Barge-in and no user input
Two speech events shape how naturally the agent takes turns with the user:
- Barge-in lets the user interrupt the agent while it is speaking. The agent stops playing its response as soon as the user starts talking and turns to what they say, instead of making them wait for the response to finish.
- No user input is an event the Bot connection generates when the user stays silent for a configured time. The agent can respond to it – asking whether the user is still there, or ending the call – instead of leaving the conversation in silence. On speech-to-speech models, use the Inactivity reminder instead.
Both are described in Speech events – see Barge-in and No user input.
Inactivity reminder
Note: Inactivity reminders apply only to speech-to-speech models. For regular (non-speech-to-speech) models, use the No user input event instead.
Speech-to-speech models do not natively handle "user goes silent" scenarios – when the user stops talking, the model simply keeps waiting. The inactivity_reminder advanced configuration parameter detects user silence and sends a reminder text to the model, prompting it to re-engage the user.
This is currently supported by:
- OpenAI Realtime models –
gpt-realtime,gpt-realtime-mini,gpt-realtime-1.5 - Gemini audio models –
gemini-2.5-flash-native-audio,gemini-3.1-flash-live,gemini-3.8-live,gemini-3.8-live-extended-thinking– only when Silero VAD is enabled (gemini_audio.vad_modeset tosilero, as described in VAD mode for Gemini models)
If the user remains silent for the configured timeout, the reminder text is sent to the model. After the model responds, the timer re-arms and may fire again for the same silent period, up to repeat times. The counter resets once the user speaks.
| Parameter | Type | Description |
|---|---|---|
timeout
|
int | Duration of user silence, in milliseconds, after which the reminder text is sent to the model. |
text
|
str | Reminder text sent to the model. Optional, defaults to Note: user has been silent for {silence_time} seconds.Supports the variables listed below. |
repeat
|
int | Maximum number of reminders sent during a single silence period. Optional, default is 3, maximum is 10. |
The text parameter supports the following variables:
{count}– number of inactivity events fired during the current silent period (resets when the user speaks){timeout}– the configured timeout, in seconds (with up to one decimal digit, e.g.2or2.4){silence_time}– elapsed time since the agent finished playing its last response, in seconds
For example:
{
"inactivity_reminder": {
"timeout": 5000,
"text": "NOTE: User has been silent for {silence_time} seconds. Please re-engage them."
}
}
Note: This is distinct from the flow-level no_user_input_repeat parameter, which repeats the last response on NO-USER-INPUT events for textual (non-speech-to-speech) agents. See Handling user inactivity.