Recognition stays local
Local inference avoids a cloud transcription dependency. Optional update checks are separate from recognition.
LOCAL AI / preview
Fixo Flow
Your voice, in the apps you already use.
The problem.
Typing interrupts the flow of an idea. Fixo Flow brings English dictation to everyday applications while keeping recognition on the computer.
My contribution.
Built the local speech workflow and desktop product: microphone capture, recognition, formatting, history, OS insertion and distribution paths.
How the parts connect.
Capture
Bounded microphone chunks enter the local recognition pipeline.
Local inference
Whisper recognition uses MLX on Mac and a CPU engine in other platform previews.
Application integration
Dictionary and offline formatting feed history, clipboard and attempted paste.
The decisions.
Recover the words
History and clipboard provide a recovery path when insertion into another application fails.
Build, test, deliver.
Recognition pipeline, privacy model and preview limitations.
Configured installer builds and checks, not a claim that a current CI run passed.
Published packages and platform requirements.
What to keep in mind.
Experimental English dictation preview; multilingual and Hinglish recognition are not delivered in the reviewed release.
Accuracy depends on speech, microphone and hardware. OS support and publisher signing have platform-specific limitations.