Caller memory
For realtime voice agents, Caller Memory → Remember Previous Conversations controls whether earlier conversations are supplied at the start. It defaults to on when no override is saved. The available history covers the caller’s last three conversations from the past 90 days. Phone calls use earlier calls with the same number; browser tests use your earlier tests of that agent. Set the toggle deliberately for your experience, save, and compare a controlled returning-caller test with the recorded earlier conversation. Memory is conversation context, not identity verification or a guarantee that the model will recall every detail. Disabling the setting is not a historical-data deletion action.Ambient noise
Enable Use Ambient Noise to use background ambience on a supported call path. Adjust Volume between 0 and 1. Use the default office ambience or select Custom Audio File and upload audio up to 10 MB. Save the settings after upload and listen on the intended call path to confirm the mix does not obscure speech. Uploading an audio asset only stores it; it does not prove the selected runtime mixed it into a call. If the sound is absent, verify the runtime supports ambience before changing the file repeatedly.Turns and interruptions
Under Turn Detection, choose Transcription-based or Smart Turn Analyzer and set any wait control shown for that strategy. Under Interruption, choose Voice activity to react to detected speech, or Minimum words to wait for a configured number of transcribed words before interrupting. Also set Allow Interruption on the individual conversation nodes. Test a short acknowledgement, a deliberate interruption, a pause mid-sentence, and background speech. Compare responsiveness and false interruptions. These pipeline controls do not establish identical behavior for every realtime model, which may manage turns through its own runtime.Call duration and idle limits
In Call Management, set Max Call Duration (seconds) and Max User Idle Timeout (seconds) to match the task. Save and test a deliberate silent period and a short duration limit on a test agent. Inspect the run’s ending reason. A limit ends or governs the conversation; it does not replace the End Call tool’s conversation-specific instructions.Transcript timing
Enable Enhanced Timestamped Transcript to request start and stop timestamps for each speaker in the uploaded transcript. Save and review a completed run’s transcript, when available. This formatting option does not create missing audio artifacts or make every provider’s timestamps equally precise.Context compaction
Enable Enable Context Compaction to summarize conversation context when transitioning between nodes in the supported pipeline. Check a multi-node test to ensure important facts remain available after transitions. The setting is ignored in realtime mode, where the speech-to-speech service manages conversation state.Dictionary
Open Settings → Dictionary, add the company names, jargon, and other words the agent should actively listen for, then choose Save Dictionary. Test those words in representative speech. Recognition support and additional cost depend on the selected provider; dictionary entries do not guarantee pronunciation or recognition for every voice/model combination.Related settings
- Phone menus and hold covers default-on menu/hold behavior and attaching Press Phone Keys.
- Voicemail and screening covers the outbound classification panel and delayed opening.
- Speech caching covers the MiniMax TTS setting and cache management.
- Outcomes and reports covers disposition classification in General settings.
- Workspace settings covers conditional PBX field updates and captured lead fields.