VoiceInk

VoiceInk is Pax's native Mac dictation application for converting speech into text at the cursor. It combines selectable transcription models, global recording shortcuts, a personal dictionary and optional AI text enhancement. It suits writers, developers and people who prefer speaking to typing; recognition quality depends on the model, language, microphone and surrounding noise.

Features

• Start recordings with configurable keyboard or mouse shortcuts and push-to-talk.
• Choose supported local speech models or explicitly configured online providers.
• Insert transcribed text into the active application using Accessibility access.
• Apply personal dictionary terms, replacements and dictionary-learning tools.
• Switch app- or context-specific modes and duplicate mode configurations.
• Review recent transcriptions and optionally enhance text with an AI provider.

Local model availability and language coverage differ. Download the appropriate model before expecting offline transcription; the app installer does not mean every model weight is already present. Transcription models and enhancement models perform different jobs and have separate capabilities and requirements.

Installation, systems and permissions

Install the official release DMG or use brew install --cask voiceink, place VoiceInk.app in Applications and complete onboarding. Current repository requirements and the cask specify macOS 15 or later. The marketing site still says macOS 14.4; use the current repository/cask baseline for this distribution rather than promising current-version support on 14.4. The publisher's site describes Apple Silicon-only support and recommends 8 GB RAM. Intel support is not claimed from the single generic DMG URL. Windows and Linux packages are not provided by the reviewed official distribution; the separately advertised iOS product is not this Mac cask.

The current official onboarding source marks Microphone and Accessibility as required: microphone records speech, while Accessibility lets VoiceInk type into other applications. Screen Recording is a separate permission for reading visible screen context to improve transcripts, and is not in the required permission list. Grant only the permissions needed for the features you intend to use. Screen context can contain unrelated private information, so be deliberate about which windows are visible when using contextual features.

Models, language runtimes or external CLI tools can have separate requirements. The source integrates whisper.cpp, FluidAudio and TranscribeCpp; these are project components, not instructions to install arbitrary Homebrew dependencies. Dictionary Auto Learn's local CLI option needs the configured compatible CLI environment; selecting a feature does not install every external provider or tool automatically.

Privacy and model providers

Local transcription models process audio on the Mac after model downloads. Optional cloud enhancement can send transcribed text and configured context to its provider. Recent releases also add cloud transcription providers, including OpenRouter and Grok, so the site's blanket claim that all voice transcription stays local is not accurate for every selectable workflow. Online transcription can send audio to the chosen service; review its credentials, network use, retention policy and costs before enabling it.

AI enhancement can rewrite meaning or produce mistakes. Inspect important text before sending it, especially in messages, terminal windows or forms. The global Auto Send option can automatically deliver Return, which may submit the active application's content; choose that behavior carefully. Dictation history and recordings can contain sensitive material, so review retention settings and keep the Mac account protected. Do not assume the app is an ephemeral-only fork that deletes every transcript automatically.

The microphone, Accessibility and screen-context capabilities give the app access beyond a plain text editor. Use trusted builds and providers, keep secrets out of prompts and review optional integrations. Open source makes inspection possible but does not guarantee a model's accuracy or a provider's data policy. Models such as SenseVoice have their own license terms separate from VoiceInk's application source license.

Trial, purchase and license

The official downloadable application offers a free trial and a paid one-time lifetime license, rather than a recurring VoiceInk subscription. The current website advertises use on up to three macOS devices; check current regional prices and license terms before purchasing. A license key is supplied after purchase, and the publisher describes automatic updates and priority support as benefits of the official compiled version.

The source is available under GNU GPL v3 and can be built following the official build guide. The open-source license and self-building option do not imply that the supported official binary is permanently free after its trial. Online model services can require their own accounts or API credentials and charge separately. Local model use does not require buying an unrelated AI-provider subscription merely to transcribe offline.

The official 2.21 DMG is 51,053,559 bytes according to HTTP HEAD. Installed files, downloaded model weights, history and recordings require additional storage.

Sources: Official source and requirements, Publisher product and purchase information, Exact releases, Onboarding permission source, Source license.