Skip to main content
A small on-screen indicator that shows SpeakoFlow is recording, so you never have to guess whether it heard you. It appears on the monitor your mouse cursor is on. All overlay settings live in Settings → Dictation → Dictation overlay.

Overlay style

A fresh install is on Auto, which is why you get the live card without configuring anything. Auto resolves to a concrete option based on the selected model. Picking any option writes that choice and overrides Auto. If Live is selected but the model cannot stream, the overlay falls back to Minimal. A non-streaming model never shows an empty live card.

Position

Six positions: Top, Top left, Top right, Bottom (default), Bottom left, Bottom right. A screen-map picker in the settings shows where each sits. Setting the position to None hides the overlay.

Keep finished text on screen

When using the Live card, you can choose how long it stays after a dictation finishes, so you can read or copy what was typed. Options: Briefly (1 second, the default), Standard (3 seconds), Longer (6 seconds), Much longer (10 seconds). Hovering over the card holds it open regardless of this setting. This row only appears while the Live card is the active style.

The compact pill

The minimal style shows a small pill. While recording, it displays a waveform. After you stop, a working indicator fills forward while the model transcribes. If AI cleanup runs, the fill continues for the cleanup pass. The fill is calibrated from this machine’s own history, not a fixed animation, so it tracks how long the work actually takes. There is no completion tick. A full, still row is the ending, because the pasted text is the real confirmation. The waveform freezes on that row while the pill fades. The pill holds briefly after the fill finishes so the run-out is visible, then fades.

The live card

The live card shows committed and tentative text in two shades as you speak. After the recording stops, it shows the final text for as long as the linger setting allows. You can scroll back through the text, select a phrase, and copy it. See Live transcription for which models stream and how cloud streaming works.

Stays on top, never takes focus

The overlay stays above other windows for the entire recording. It never takes keyboard focus from the app you are dictating into. On Windows this is WS_EX_NOACTIVATE; on macOS a non-key panel; on Wayland compositors with layer-shell support, a surface with no keyboard mode. When a conversation is active, the recording overlay rises above the conversation bar so it does not cover the controls.

Overlay states

A short notice appears when something did not work, for example “Flow needs an assistant model, pasted as dictation.” It auto-hides after a moment.
The overlay has no cancel button. To abandon a recording, press Esc (the default Cancel shortcut on Windows and macOS). See Hold or tap for more on cancelling.
On a native GNOME/Wayland session the overlay cannot stay above other windows. Mutter does not implement wlr-layer-shell, and Wayland gives apps no way to raise themselves.SpeakoFlow detects GNOME on Wayland and runs under XWayland, where “keep above” works. That is automatic and needs no setup. X11 and KDE/wlroots Wayland sessions already work.To force native Wayland anyway, launch with SPEAKOFLOW_ALLOW_WAYLAND=1. To disable GTK layer shell under a layer-shell compositor, use SPEAKOFLOW_NO_GTK_LAYER_SHELL=1. More in Troubleshooting.
Last modified on August 30, 2026