Dictation for Linux that runs entirely on your machine

The mainstream dictation market is Mac-first. Several popular apps ship macOS and Windows builds and treat Linux as a later platform, if they support it at all. Linux also has no single input API: X11, Wayland, different desktop environments, and different text-input tools all change how an app can insert text.

Cloud dictation sidesteps input complexity by pasting into the clipboard. Local dictation has to solve the last mile properly — or it becomes a clipboard tool with extra steps.

The checklist is short, but most apps fail at least one item.

• Local recognition by default, not a cloud engine with a local option hidden in settings.
• A narrow, predictable text channel: characters and Backspace rather than synthetic shortcuts.
• X11 and Wayland support, with honest documentation about limitations.
• Native packages — AppImage, .deb, .rpm — instead of an unmaintained script.
• CPU-friendly streaming so you do not need a discrete GPU.
• No account, no subscription, no telemetry.

Saype ships as an AppImage, .deb, and .rpm for 64-bit Linux. It uses NVIDIA Nemotron 3.5 ASR Streaming on the CPU, so words stream while you speak, and it never uploads audio. The Whisper mode handles harder audio offline.

Typing works in X11 and Wayland sessions. On pure Wayland, text injection is best-effort because the compositor controls input; Saype fails safe rather than firing arbitrary key events when focus cannot be verified.

The blind typer emits only transcript characters and Backspace. That makes it usable in terminals and secure environments where a general-purpose automation tool would be unacceptable.

Handy is the strongest free option: MIT licensed, offline with Whisper or Parakeet, and available for Linux. It is a good choice if you want open source and do not mind choosing models.

OpenWhispr is another open-source cross-platform option with local Whisper or Parakeet models and an optional hosted plan. Vocalinux focuses specifically on Linux, with whisper.cpp and several engines, and supports X11 and Wayland.

Compared with those, Saype is not open source and is paid after the beta. The trade-off is a focused streaming experience, a narrow blind-typer input path, and one hotkey flow across Linux, macOS, and Windows.

Installation is the same on Debian, Ubuntu, Fedora, and most other distributions.

• Download the AppImage, .deb, or .rpm for your distribution.
• Make the AppImage executable or install the package with your package manager.
• Launch Saype, download the voice model once, and set your hotkey.
• Dictate into any app — no root service, no account, no cloud.

Shell history, incident notes, and credentials can all pass through dictation. Local recognition keeps audio on the machine; the blind typer keeps the app from emitting anything except your words and Backspace. If you work in a regulated environment, test the exact build and review the privacy documentation before rolling it out.