Dictate with a mouse button on Linux

Bulbul mouse mode gives Linux users a built-in mouse trigger instead of a keyboard chord or an external button-remapping utility. Click once to record, speak, then click again to stop. Bulbul transcribes and cleans up your words, then inserts the result into the focused app. Middle-click is the default.

This is a recording trigger, not voice control of your pointer. It does not move the cursor, click targets by voice, or turn speech into mouse commands. Its job is to keep the dictation control on the device already under your hand.

The mouse trigger is independent of your keyboard hotkey. You do not need to map a mouse button to a keyboard combination to use Bulbul mouse mode.

Install the Linux build for your system: .deb for Debian, Ubuntu, or Mint; .rpm for Fedora or openSUSE; or AppImage. Then turn on Mouse mode in the desktop sidebar. To change the trigger, open Settings > Hotkeys and find Mouse button.

The .deb and .rpm installers configure Linux input access. An AppImage may need manual uinput access, especially on Wayland. Follow the current platform docs for your installed version rather than assuming every distribution is configured identically.

The current source confirms these control names. Labels and packaged behavior can differ between releases. If setup fails, check microphone access and input permissions separately; a working keyboard shortcut does not prove the mouse hook is ready.

The current Bulbul button recorder offers Middle click, Side button (back), and Side button (forward). It does not offer ordinary left or right click.

Middle-click is the default, so you do not need a mouse with extra side buttons. A chosen side button can replace its normal job while mouse mode intercepts it. If you rely on browser Back or Forward, choose another button or switch mouse mode off when you need that action.

A trackball or pen device can suit this workflow if it exposes a supported button, but not every device is certified. A trackpad gesture is not automatically a middle-click or a supported mouse button.

On Linux, test the button in your installed build. If the virtual mouse cannot be used, Bulbul can observe the click without suppressing its normal action.

Bulbul mouse mode is click-to-toggle: one click starts recording and the next stops it. Holding the mouse button is not the documented mouse-mode behavior.

The keyboard offers separate hold-to-talk and tap-to-talk options. Do not confuse those keyboard modes with the mouse trigger. The current desktop mouse-hook source routes button presses as toggle actions.

You can speak without keeping a button pressed. You still need the second click to stop the recording; mouse mode is not an always-listening voice activation system.

Bulbul documents Linux support on both X11 and Wayland. Its current Linux mouse implementation reads button events through evdev and mirrors other mouse events through a uinput virtual device.

This input-device approach is separate from binding a shortcut inside an X11 or Wayland desktop. Device permissions and the installed release still matter. Support for both display systems is not a guarantee that every distribution, compositor, mouse, or old build works without setup.

If the virtual mirror or exclusive input access fails, the current code falls back to observing clicks without suppressing them. The chosen button can then also reach the foreground app: middle-click may paste or open a link, and side buttons may navigate Back or Forward. Test in a harmless field before relying on mouse mode.

Basic dictation support does not promise identical per-app integration on every compositor. Bulbul also has mouse mode on Windows and macOS; Android uses a separate floating bubble, not the desktop mouse control.

Use it when your hand is already on the mouse and reaching for a keyboard shortcut interrupts the task.

These are workflow examples, not measured productivity gains or medical claims. Dictation can still make mistakes. Names, commands, punctuation, and sensitive text deserve a review before you use the result.

Bulbul is the only app in this comparison with built-in mouse-button dictation on Linux. Other apps have mouse triggers on different platforms, so this is a Linux-specific distinction, not a claim that Bulbul invented mouse dictation.

Wispr Flow has no Linux app in this check. OpenWhispr does support Linux, but its reviewed mouse-hotkey validator rejects mouse bindings outside macOS. FluidVoice is macOS-only. Do not confuse a cross-platform app with a cross-platform mouse-trigger feature.

Handy also documents external CLI and Unix-signal triggers. A mouse remapped to a keyboard shortcut or an external command may start dictation, but that is not a built-in mouse binding. This comparison does not claim there is no possible mouse-driven workaround.

This is a dated comparison of the named apps, not an exhaustive survey of every voice-typing project. Development source can be ahead of packaged releases; check the version you install.

Yes, if it meets Bulbul's platform requirements and has a microphone and internet connection. No local speech-model download or GPU is needed for transcription: the mouse button controls recording, while Groq runs speech recognition and text cleanup using your API key.

Bulbul is a thin client for hosted speech inference. It records audio, sends it directly to Groq, and inserts the returned text. No Bulbul-owned transcription server sits between your device and Groq. History, dictionary, snippets, and settings live in a local SQLite database.

The speech-model workload runs on Groq rather than your laptop. That makes the workflow practical on supported cheap or older Linux hardware without setting up local Whisper or Parakeet. It does not promise that every machine or connection gives the same latency.

Your device still runs the interface, records audio, and uses the network. There is no zero-RAM or zero-battery claim, and no measured battery comparison here. Internet and a Groq key are required. Bulbul is free software; Groq free-tier limits or paid usage still apply. Transcription and cleanup are separate requests.

If offline speech or keeping audio entirely on-device matters more, use a local workflow such as Handy or OpenWhispr with local models and cloud cleanup disabled. Bulbul mouse mode does not make its hosted speech pipeline offline.

Mouse mode uses the same hosted pipeline as keyboard dictation. Audio leaves your device for Groq transcription, and text goes to Groq for cleanup. Changing the trigger does not change that privacy boundary.

Section 4.2 of the Groq Services Agreement says: "Groq is not permitted to use Inputs or Outputs for training or fine-tuning any AI Model Services or other models, unless explicitly granted permission or instructed by Customer."

Groq says inference content is not retained by default. Reliability and abuse-monitoring exceptions can retain content for up to 30 days, or longer if legally required. Usage metadata is collected and does not include customer inputs or outputs.

Groq's data documentation says all customers may enable Zero Data Retention (ZDR) in Data Controls to stop retention for reliability and abuse monitoring. The agreement refers to eligible customers. Review your own account settings and current terms; Bulbul does not automatically enable ZDR.

The Bulbul README discloses anonymous usage statistics on by default, with an off-toggle in Settings > Privacy. It says these exclude audio and transcripts. Local history storage and a no-training rule do not make the speech pipeline offline or remove every form of data collection.

If your rule is that audio must never leave the device, choose a fully local speech workflow and keep any cloud cleanup disabled. Mouse-button convenience is not a reason to ignore that requirement.

Product controls, Linux input behavior, competitor platforms and mouse triggers, and provider policy were checked against official documentation and repository source on October 9, 2026. Hardware behavior was not tested across every Linux distribution or device.