General
Agent Name The display name for this agent — shown on the canvas, in call logs, and in reports.Ambient Noise Add a subtle background sound to make the conversation feel more natural. Use the slider to control the volume level. You can upload your own audio file (max 10 MB) or use the built-in default office ambience.
Turn Detection Controls how the agent decides when the caller has finished speaking.
- Transcription-based — Ends the turn as soon as the transcription indicates the caller has stopped. Best for short responses like confirmations or one-word answers.
- Smart Turn Analyzer — Uses a machine learning model to detect end of turn. Best for longer responses with natural pauses mid-sentence, where transcription-based detection might cut in too early.
Context Compaction When turned on, the agent automatically summarises the conversation history as it moves between nodes. This keeps the context window efficient on long calls without losing important information.
Context Compaction does not apply in Realtime mode — in that mode, the speech-to-speech service manages its own conversation state and this setting is ignored.
Call Management
- Max Call Duration — The maximum length of any single call in seconds. Default is 600 seconds (10 minutes). The call ends automatically when this limit is reached.
- Idle Timeout — How many seconds of complete silence the agent tolerates before ending the call. Default is 10 seconds. If neither party says anything for this long, the call is terminated.
Voice & Language
Language The spoken language the agent uses for conversation. Voicerix supports multiple languages — select the one your callers speak. Voice The voice the agent speaks in. Available voices depend on the selected language. Use the Preview button to hear how the voice sounds before saving. Transcriber The service that converts the caller’s speech to text. This is selected automatically based on the language but can be changed if needed.Interruption
Controls how the agent decides when the caller is interrupting while it is speaking. Choose one of three strategies: Default Uses the platform default — relies on the transcription service’s own turn signals when available, and falls back to local voice activity detection otherwise. Works well in most cases without any extra configuration. Minimum Words The agent waits until the caller has spoken a minimum number of words before treating it as an interruption. Reduces false triggers from short sounds or background noise.- Minimum Words Before Interruption — How many words the caller must speak before the agent stops. Default is 3.
- Provisional Pause (seconds) — How long the agent pauses while waiting for transcription confirmation. Default is 1.5 seconds.
Template Variables
Variables you define here are available in all prompt fields of this agent using{{variable_name}} syntax. Use this to test your agent with sample data — for example, setting customer_name to Rahul so the greeting uses a real-looking name during testing.
Add a key and value pair, then click Add Variable. These variables are for testing only and do not affect live calls.
Dictionary
Add words the agent should actively recognise during calls — company names, product names, industry terms, or any jargon that standard speech recognition might miss. Enter words separated by commas. Example:FinQuick, EMI Plus, NACH mandate
Adding words to the Dictionary may incur additional cost depending on your speech provider.
Voicemail Detection
When turned on, the agent automatically detects if the call connects to a voicemail system and ends the call instead of leaving a message. Recommended for outbound campaigns where you want to speak only to a real person.Recordings
Recordings are managed at the organisation level and are shared across all your agents. Once uploaded, you can insert a recording into any prompt field by typing@ and selecting from the list.