Skip to content
Menu-bar dictation, v1.0

Press. Speak.
It is written.

QuillMic lives in your menu bar. One hotkey turns the microphone on, another opens Nika, an in-app voice assistant. Words land in whatever app you were already using — Slack, Mail, your editor, your terminal. No window switching. No subscription required.

Download for macOSHow it works
  • Apple silicon native
  • 100% offline mode
  • Mac App Store ready
1 chord
From idle to dictating
<400ms
On-device latency
0
Telemetry pings
100+¹
Languages, model-dep.

¹ Tuned for non-native accents on the Best preset with the largest on-device or cloud model selected. Results on Light / Balanced presets, smaller Macs, or noisier rooms will vary — accuracy and language coverage both depend on the model you choose.

The pipeline

A hotkey, a microphone, and an obsession with not breaking your flow.

  1. 01⌃ ⌥ ⌘ Space

    Press the chord

    Anywhere on macOS. Slack, Mail, a terminal prompt. QuillMic flips the menu-bar dot from grey to ember.

  2. 02you speak

    Speak naturally

    Whisper, Parakeet, or OpenAI handles the audio. Filler words, repeats, and stray commas are quietly removed.

  3. 03↵ inserted

    It lands in place

    Cursor returns to the focused field with clean text. Accessibility path when granted, pasteboard when not.

Nikaanswers in your voice, too. A second hotkey opens her — draft a reply to the email you just read, reshape an awkward paragraph, or ask an external model for today's news, a stock price, or anything else, and get it spoken back without leaving the field.

⌃ ⌘ Space
What's inside

Seven small ideas, executed with care.

QuillMic is not trying to be every dictation app. It is trying to be the one you reach for without thinking, twenty times a day.

  • 01

    Hotkey-driven

    Three global chords, fully remappable. Toggle dictation, summon Nika, replay the last reply.

  • 02

    On-device by default

    Bundled Whisper base.en works the moment you finish onboarding. No account, no network round-trips.

  • 03

    Cloud when you want it

    Bring your own OpenAI key for gpt-4o-transcribe or polish. Off by default, clearly labelled when on.

  • 04

    Tuned for accents*

    On the Best preset, QuillMic handles non-native English and regional accents that trip lighter dictation tools — names, place names, code-switching included.

  • 05

    Nika, the resident assistant

    Draft an email reply, reshape a clumsy paragraph, or ask an external model for the news, a stock quote, or a quick fact — all by voice, all in place.

  • 06

    Custom vocabulary

    Teach QuillMic your jargon — names, acronyms, codes — and it stops mishearing them on the first try.

  • 07

    Polish styles

    Conservative, light, aggressive, or off. The LLM cleans punctuation and capitalization — never rewrites your voice.

  • 08

    Designed for the menu bar

    No Dock icon. A floating overlay shows state without stealing focus. Quit and it leaves no breadcrumbs.

* Accent and language quality depend on the selected ASR model and your Mac's available memory. The Best preset (FluidAudio Parakeet on-device or gpt-4o-transcribe in the cloud) needs ≥16 GB free RAM for on-device, or an active OpenAI key for cloud. Lighter presets and smaller Macs trade some of this accuracy for speed and footprint — QuillMic tells you exactly which model is running in Settings → Models.

The accents work

Built for the way the world actually speaks.

Off-the-shelf dictation models are trained on a narrow slice of speakers, then quietly fail on everyone else. We did not want that. So a meaningful chunk of QuillMic's development went into the things that decide whether your accent gets respected:

  • Model curation. Whisper, Parakeet, and the OpenAI tier are picked for accent coverage first, raw WER second.
  • Vocabulary you control. Names, places, jargon, code-switched words — all addable so the model stops Anglicising what you actually said.
  • Per-language polish. The deterministic clean-up step is tuned per-locale, so it does not "fix" your phrasing into something it isn't.
  • Honest fallbacks. When a smaller model is loaded, QuillMic tells you and offers to upgrade. No silent guessing.

Accent quality scales with the model you load. The Best preset — FluidAudio Parakeet on a 16 GB+ Apple silicon Mac, or gpt-4o-transcribe in the cloud — is where the heavy work shows. Lighter presets are still solid, just lighter.

  • Indian Englishtranscribed clean

    “Schedule a meeting with Aarav and Priyanka for next Tuesday at half past three.”

  • Brazilian Portuguese accenttranscribed clean

    “I will deploy the staging build to São Paulo region after the QA pass.”

  • Slavic accenttranscribed clean

    “Open the document called road-map dot key and jump to slide twelve.”

  • Scottish Englishtranscribed clean

    “Send Iain the contract draft and ask him about the Aberdeen pricing.”

Why it matters

Names should be spelled right. Places should not be turned into nearby-sounding English words. Your accent is not a bug.

The voices

Two ways to be heard.

QuillMic uses text-to-speech only when Nika speaks a reply — never to read your dictation back. Pick the engine that suits the moment: the system voices already on your Mac, or the curated neural library that ships with FluidAudio.

01 / Native

macOS system voices

Free. Instant. Already installed.

Whatever voices you already have under System Settings → Accessibility → Spoken Content are immediately available to QuillMic. No download, no model load, no extra disk footprint. The catalogue is whatever Apple has shipped or you have installed — including the premium "Enhanced" and "Premium" voices Apple offers for free in modern macOS.
  • Runs entirely on-device.
  • Zero cold-start time on first use.
  • Inherits any voice you have already set up system-wide.
  • Works offline. Always.

Best for: lowest friction, smallest disk footprint, or when you have already curated the macOS voice you like.

02 / Neural

FluidAudio voice library

16 hand-picked neural voices.

A curated library that ships through FluidAudio. The voices sound markedly more natural than the system catalogue and cover a range of accents and timbres — from soft and conversational to confident and upbeat. We update the library periodically as new voices land upstream, so we keep the exact roster off this page on purpose. The current count is below.
  • Runs entirely on-device after a one-time ~900 MB model download.
  • Pick any voice from the QuillMic Settings → Nika pane.
  • Each voice has a one-tap Preview button before you commit.
  • Catalogue is curated — not every upstream voice ships; we drop the ones that fail the quality bar.
16
Voices total
9
Female
7
Male

Best for: when you want Nika to sound like a person, not a phone tree.

Honest noteThe exact voice roster in the FluidAudio library is not enumerated here because it is maintained by the upstream library and is subject to change. The number of voices may go up over time. The picker inside QuillMic always reflects the live catalogue with a Preview button next to every entry — that is the source of truth.
Apple silicon nativeSandbox compliantHardened runtimeApp Store readyLocal Whisper · ParakeetOpenAI optionalKokoro neural voiceUp to 100 languages · model-dep.Apple silicon nativeSandbox compliantHardened runtimeApp Store readyLocal Whisper · ParakeetOpenAI optionalKokoro neural voiceUp to 100 languages · model-dep.
Available now

Stop typing what you already said.

Free during beta, free to keep your audio on-device, paid only if you ask for the cloud-quality polish. Built for the menu bar, not for our roadmap.

Download for macOSRead the manual

macOS 13 Ventura or later · 12 MB · code-signed & notarized