FAQ

FAQ

Is my voice sent anywhere? No. Transcription runs entirely on your computer (faster-whisper). After the one-time model download, AgentWhisper works fully offline. Nothing you dictate ever leaves your machine — dictated text isn't even written to the log.

Which languages are supported? English, currently. The architecture is prepared for more languages (multilingual Whisper models); it's on the roadmap.

Does it work on Wayland? Not yet — X11 only. All display-server code sits behind one interface, so Wayland support is a planned module, not a rewrite. Check your session with echo $XDG_SESSION_TYPE.

Why does F12 stop working in other apps while AgentWhisper runs? By design: AgentWhisper reserves the key exclusively so dictation never collides with an app's own F12 action. Combinations (Ctrl+F12, Alt+F12) are untouched. Quit AgentWhisper and F12 is instantly released. You can pick a different key in the Configuration.

Where are the speech models stored? ~/.cache/huggingface/ — the standard shared cache. If another Whisper tool already downloaded a model, AgentWhisper reuses it (no second download).

How much CPU/RAM does it use? Idle: negligible. While transcribing: a burst of CPU for roughly 1–3 seconds per sentence with the default base.en model. Bigger models are more accurate but slower — see Configuration.

Can it start automatically when I log in? Not yet — autostart is part of the next milestone. Until then, start it from the applications menu.

Why "Agent"Whisper? The long-term vision includes an optional agent mode — speaking instructions to an AI assistant instead of dictating text. The architecture has a designed-in seam for it; today the app is a focused dictation tool.

Something's broken and it's not in Troubleshooting. Open a GitHub issue with the tail of ~/.local/state/agentwhisper/daemon.log.