Troubleshoot Voice Control
Cannot reach Model Vista
Section titled “Cannot reach Model Vista”For Cannot reach Model Vista. Start Model Vista and enable AI (MCP)., start Model Vista on this computer and enable AI connection enabled under Settings > External Connections. Check that the MCP port is 8750, or matches your configured endpoint. Select Connect Model Vista again.
Enabling Web Remote alone does not enable MCP. After changing a custom endpoint, fully close and reopen Voice Control. See Install and connect.
A command is missing
Section titled “A command is missing”Update Model Vista to a preview build containing the required MCP tools, then reconnect Voice Control and start a new voice session. Updating only the companion does not add APIs to an older Model Vista build.
Check the tool count after connecting. Ask the assistant to list the relevant models, states, windows or supported settings before choosing an operation. Features may also be unavailable because of the current rendering configuration, XR mode or license.
Voice will not start
Section titled “Voice will not start”- Check API key settings and API account access to the selected mode’s models.
- Check internet access, Windows microphone permissions and the selected microphone.
- Allow microphone access when prompted. Check speakers or headphones if replies are silent.
- If startup times out or the connection is lost, start a new session after resolving the cause.
The installed window requires Microsoft Edge WebView2 Runtime. If the app reports that it is missing, install the runtime and reopen the app.
Speech uses the wrong language
Section titled “Speech uses the wrong language”End the session, choose English or Japanese under Spoken language, then start again. Changing EN / 日本語 changes the interface only. Use short, clear requests and reduce background speech during GPT-Live sessions.
The assistant says an action succeeded, but nothing changed
Section titled “The assistant says an action succeeded, but nothing changed”Enable Show actions and tool results and inspect the result at the corresponding timestamp. Check the actual scene. If no action was issued, ask explicitly for the operation again. If an action timed out, inspect the current state before retrying.
For repeated model names, ask for loaded instance IDs. For repeated state option names, specify both the group and the option ID. After reloading a model, list its state groups again because IDs may have changed.
For animation, ask whether playback is advancing. A selected animation option does not prove visible motion. For camera movement, exit XR first. For scrolling, confirm that the intended window is visible and that its content exceeds the available space.
An idle session stays connected
Section titled “An idle session stays connected”Confirm that Mode is GPT-Live conversation and Auto-disconnect is not Never. Recognized background speech can reset the timer. Muting does not pause it, but background scheduling can delay the callback. Select End session when you want to disconnect immediately.
Information for a support report
Section titled “Information for a support report”Record the Voice Control and Model Vista versions, selected voice mode and language, the request, its timestamp and the relevant action/error result. Mention the visible window or selected model where relevant. Do not include API keys; review logs for private model names, paths or other sensitive project data before sharing them.
