Stand-up ends with three blockers you promised to relay. Your cursor is already in the team channel, so you hold the dictation key, say all three the way you would say them out loud, and let go: they arrive punctuated, without the “so basically” you started each one with.
11:40.A stakeholder catches you between rooms and describes a change they want. You do not open a document. You hold the shortcut of your “Draft user story” agent, which is set to listen, and describe the request in your own words. What lands is a story with acceptance criteria, ready to be edited rather than started.
15:05. A tester walks you through a bug at their screen. Your voice-triggered workflow drafts the report and its last step files it, so the issue exists before you are back at your desk: the spoken version of writing to Jira from anywhere.
17:30.You open the palette, hold the key, and say what you want done with the day’s notes. The words land in the idea bar; you read them back, correct one product name, and press Enter. Everything spoken today ends up where the typed version would have, minus the typing.
What happens to the audio
Dictation is opt-in and does nothing until you bind a shortcut. The microphone is open only while the key is held, never before, never after. On release, the audio clip and the short vocabulary hint travel through Khint’s backend to a speech-to-text provider and come back as text: Zero Data Retention is enabled on that account, so the audio and the transcript are not stored there and are not used for training, and Khint keeps no server-side copy of the recording. If a dictation feeds an agent whose result you would rather not send as-is, the same per-action controls apply as everywhere else in Khint, including redacting personal data before an action runs. The full list of who sees what is on the privacy page, and the reasoning behind it in privacy-first AI for product work.
Dictation is an input, not a separate mode
Most dictation tools end at the transcript: they give you your words back and leave the work of turning them into something usable to you. In Khint, your voice is simply another input source, sitting beside selected text, a document, and your active session. That is why an agent can be pointed at it without inventing a new gesture, why a workflow can start from it and still finish inside Jira or Confluence, and why a spoken note can land in a Memory session as context the rest of your afternoon can draw on. The gesture is small. What it is wired to is the point.
Common questions
Do I need a separate dictation app?
No. Dictation is part of Khint, but it is opt-in and starts switched off: there is no dictation shortcut until you bind one in Settings. Nothing listens before that, and once bound, the microphone only opens while you hold the key down.
Does dictation use up my plan?
One dictation costs one credit, the same as a light AI action, and the tidy-up pass that strips fillers and adds punctuation happens inside that same request rather than as a second charge. When you point an agent or a workflow at your voice, the run it triggers is billed like any other run of that agent or workflow.
Will it spell our ticket keys and product names correctly?
That is the error class Khint works hardest on. On clear speech, ordinary words come back fine and proper nouns are what break: project keys, product names, internal jargon. Before transcribing, Khint primes the model with the names of the agents and workflows in your active pack, so the vocabulary you actually work with is already in front of it.
Can I see what it heard before anything runs?
Yes, and in the one place it matters most. With the palette open, dictation fills the idea bar instead of typing behind the window: the transcript is placed in the field, not executed, so you read it back and press Enter yourself. When a workflow is voice-triggered from its own shortcut, the transcript is shown in the status HUD as the run starts.
Stop typing what you could say
Free with 300 credits a month, about 10 AI actions a day. No credit card. Bind one key and dictate the next thing you were going to write down.