> ## Documentation Index
> Fetch the complete documentation index at: https://docs.osvi.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Call Settings

> Tune how your agent handles interruptions, silence, audio, and response timing

Call settings control how an agent behaves *in the audio stream* — when it stops talking, how long it waits before responding, what it does when the line goes quiet. Defaults work well for most agents; reach for these when an agent talks over people, pauses awkwardly, or responds too eagerly.

All settings live on the agent's **Call Settings** tab.

## Interruptions and turn-taking

| Setting                         | What it does                                                                                                                                                                                                                                                                                                                         |
| ------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Allow interruptions**         | Whether the caller can cut the agent off mid-sentence. On by default — turning it off makes the agent finish every sentence, which feels robotic but suits compliance scripts that must be read in full.                                                                                                                             |
| **Min interruption duration**   | How long the caller must speak before it counts as an interruption (0.1–2.0 s). Raise this if background noise or short "mm-hmm"s keep cutting the agent off.                                                                                                                                                                        |
| **Minimum interruption words**  | Alternative threshold in words — the caller must say at least this many words to interrupt.                                                                                                                                                                                                                                          |
| **Min / max endpointing delay** | How long the agent waits after the caller stops speaking before replying. The minimum (default \~0.4 s) prevents the agent jumping in on a breath pause; the maximum (default \~2 s) caps how long it can hold back. Raise the minimum for callers who speak with pauses (numbers, addresses); lower it for snappier back-and-forth. |

## Silence handling

When the caller goes quiet, the agent checks in rather than sitting in dead air:

* **Voice activity detection (VAD)** — on by default, with a sensitivity slider. Higher sensitivity detects quieter speech but picks up more background noise.
* **Silence message** — what the agent says after a sustained silence. Use a **static message** (e.g. "Hey, are you still there?") or have one **generated from the conversation context** so it fits naturally.
* If the caller still doesn't respond after the check-in, the agent ends the call (the conversation record shows the silence timeout as the end reason).

## Voice and audio

* **Noise cancellation** — on by default; filters background noise from the caller's side before the agent hears it.
* **Background audio** — play ambient sound behind the agent's voice to make the call feel like it comes from a real place. Options: none, city ambience, forest ambience, office ambience, crowded room — with a volume slider. Subtle office ambience makes long pauses feel less artificial.

## Response timing

Presets (**fast / medium / slow / custom**) that trade snappiness against the risk of talking over the caller. Fast suits short transactional calls; slow suits conversations where callers think out loud.

## Supporting LLM

For agents where every millisecond of response delay matters, you can enable a second, smaller LLM that generates a brief acknowledgement ("Sure, let me check that…") while the main model composes the full answer. You choose its provider and model, set its temperature and token cap, and give it a **filler instruction** describing the kinds of acknowledgements it should produce. The result is an agent that never leaves a long thinking gap.

<Tip>
  Change one setting at a time and re-test in the web-call playground. Interruption and endpointing settings interact — adjusting both at once makes it hard to tell which change helped.
</Tip>