For a quick thought.
Keep fn space down while you speak. Let go and it's pasted.
Just say it.
Hold fn space, speak, let go. Dimmela types it wherever your cursor is.
Download for Mac
$ brew install --cask mameli/dimmela/dimmela
Try it here
Hold fn space.
We'll do the talking.
This is how Dimmela works, played in your browser. Hold fn space, let go, and the text lands at the cursor. Want it rewritten? Press shift while you talk and pick a Mode.
Click or scroll here to use your keyboard
Monday
Your Dictations land here.
Hold fn space on your keyboard, or press and hold it below.
- 1Hold fn space and talk
- 2shift opens the Modes
- 3A digit or ▲ ▼ to choose
- 4shift or return to confirm
Let go to paste. A quick tap of fn space Locks it: it keeps listening until the next tap. esc cancels.
One shortcut. Nothing to configure.
Hold to talk.
Tap to keep talking.
Recording starts the instant you press, so your first word is never cut off while Dimmela works out what you meant.
For when you have a lot to say.
Tap once and talk hands-free, for as long as you like. Tap again to finish.
Changed your mind?
Press esc and nothing is pasted, nothing is kept. An accidental tap with no speech costs nothing either.
Modes
Your prompts and skills,
a digit away.
Point Dimmela at the folders where you keep prompt files and skills (SKILL.md). Turn on the ones that make sense for dictation and drag them into order: the order is the digit. Edit a file and the next Dictation uses it.
Post-processing goes to any OpenAI-compatible provider, such as OpenRouter or a model running on your own machine. Every Dictation starts with your Default Mode, which can be none. If the provider is unreachable, you get the Transcript and a notification, so you never lose what you said.
Native Swift
Light enough
to forget it's there.
Dimmela sits in the menu bar all day and uses almost nothing while it waits.
40MB
of memory, with the speech model loaded.
Dimmela
~40 MB
Handy
~900 MB
0% CPUwhile idle. Your battery won't notice it.
Loaded at loginThe model is ready before your first Dictation.
No Dock iconIt lives in the menu bar, and only there.
Resident memory with Parakeet v3 loaded, compared with Handy running the same model. The model's weights live with the Neural Engine and aren't counted; even counting them, Dimmela stays well under half.
On-device
Your voice never leaves your Mac.
Speech is transcribed by Parakeet v3 on the Neural Engine. No account, no server, and it works offline.
- Audio stays local.Recordings are transcribed and kept only for your last five Dictations.
- Post-processing is opt-in.Only text is sent, and only to the provider you choose.
- Your key, in your Keychain.Not in a preferences file.
The small things
Built for dictating all day.
Vocabulary
Names and terms that speech recognition gets wrong are fixed in the Transcript, then handed to the LLM too.
No more “ehm”
Fillers like “uhm” and “mmh” disappear. Words with meaning, like “cioè” or “tipo”, never do.
History
Your last five Dictations, with their audio. Copy one, or Post-process it again with another Mode.
Many languages, no setting
Speak Italian, English, and other European languages, and mix them, without picking one first.
Dimmela.
/ˈdim.me.la/ Italian for “tell it to me.”
Download for Mac
$ brew install --cask mameli/dimmela/dimmela