FluidVoice

FluidVoice is a free macOS dictation app that keeps everything local. Speech is transcribed on-device, text lands in whatever app you're focused on, and no audio or transcribed content leaves your Mac. It's a strong pick if you've looked at tools like VoiceInk or OpenWispr and want something with a built-in AI polish layer that still runs offline.

The app pairs with Fluid-1, an optional local AI model that post-processes raw dictation. It cleans up filler words, fixes capitalization and formatting, handles dates and numbers, and adjusts tone based on which app is active. Slack gets casual language. Mail gets formal phrasing. GitHub issues get structured output. Same voice, different register, no manual editing.

Key capabilities:

• System-wide input – one hotkey activates dictation in any text field: terminals, code editors, chat apps, browsers
• Multiple speech models – Nemotron Speech 3.5, Parakeet TDT v2/v3, Apple Speech, Whisper (Tiny through Large), and others; model choice affects language support and speed
• 40+ languages – coverage varies by model, with Whisper reaching up to 99 languages at larger sizes
• Three dictation modes – Write Mode, Command Mode, and Direct Dictation for different use cases
• AI post-processing options – Fluid-1 locally, or optionally route to OpenAI, Groq, or a custom provider
• Apple Silicon optimized – CoreML and Metal keep latency under 100ms and preserve battery
• Long-form sessions – handles extended dictation without manual chunking

For those browsing the input and dictation space, FluidVoice stands out because it doesn't force a choice between privacy and polish. Most open-source dictation tools stop at raw transcription. Most polished commercial apps send data to the cloud. FluidVoice does both locally, with the Fluid-1 model as an optional download around 3.5 GB.

It's free forever, GPLv3 licensed, and runs on both Apple Silicon and Intel Macs (macOS 15.0 required).