Skip to main content
Sometimes the agent needs a moment to think or process before replying. Instead of awkward silence, it fills that gap with a natural phrase — like “Sure, give me just a second!” or “Great question, let me check that for you.” This keeps the conversation smooth and the customer engaged, just like a real human would. There are 3 types of filler phrases:

​
None

The agent has no filler phrase. If it needs time to think, it simply stays silent during that time.

​
Dynamic

The assigned LLM generates the filler phrase to fit the situation. It has these settings:
  • Trailing Silence: how long the agent stays quiet after saying a filler, before it starts speaking to the customer again.
  • Dynamic Context Wait: how long the agent waits to gather enough context before deciding which filler to say.
  • Filler Delay: how long the agent waits before playing the filler. Set it to 0 and the agent decides the best time on its own.
  • LLM Model: choose which AI model generates the fillers. Pick the one that responds fastest and sounds most natural — low latency and good phrase quality is what you want.
  • LLM Model Temperature: controls how the fillers sound — lower means similar phrases every time, higher means more variety.
  • Filler Prompt: instructions for how the agent should generate fillers — the tone, style, and kind of phrases to use.

​
Static

The agent picks a phrase at random from a list you provide and plays it whenever it needs to fill a silence. No LLM needed — just the phrases you gave it. It has Trailing Silence, Dynamic Context Wait, and Filler Delay (same as above), plus:
  • Filler Phrase: add the different phrases the agent can use. The agent will rotate between them.