What is Silence Tolerance?
During a conversation, users may pause for various reasons:- Thinking about their response
- Distracted by something
- Unsure how to proceed
- Technical issues with the call
Key Timers
How It Works
The
Silence Tolerance property on each step controls silence tolerance behavior:
When to Use Each Setting
Hurried (Default)
Best for: almost every step — the recommended starting point, including questions that need a moment’s thought (that is Response Eagerness, not silence tolerance). Example: “Is Tuesday at 2pm okay?”Patient
Best for: the assistant has asked the caller to go and find something. Example: “Can you grab your policy number?”Extremely Patient
Best for: the assistant has asked the caller to do something and report back when it’s done. Examples: “Reply to the text I’ve just sent and let me know when it’s done”, “Can you check the serial number on the back of the device and read it out?”Infinite
Best for:- The assistant is waiting on something outside the conversation: the caller is on hold, a third party is being fetched to the phone, or the caller has asked the assistant to hold while they step away
Silence Prompts
When the response delay expires, the agent prompts the user to re-engage:idle_offset(seconds).
The re-engagement prompt fires once per quiet period. If it is cancelled before the caller ever hears it — because the assistant was interrupted, or the call was busy doing something else — it is re-armed and spoken as soon as it can be, rather than being counted as already said. Silence tolerance is applied per step, so each step waits for as long as its own setting says.
Interaction with Other Settings
Silence tolerance works alongside other dynamics:Best Practices
- Configure per step — Set silence tolerance on each step based on the expected response
- Match tolerance to the task, not the question — Longer tolerance is for callers who have gone to fetch something or do something; a hard question needs a slower Response Eagerness, not a longer silence timer
- Consider your audience — Elderly users may need more time
- Test realistic scenarios — Try with actual users, not just ideal cases
Next Steps
- Response Eagerness — Control how quickly the agent responds
- Thinking Effort — Adjust LLM reasoning depth
- Fast Response — Decide whether a step waits for a reply at all
- Overview — See all conversation dynamics

