Every Ask Dockli input has a microphone button so you can speak instead of type. Speech is transcribed using Azure Speech, and — like the rest of Dockli’s cloud access — the Speech key never reaches your machine.
The microphone button
Click the mic to start; click it again to stop. What happens between depends on where you are:
| Mode | Where | Behavior |
|---|---|---|
| Dictation only | The search bar (search, not AI Mode) | Your speech is transcribed into the field and appended as you talk. Click the mic to stop, then send it yourself. The button’s tooltip reads “Dictate with your voice”. |
| Conversational | An Ask Dockli input (AI Mode or the folder-scoped panel) | Talk like you would to a voice assistant. After a short pause, Dockli submits what you said as a prompt, clears the field, and keeps listening for your next one. The button’s tooltip reads “Talk to Dockli AI”. |
In conversational mode, Dockli waits for about 1.6 seconds of silence after you stop speaking before it submits — long enough that a brief mid-thought pause won’t send early, but short enough to feel responsive. Each utterance goes out as a fresh prompt, so your follow-ups are never glued onto the message you just sent.
Listening states
The mic reflects exactly what’s happening, so you always know when it’s safe to speak:
- Connecting… — Dockli is setting up (fetching a token, opening the microphone). The input placeholder reads “Please wait…”. Don’t speak yet.
- Listening — speak now — capture is genuinely live. The input placeholder reads “Speak now…”. Go ahead and talk.
If something goes wrong, the mic’s tooltip surfaces the error, prefixed “Speech error:”.
Stop, not mic, while Dockli is working
While Ask Dockli is generating a response, the microphone temporarily becomes a Stop button so you can cancel the request. Once the answer finishes, it returns to a microphone. See Using the assistant.
How the speech connection works
Dockli is built so your Azure Speech key is never exposed in the app:
- When you start dictation, the browser layer asks your local Dockli service for a short-lived Speech token (from
/api/speech/token). - The service — which alone holds the Azure Speech key — mints a token that is valid for roughly 10 minutes and returns just that token plus the Azure region (eastus).
- Your microphone audio is streamed to Azure Speech using that temporary token. The key itself stays server-side.
The token is reused for a few minutes rather than re-fetched on every click, and the speech engine is pre-warmed when you move toward the mic, so starting dictation feels near-instant.
Microphone permission
Dockli runs inside a Microsoft Edge WebView2 host. That host is configured to auto-grant only the microphone permission and nothing else — so dictation works out of the box without a browser-style permission prompt, and no other device permission is silently granted. Dictation also requires a secure context, which Dockli provides.
If your microphone doesn’t start, the most common causes are a missing or disabled input device, or the OS-level microphone privacy setting blocking desktop apps.
Troubleshooting
If dictation won’t connect, keeps failing, or the mic button stays on “Connecting…”, see Troubleshooting AI and dictation for step-by-step fixes.