figlet molvi
███╗ ███╗ ██████╗ ██╗ ██╗ ██╗██╗
████╗ ████║██╔═══██╗██║ ██║ ██║██║
██╔████╔██║██║ ██║██║ ██║ ██║██║
██║╚██╔╝██║██║ ██║██║ ╚██╗ ██╔╝██║
██║ ╚═╝ ██║╚██████╔╝███████╗╚████╔╝ ██║
╚═╝ ╚═╝ ╚═════╝ ╚══════╝ ╚═══╝ ╚═╝
molvi v1.3.0 — local voice input for Windows
# hold a key → speak → text appears. whisper large-v3, 100% on your own hardware
asciinema play demo.cast
rec 0:03
>
man molvi | head
- NAME
- molvi — you speak, not type
- SYNOPSIS
- hold Ctrl (configurable) → speak → release
- ENGINE
- faster-whisper / large-v3 on CUDA; CPU fallback to base
- NETWORK
- zero bytes out. model downloads once, then fully offline
- TELEMETRY
- does not exist, period
- LICENSE
- MIT, source on github
cat features.txt
- types into the terminal — detects the console and sends char by char, where Ctrl+V is dead
- perfect for Claude Code / Codex: a long prompt in seconds, by voice
- live equalizer: the wave moves with your voice — you can see the mic is listening
- any hotkey, any microphone, ru/en/auto
- overlay lands on the monitor with the active window (multi-monitor done right)
- autostart in the tray, ~0% CPU while you're silent
diff --side-by-side wispr-flow molvi
wispr-flow / aqua-voice
- $12–15/mo, subscription
- audio goes to the cloud
- doesn't work offline
- closed source — take their word for it
cat INSTALL.md
- download the installer (~70 MB) — button below
- SmartScreen: More info → Run anyway # no signature — but there's source code
- the wizard finds your GPU on its own and downloads the model (
large-v3 ≈ 3 GB on NVIDIA 6+GB, otherwise base)
- hold
Ctrl → speak → done
▼ Molvi-Setup.exe
windows 10/11 x64 · full guide with screenshots
molvi --donate
free, no ads, no telemetry. if it saved you time — buy me a coffee
# tribute, one-time or monthly
♥ donate via telegram
or pay by card on the web: web.tribute.tg/d/Nwy