On-Device Dictation for Linux: Local Voice-to-Text on the Desktop

Local on Linux is the same architecture

Speech recognition runs on the machine. Raw voice audio is not uploaded for transcription. Hardware-tier model selection picks a local engine that fits the box you have. The pipeline is documented in how on-device dictation works.

Account status, billing, consented model downloads, and optional support still use the network when you use those features. That split is the same as Trust & Privacy.

Handy, nerd-dictation, and packaged apps

Handy is free, MIT-licensed, and genuinely local on Linux. It is the right pick if you want an open-source tool you can fork. Wayland insertion can need extra setup depending on the compositor.

nerd-dictation and whisper.cpp remain good if you want a script you control. Spokenly now ships a Linux build with local models and an optional cloud path. Wispr Flow and Superwhisper do not ship Linux.

VeloxWaves is the packaged private assistant: local dictation plus a second brain, not a CLI you assemble. Choose it when you want that product on the Linux box you already use for work.

The loop

• Install the Linux desktop build from the download section.
• Grant microphone access for the session.
• Hold the shortcut, speak, release. Text lands in the focused field.
• Transcripts can join the local knowledge graph, same as on Windows and Mac.