FluidVoice

FluidVoice is a free, open-source macOS dictation application developed by altic-dev. It converts speech into text, inserts text into other applications, and offers optional AI post-processing. It is designed for writing messages, documents and notes by voice, with additional Write and Command modes for rewriting selected text or automating supported Mac tasks.

Dictation and transcription

Choose a speech model, microphone and global hotkey, then dictate into the target text field. Live preview shows the recognized words before insertion. File transcription supports optional offline speaker labeling, timestamps and text or JSON export. A custom dictionary can handle vocabulary and spoken formatting. Check names, numbers and punctuation before sending or saving important text; recognition accuracy and speed depend on the recording and model.

Models, hardware and installation

FluidVoice 1.6.9 requires macOS 15 Sequoia or later. The official project supports Apple Silicon and Intel Macs, with different model choices. Parakeet and the newer Apple Silicon-specific engines require Apple Silicon; Intel support uses Whisper. Apple Speech is also listed for both architectures, with language availability dependent on the system.

The application DMG is 49,299,078 bytes before installed storage. Copy FluidVoice.app to Applications, grant the required permissions and select a model during onboarding. Speech-model downloads are additional to the app: the documented choices range from roughly 75 MB to 2.9 GB, and the optional Fluid Intelligence model requires approximately 3.5 GB of disk space. Allow sufficient storage and choose a model appropriate to your hardware and language. Model downloads require internet access; supported downloaded local models can then perform transcription on the Mac.

The packaged app uses native speech libraries; its documented installation does not require a separate Python environment. Building from source uses Xcode and Swift Package Manager. The catalog declares no additional package dependency or conflict and marks the app as self-updating.

Permissions and automation

Microphone permission allows voice capture. Accessibility permission supports global hotkeys and insertion into other applications. The source also declares speech-recognition access and Apple Events access for managing launch-at-startup settings. Follow the prompt for the selected feature; Input Monitoring is not listed as a general installation requirement.

Write mode can replace selected text, while Command mode can launch apps, invoke shortcuts or perform other actions. Review what a command will change and use trusted enhancement instructions. Accessibility access is broader than simply recording audio, and transcripts can include sensitive information. Optional local audio history creates recordings on disk; review its storage budget and export or retention settings.

Local and cloud processing

Local speech engines and optional local enhancement provide an on-device workflow. Choosing OpenAI, Groq or a custom cloud enhancement provider sends the relevant content to that service and requires its credentials and terms. Provider API keys are stored through macOS Keychain. Cloud-provider usage may incur separate charges; a paid provider account is not required for the core local dictation workflow.

The project describes local-first processing, but that does not mean the app makes no network connections: downloads, updates, optional cloud services and analytics are distinct functions. Its privacy documentation says anonymous app-health and feature-usage analytics are enabled by default and can be disabled under Share Anonymous Analytics. It states that those analytics exclude audio, transcripts, selected text and prompts. Choose settings appropriate for the information you dictate and the policies of any connected provider.

License and cost

The FluidVoice source is GPLv3 from February 23, 2026 onward; earlier versions used Apache License 2.0. The core application is free, with optional sponsorship. Fluid Intelligence is a separately maintained private local runtime, so the application's open-source license should not be assumed to cover that runtime or every downloaded model. Model and cloud-service terms are separate. No FluidVoice account or paid subscription is specified for core local dictation.

Official sources: FluidVoice, Version 1.6.9 documentation, License, Releases, Whisper integration.