FluidVoice is a local-first dictation app that turns speech into polished text anywhere on your Mac, positioning itself as an open-source alternative to Wispr Flow. It pairs on-device speech recognition with its own custom-trained Fluid-1 models that clean up your words before they land in any app, and it has earned about 12,300 stars on GitHub with more than 350,000 downloads.
Most dictation tools send your voice to the cloud. FluidVoice keeps the whole pipeline on your machine by default: a speech model such as its Fluid Speech Mini transcribes locally, then a Fluid-1 model — available in full, Mini (under 1.5 GB) and Pico (600 MB) sizes — rewrites the raw transcript into finished text, removing filler, fixing phrasing and adapting tone per application. The model roster is broad: Parakeet variants, NVIDIA Nemotron Speech, Cohere Transcribe, Apple Speech and Whisper models are all supported, covering up to 40+ languages depending on the model chosen. Everything is one hotkey away in any app, and even a base M1 MacBook Air can run the lightweight model pairing in under 1 GB of memory. The Mac app is the flagship; Windows is in beta, an iPhone beta runs through TestFlight, and Android and Linux are on the waitlist.
FluidVoice is completely free and open source under the GPL-3.0 license — the project states ‘free forever, no tiers’. Optional cloud AI providers can be enabled for enhancement if you want them, at your own cost, but nothing in the app requires payment.
FluidVoice is for writers, developers, lawyers, finance teams and other privacy-conscious professionals who dictate a lot and want fast, polished text without sending their voice to a cloud service — especially Mac users looking for a free, local Wispr Flow alternative.