Hold fn and talk.
Recording starts the moment the key goes down, so your first word is never clipped. Talk the way you would to a person.
Hold fn, say what you mean, let go. FlowLite transcribes on your Mac, cleans up the ums and grammar, and types it where your cursor is.
Free & open source · GPL-3.0 · macOS 15+ · Apple Silicon
Recording starts the moment the key goes down, so your first word is never clipped. Talk the way you would to a person.
Whisper large-v3-turbo runs on the Apple Neural Engine. Your audio never leaves the machine, and your vocabulary keeps names spelled right.
Claude Haiku trims the ums, applies your corrections, fixes grammar and matches the app you're in. Then FlowLite types it in for you.
umsomeandhimwasuhgoingtoshipitfridaynowaitthursday
Choose Standard editing to fix grammar like this, or Light to keep your grammar exactly as spoken. If cleanup can't run, you still get the raw transcript.
No plugins, no per-app setup. FlowLite types into the field you're in, and in apps that need it, pastes and puts your clipboard back.
The build is green now. We can ship it tomorrow morning.
Hi Priya, thanks for the notes. I'll send the revised deck by Thursday.
Blockers: he and I were going to fix the login bug, but he doesn't have access to prod.
Run the tests with swift test before opening a pull request.
Thanks for reaching out. I'm free on Tuesday after two.
Yeah, sounds good. See you at seven.
Hold fn, speak, release. Text lands at your cursor in about a second and a half.
Double-tap fn to keep listening while you pace around. Tap again to finish.
Select text, hold Control and fn, and say “make this friendlier” or “translate to Spanish”.
Casual in chat, polished in email, careful with code. It continues the sentence you were writing.
Teach it names and jargon, and set replacements like “my email” for your address.
Offline, no key, or a slow reply: the raw transcript goes in instead. Every dictation is in History.
No account, no telemetry, no FlowLite server. The only network call is cleanup, made with your own API key.
Whisper runs locally on the Neural Engine. Audio is never uploaded, and History is a local database you can clear any time.
The transcript, the app's name and about 2,000 characters around your cursor go to Claude Haiku. Switch cleanup off and nothing is sent.
FlowLite doesn't read them, doesn't send them for cleanup, and doesn't keep them in History.
FlowLite is built from source with Xcode. On first launch it downloads the 1.5 GB speech model.
$ git clone https://github.com/Jenish109/flowlite.git$ cd flowlite$ brew install xcodegen# once: a local signing certificate so permissions survive rebuilds$ Scripts/make-signing-identity.sh# build, sign, install to ~/Applications and launch$ Scripts/build.sh --run