The questions people actually ask.

QuillMic itself is free during the current beta and free to use entirely on-device after that. The only thing that costs money is bringing your own OpenAI or OpenRouter API key for the optional cloud features — and those are off by default.

No. There is no QuillMic account. The app works the moment you finish onboarding. The only credentials it ever asks for are your own third-party API keys, and only if you decide to enable the optional cloud tier.

Only for cloud features (cloud speech-to-text, the Nika assistant, cloud transcript polish) and for the first download of an additional local model. With the bundled Whisper base.en model that ships in the installation, dictation works offline from the very first launch.

Drag the app to the Trash. The container at ~/Library/Containers/com.aktica.QuillMic/ holds your settings and downloaded models — delete that folder if you want every trace gone, or keep it for a future reinstall. Keychain entries clean up automatically when the container is removed.

No. Dictation history is held only in memory for the current session and is discarded when the app quits. Audio buffers live in volatile memory only for the duration of one utterance, then are released. We do not write a transcript log to disk.

Never to ours — we run no servers that touch your content. If, and only if, you turn on an Optional Cloud Service and enable the corresponding feature, then audio or text travels directly from your Mac to the third-party provider you have chosen, over TLS. We do not proxy, mirror, log, or retain any of it.

In the macOS Keychain, with the WhenUnlockedThisDeviceOnly accessibility class. Nothing reads them while your screen is locked, and they never leave the Keychain except to attach to a request you have explicitly authorised by enabling a cloud feature.

Yes. Settings → Diagnostics surfaces a recent activity log, and presets that depend on a cloud service are clearly labelled in the UI. There is no silent path that flips cloud on without you doing it. The full local-vs-cloud breakdown lives on our Security page.

It depends on the model you select. The bundled Whisper base.en is English-only. Multilingual Whisper variants and FluidAudio Parakeet add European languages. Cloud Whisper-1 covers the entire Whisper language matrix — close to one hundred languages in total, depending on the model you load.

Better than off-the-shelf dictation tools, but model-dependent. On the Best preset (FluidAudio Parakeet on a 16 GB+ Apple silicon Mac, or gpt-4o-transcribe in the cloud), QuillMic handles non-native English and regional accents that lighter tools tend to fail on. On Light or Balanced, accuracy on heavy accents falls off — switch up a tier if you notice it. Adding the names and code-switched words you actually use to the custom vocabulary makes a measurable difference at any tier.

Yes — Settings → Language & Automation → Speech substitutions. Add the heard form ("open ai") and the inserted form ("OpenAI"). Substitutions can be scoped per app, applied as exact match, word boundary, or regex, and learned automatically when you correct the same word twice.

Mixed-language utterances are preserved as-is by the recognizer; the deterministic cleanup is English-tuned and the LLM polish step is skipped automatically when mixed scripts are detected, so it does not try to anglicise what you said.

Apple silicon Macs (M1 and later) running macOS 13 Ventura or later. We do not support Intel Macs.

Anywhere from 8 GB up. The Light preset runs comfortably on 8 GB. Balanced is ideal at 16 GB. The Best preset (the largest local models) needs 16 GB free RAM at minimum, and we recommend 24 GB+ for the largest Whisper variants. If you choose Best on a smaller Mac, Simple Setup downgrades automatically and tells you why.

The bundled installation is small. Optional models add to that — Whisper variants range from 75 MB (tiny) to about 1.6 GB (large-v3-turbo), Parakeet is around 600 MB, Kokoro neural voices are around 900 MB. Settings → Models shows the exact disk and RAM footprint of every model with one-click install and unload.

No. It runs as a single menu-bar agent with no Dock icon. Quit it from the menu bar and nothing remains running.

QuillMic does not bill you. You bring your own OpenAI or OpenRouter API key, the request goes from your Mac directly to that provider, and the provider bills you under their own pricing. We never see, charge for, or take a cut of those calls.

Yes. Settings → Nika → Daily spend cap (USD). Set a positive number and Nika refuses cloud calls once today's recorded spend hits it. The cap is enforced from QuillMic's local cost ledger — no provider-side billing API needed. Set 0 to disable.

Settings → Usage shows daily / weekly / monthly rollups with a per-model breakdown. The provider's billing dashboard is the source of truth, but the local ledger is more than enough for day-to-day awareness.

It is a toggle. Local mode keeps working unchanged whether or not you also have cloud configured, and disabling cloud anywhere returns the app to local-only on the next dictation.

Yes. Nika is the only feature that always requires an LLM call to a remote provider, so it needs either an OpenRouter or an OpenAI key. Without one, Nika refuses to run and surfaces a clear error.

Draft a reply to the email you just read, reshape an awkward paragraph in your own voice, or pass a request to an external model for the news, a stock quote, or a quick fact — and speak the answer back without leaving the focused field. She also has a small set of built-in commands (open settings, control Apple Music, free local model memory) and can run any HTTPS webhook you wire up.

Only what you explicitly allow. Two privacy toggles control this: "Allow selected text" lets her see the text you have selected (used for "rewrite this"), and "Allow app context" lets her see the focused application name and editable-field state (so replies fit the app you are in). Both default off.

No. Her responses always require a remote LLM call. Local dictation, local transcript polish, and local TTS all work offline; the assistant does not.

Settings persist locally in UserDefaults under com.aktica.QuillMic. iCloud sync is not enabled in the Mac App Store release. Custom vocabulary, automation actions, and command catalogs are file-based under Application Support and can be copied between Macs manually.

Not in the Mac App Store release. The dictation pipeline is real-time only — microphone in, focused field out.

Not at this time. QuillMic is built for the macOS menu bar; an iOS port would be a different product, not a build target.

Use the contact form on this site. Include reproduction steps; if you can attach the redacted log file from Settings → Diagnostics → Export logs, that helps a lot. Sensitive content (transcripts, keys, file paths) is stripped automatically before export.