Hold Fn, speak, let go.
Your words, at your cursor.
Voice typing is part of Kesami for Mac. What you say is pasted where your cursor is, in any app, and speech recognition runs on your Mac.
macOS 13 or later · Apple silicon
Voice typing is macOS only: Kesami’s Windows (beta) build doesn’t include it.
Kesami needs a Mac with Apple silicon (M1 or later).
“Thanks for the call today. I’ll send you the pricing proposal by Friday.”
Your text field · after you let go
Thanks for the call today. I’ll send you the pricing proposal by Friday.
Everything starts with Fn.
Hold it to talk, tap it twice to go hands-free, and add Control to give an instruction.
Push to talk
Fn hold
Speak while you hold it. When you let go, the text is inserted where your cursor is.
Hands-free
Fn twice or Fn + Space
Speak without holding the key, then press Fn or Fn+Space again to insert. Pressing again within half a second cancels instead.
Command
Fn + Control hold
Your words become an instruction, for the selected text or for something new at your cursor. Double-tap Fn+Control to lock it hands-free.
Cancel
Esc
Stops listening, processing or a running command.
Paste the last dictation
Control + Command + V
Pastes your last dictation again, including what was heard before a cancel.
One setting first. macOS also acts on the Fn key. In System Settings › Keyboard, set what pressing the Globe (Fn) key does to “Do Nothing”, and don’t use a Dictation shortcut that relies on Fn.
Spoken loosely, typed cleanly.
On macOS 26 or later with Apple Intelligence turned on, clean-up can tidy a dictation on your Mac before it’s inserted.
What clean-up fixes
- Filler words and false starts are removed.
- Self-corrections keep only the correction: “at 2, actually 3” becomes “at 3”.
- Punctuation, capitalisation and grammar are fixed.
- Spoken lists become one item per line.
- Spoken commands such as “new line” and “comma” are applied.
Skipped for phrases under three words, Hindi or mixed Hindi-English speech, very long dictations, and languages Apple Intelligence doesn’t support; the recognised text then goes in with only your snippets expanded.
um so i’ll send acme the pricing proposal on thursday actually friday comma with sso included
Your text field
I’ll send Acme the pricing proposal on Friday, with SSO included.
Choose a level under AI cleanup in settings
- Light
- The default. Makes the changes above.
- Medium
- May also reword for clarity.
- Off
- Inserts what was recognised, with your snippets expanded.
Add Control.
Say what to change.
Hold Fn and Control, and what you say becomes an instruction: for the text you’ve selected, or for something new.
Rewrite the selection
Select text, hold Fn and Control, and say what to change, such as “make this more formal” or “translate into Hindi”. The result replaces the selection.
Draft to Priya · selected
Priya, proposal’s coming Friday and SSO is in it.
“Make this more formal.”
Draft to Priya · after the command
Hi Priya, the proposal will be with you by Friday, and SSO is included.
Draft from a description
With nothing selected, describe what to write, such as “a polite note declining Thursday’s meeting”. In most apps, the result is inserted at your cursor.
“A polite note to Priya declining Thursday’s meeting.”
Your text field
Hi Priya, thank you for the invitation. I’m afraid I can’t make Thursday’s meeting, but I’d be glad to find another time.
Your configured AI
The selected text and your instruction are sent to your configured meeting AI: Kesami’s cloud AI when you’re signed in, your own key’s provider, or Claude Code. Results are inserted as the AI returns them, without clean-up.
What it sends
The selected text, or none for a draft, and your instruction. Commands don’t search your meetings or create one. In an editor that copies the whole line when nothing is selected, that line is sent as the selection.
Counted on Free
Each command uses one of the Free plan’s 20 AI uses a month, the same allowance as chat replies. Pro is unlimited, and a request you cancel with Esc isn’t counted.
Down to the details.
Which languages it hears, the words it should know, and the line that shows what it’s doing.
Languages by macOS version
On macOS 26 or later, Apple’s newer recogniser runs on your Mac and covers more languages, Hindi included. A language whose model isn’t installed yet offers a one-time download in settings; after that, a session can run up to 20 minutes. On macOS 13 to 25, the older recogniser stops after 55 seconds and doesn’t offer Hindi.
Last dictationSettings
Needs a one-time download from Apple before you can dictate.
Words it should know
Add names and terms to Custom words, separated by commas: recognition leans towards them, and clean-up is told to keep their spelling. Snippets turn a spoken trigger into fixed text, inserted exactly as written. Triggers match whole words in any case, and the longest one wins.
Custom words (comma-separated)
Acme, Priya, Kabir, SSO
Snippets: one per line, say the trigger to insert the text (my link => https://…)
sign off => Thanks, and speak soon.
“sign off”
Your text field
Thanks, and speak soon.
A status line above the Dock
A thin line shows what voice typing is doing, and turns cobalt while it works. Hover and it expands into a bar that names the step, such as Listening or Writing. Click it to dictate hands-free, or drag it anywhere.
- At rest
- While it works
Dictation audio is processed on your Mac.
Audio never leaves your Mac, and speech recognition and clean-up never call Kesami’s backend.
Stays on your Mac
- Your audio.
- Your last dictation, in memory only. The helper doesn’t save transcripts, instructions or AI results on your Mac.
- Your settings: spoken language, clean-up level, sounds, review preference, custom words, snippets and the status line’s position.
Sent when you ask
- Fn+Control commands and Ask AI send the selected or panel text and your instruction.
- They’re sent to your configured AI: Kesami’s cloud AI, your own key’s provider, or Claude Code.
- They don’t search your meetings or create one.
Permissions it asks for
- Accessibility, Microphone and Speech Recognition. Accessibility is for the shortcuts, reading the focused field and pasting.
- Input Monitoring, only if macOS refuses the keyboard tap.
- macOS lists them under “Kesami Voice Typing”, a separate helper app, rather than Kesami itself.
Before you hold Fn.
Which Macs, how to turn it on, what counts toward Free, and what happens to your clipboard.
Which Macs does it work on?
On a Mac, Kesami needs Apple silicon (M1 or later) and macOS 13 or later, and voice typing runs wherever the Mac app does. Hindi needs macOS 26 or later and a one-time download; clean-up needs macOS 26 or later with Apple Intelligence turned on, in a supported language. On macOS 13 to 25, a dictation stops after 55 seconds.
Is voice typing available on Windows?
No. Voice typing is part of the Mac app only; the Windows build doesn’t include it. Kesami’s meeting notes run on both macOS and Windows (beta).
How do I turn it on?
It’s off until you turn it on: use the Voice typing switch on Kesami’s Home screen, or choose Voice Typing… in the Kesami menu or the menu bar dropdown. Home then offers Set up voice typing, for the Accessibility, Microphone and Speech Recognition permissions it needs.
macOS already uses my Fn key. What do I change?
In System Settings › Keyboard, set what pressing the Globe (Fn) key does to “Do Nothing”, and don’t use a Dictation shortcut that relies on Fn. Voice typing settings show a warning, with a button to that page, while the macOS setting conflicts.
What counts toward the Free plan?
Only Fn+Control commands and Ask AI requests. Each uses one of the Free plan’s 20 AI uses a month, the same allowance as chat replies, and Pro is unlimited. A request you cancel with Esc isn’t counted. Dictation and clean-up run on your Mac and never call Kesami’s backend.
What is Ask AI?
Review before inserting, off by default, shows each dictation in the helper’s panel first. There you can edit it, type an instruction for Ask AI, undo the AI’s change, or click Insert text to paste it into the app you dictated into. Ask AI sends the text and your instruction to your configured AI, like an Fn+Control command.
Can clean-up change what I said?
Light only removes and tidies; Medium may also reword for clarity. Every result is checked first: one that adds words your level doesn’t allow, changes a number, answers a question you dictated, drops more than a self-correction explains or starts like an assistant’s reply is discarded, and the recognised text goes in instead, with your snippets expanded, so a dictated instruction is typed, not obeyed.
Does it work in every app?
It pastes into whichever text field has your cursor, in any app. If nothing takes the paste within four seconds, for example because no text field had focus, the text stays on your clipboard and the status line says Copied to clipboard. If you switch apps while text is being prepared, it waits in the helper’s panel, where Insert text pastes it back into the app you started in.
What happens to my clipboard?
Text goes in by pasting. Once the app has read it, your previous clipboard is put back, and anything you copied in the meantime is kept. If nothing reads it within four seconds, the text stays on your clipboard and your previous clipboard isn’t put back. PDF, rich text with attachments, audio and file-promise contents aren’t restored, and a very large copy may come back only in part. In apps that don’t share their selection, a command copies it with Command+C and restores your clipboard afterwards. Editors that copy the whole line when nothing is selected send that line instead.
Can it type into password fields?
No. Dictation doesn’t start in a field macOS reports as secure, and a paste into one is refused. Voice typing also never presses Return or submits a form.
Write your next reply by voice.
Download Kesami for Mac, turn on Voice typing from the Home screen and allow its permissions, then hold Fn and speak.
macOS 13 or later · Apple silicon · .dmg disk image
Kesami needs a Mac with Apple silicon (M1 or later).