What this page covers
Call Settings control the mechanics of the conversation rather than its content: how long the agent tolerates silence, what it says to break that silence, how readily it yields when the caller talks over it, and when the call is cut off regardless of what is happening.
These are configured in the dashboard under the Finn's Call Settings tab. There is no API for editing them independently of the Finn record.
For what the agent says, see prompting and guardrails. For voicemail handling, see voicemail.
Fields
| Field | What it controls | Sensible default |
|---|---|---|
max_idle_duration | How long silence is allowed before the Finn re-prompts | 10s |
idle_reminder | What the Finn says during a re-prompt | "Are you still there?" |
limit_call_duration | Hard ceiling on call length | 5 mins |
voicemail_detection | What to do if voicemail picks up | End call if voicemail detected |
handoff_number | Number to transfer to if escalation triggers | (optional) |
Interruption Sensitivity is a separate control in the same tab. Lower values make the agent more patient and less likely to cut in while the caller is still speaking.
Silence and re-prompts
When the caller stops speaking, the agent waits max_idle_duration before saying idle_reminder. The trade-off is direct and there is no setting that avoids it.
Set it too high and callers sit through long dead air. They assume the line dropped and hang up, which shows up in your logs as a short call with no clear outcome rather than as a silence problem. Set it too low and the agent talks over people who are thinking, reading a number off a card, or fetching a document. That reads as rude and pushes callers into repeating themselves.
If you see long awkward silences in call logs, lower max_idle_duration (5s instead of 10s is a common adjustment). If callers complain about being rushed, raise it.
Write idle_reminder to match the agent's identity. The default "Are you still there?" is neutral but generic. A reminder that references the pending question tends to recover the conversation faster, for example "Take your time. I just need the appointment date."
The context does not establish how many consecutive re-prompts fire before the call ends, or whether the reminder text varies between attempts. Treat that as not yet settled and verify against your own call recordings before relying on a specific count.
Interruption
Interruption sensitivity governs barge-in: whether the caller speaking mid-utterance stops the agent.
High sensitivity means the agent yields quickly. That is usually what you want for conversational inbound calls. It also means background noise, a cough, or a second person in the room can cut the agent off mid-sentence, and the agent will not repeat what it lost.
Low sensitivity means the agent finishes its sentence. That is what you want for mandatory disclosures and recording notices, where a partially spoken compliance line is a compliance problem. It also means an impatient caller repeats themselves three times before being heard.
If troubleshooting shows the Finn interrupting callers, lower the sensitivity first before rewriting prompts. This is a mechanical problem and prompt changes will not fix it.
Call length
limit_call_duration is a hard ceiling. When it is reached, the call ends. The agent does not get a graceful closing turn, and any task that had not yet fired may not fire.
The default is 5 minutes. That is short for anything involving multi-step intake or a knowledge base lookup chain. Before raising it, check what your longest successful calls actually run to, then set the ceiling above that with margin rather than at the average.
Two consequences worth planning for:
- Post-call analysis on truncated calls. Fields extract from the transcript. A call cut off before the caller confirmed anything leaves those fields blank or wrong. See post-call analysis.
- Tasks that never run. If a booking or webhook task fires at end of call, a hard cutoff can skip it. See tasks.
Very short calls have their own effect on reporting. Some metrics filter out calls under 5 seconds and treat them as no-answer, so a run of instant hangups can make analytics look emptier than the raw call log.
Escalation
handoff_number is the destination when escalation triggers. It is optional. If it is empty and your prompt tells the agent it can transfer the caller, the agent will promise a transfer it cannot perform, which is worse than telling the caller it cannot help.
Set the number and the prompt together, or set neither. See transfer.
Applying changes
Edits apply to new calls only. Calls already in flight finish under the configuration they started with. If you change max_idle_duration mid-campaign, expect mixed behaviour in the same deployment's logs for the length of the longest in-flight call.
Test the change before launching. Test on Call exercises real timing including silence and barge-in. Test on Chat does not, since there is no audio channel and no silence to measure. See testing.