FlowLite
Rust · Tauri · Desktop development · 2026
A free, local voice-to-text app for Windows. Hold a key, speak, and the transcript is typed into whichever app has focus, with Whisper running on your own machine.
Status: ACTIVE DEVELOPMENT
Real FlowLite v0.3.0 interface rendered at 1100 × 720 with the backend mocked: all history, word counts, streaks and settings shown are fictional demo data, not my dictations. Live dictation is not shown.
Problem
Cloud dictation tools meter words and send audio to servers. I wanted push-to-talk dictation that works in any text field, stays private and has no subscription or word limit.
Approach
A Rust backend handles hotkeys, microphone capture and clipboard-paste insertion, and runs the prebuilt whisper.cpp CLI as a subprocess. A React and Tailwind interface inside a Tauri v2 shell manages settings, history and insights.
Features
- Push-to-talk and toggle modes — Hold Ctrl+Win or tap F9 to dictate into any application; Esc cancels a recording.
- Local speech models — Six Whisper model sizes, an optional CUDA engine for NVIDIA GPUs, and no audio leaving the machine during use.
- Transcript post-processing — Custom dictionary, find/replace rules, voice commands such as "new line", and optional filler-word cleanup.
- History and insights — Local dictation history with words-per-minute, totals, a streak and an activity heatmap.
- Optional local AI rewrite — A small language model run through llama.cpp can reformat transcripts after a one-time download.
Outcome
Version 0.3.0 works end to end on my own Windows PC. It is unsigned and has not been tested widely across other machines, microphones or applications.
Stack
Rust, Tauri v2, React, TypeScript, Tailwind CSS, whisper.cpp
Links
GitHub
All projects · Harshil Patel