Voice Agent Configuration - SubVerse AI

Identity

Field	Type	Default	Description
Name	Text	— (required)	Unique identifier. Lowercase letters, numbers, and underscores only (e.g. `sales_caller`). Must start with a letter.
Description	Text	`""`	Internal description of what this agent does.
Architecture	Dropdown	`Prompt Based`	Prompt Based uses a system prompt. Conversational Flow uses a visual node-based builder.
Debug Mode	Toggle	Off	Outputs extra internal reasoning logs. Use during development; disable in production.

Agent Settings

Field	Type	Default	Description
Greeting Message	Text	`""`	What the agent says when the call connects. Leave blank if the caller speaks first.
Greeting Type	Dropdown	`Prompt`	Prompt — agent generates the greeting dynamically based on context. Say — agent reads exactly what you typed, word for word.
Who Speaks First	Dropdown	`SubVerseAI Agent`	SubVerseAI Agent — agent speaks first (outbound style). User — agent waits for the caller to speak (inbound style).
System Prompt	Text area	Empty	The main system prompt. Define the agent’s identity, role, tone, capabilities, and rules. Use the Refine using AI button to get AI assistance writing or improving the prompt.

Large Language Model Preferences

Field	Type	Default	Description
LLM Model	Dropdown	— (required)	The AI model powering this agent. Each model shows its speed and reasoning ratings.
Thinking Level	Dropdown	`Default`	Extended reasoning depth before responding. Options: Default, Low, Medium, High, Very High. Higher = better reasoning but slower and more expensive.
Max Tokens	Number	None	Maximum tokens the LLM can generate per response. Leave blank to use the model’s default.
Temperature	Slider	`0.5`	Response creativity. `0.0` = consistent and precise. `1.0+` = more varied and creative.

Speech Detection Settings

Field	Type	Default	Description
Transcript Model	Dropdown	— (required)	The speech-to-text model used to transcribe the caller’s speech (e.g. Deepgram Nova 3).
Transcript Language	Dropdown	— (required)	Language for transcription (e.g. English (US), English (India)).
Custom Keywords	Text	None	Words or phrases the STT model should pay extra attention to (e.g. product names, brand terms). Separate multiple entries with commas.
Smart Format	Toggle	Off	Automatically format numbers, currencies, dates, and addresses in transcripts (e.g. “twenty dollars” → “$20”).
Numerals	Toggle	Off	Convert spoken numbers into digits (e.g. “five” → “5”).
Allow Interruptions Overall	Toggle	On	Whether the caller can interrupt the agent mid-sentence. Turn off for strict scripted playback.
Allow Interruptions Initially	Toggle	Off	Whether the caller can interrupt the very first message the agent speaks. Useful for letting eager callers skip the intro.
Interruption Sensitivity	Slider	`2`	How easily the agent is interrupted. Higher = more sensitive. Range: 0–4.

Voice Settings

Field	Type	Default	Description
TTS Model	Dropdown	— (required)	The text-to-speech model for generating speech (e.g. ElevenLabs Flash V2.5).
TTS Language	Dropdown	— (required)	Language for speech synthesis (e.g. English (US)).
Gender	Radio	None	Filter voices by Male or Female. Optional.
Voice	Dropdown	— (required)	The specific voice to use. Click Browse to preview available voices from the selected provider.
Speed	Slider	`2`	Speaking speed. Higher = faster. Range: 0–3.
Voice Stability (Expressiveness)	Slider	`0.5`	How consistent the voice sounds. Lower = more expressive and varied. Higher = very consistent.
Similarity Boost	Slider	`0.8`	How closely the voice matches its original model. Higher = more faithful but may sound less natural.
Voice Pitch	Slider	`2`	Voice pitch adjustment. Range: 0–4.
Voice Volume	Slider	`2`	Output volume level. Range: 0–4.

Call Settings

Field	Type	Default	Description
Communication Method	Dropdown	`Two Way`	Two Way — agent speaks and listens. One Way — agent only speaks (e.g. announcements).
DTMF Enabled	Toggle	Off	Enable touch-tone input. The agent can detect digit presses (e.g. “Press 1 for sales”).
Background Ambient	Toggle	Off	Play subtle background audio (e.g. office sounds) to make calls feel less robotic.
Background Thinking	Toggle	Off	Play a brief sound while the agent is generating a response, signaling to the caller that something is happening.
Transfer To	Text	None	SIP URI or phone number to transfer the call to when a transfer is triggered.
Display Phone Number	Radio	`Show Agent Number`	Show Caller Number — displays the caller’s number to the recipient. Show Agent Number — displays the agent’s number.
Transfer Type	Radio	`Cold Transfer`	Warm Transfer — agent introduces the caller before transferring. Cold Transfer — call is transferred immediately without introduction.
Warm Transfer Prompt	Text	`""`	The message the agent reads during a warm transfer (e.g. “I have a customer calling about a billing issue”). Only applies when Transfer Type is Warm Transfer.
Whisper	Toggle	Off	The agent can hear the caller but the caller cannot hear the agent. Useful for live coaching scenarios.
Silence Time	Number	`8`	Seconds of caller silence before a silence action is triggered.
Silence Turns	Number	`2`	Consecutive silence events before the agent takes action.
Max Duration	Number	`600`	Maximum call length in seconds. The call ends automatically at this limit.
Silence Prompt	Text	`""`	What the agent says when a silence timeout occurs (e.g. “Are you still there?”).
Agent Responsiveness	Slider	`4`	Overall responsiveness after end-of-turn is detected. Higher = responds more quickly. Range: 0–4.
Preemptive Response	Toggle	On	Agent starts generating a response before the caller fully finishes speaking. Significantly reduces perceived latency. Turn off only if you need strict full-turn capture.
Session Timeout (mins)	Number	None	Minutes of inactivity after which the call session ends automatically. Leave blank for no timeout.

Agent Transfers

Field	Type	Default	Description
Select Agents	Dropdown	None	Other Subverse agents this agent can transfer the conversation to during a call.

Knowledgebase Settings

Field	Type	Default	Description
Knowledge Base	Dropdown	None	Attach a Knowledge Base so the agent can look up answers from your documents during a call.

Function Settings

Field	Type	Default	Description
Pre Call Tools	Multi-select	Empty	LLM tools that run automatically before the call starts (e.g. fetch customer account info).
During Call Tools	Multi-select	Empty	Tools the agent can call mid-call based on the conversation (e.g. look up order status, book an appointment, transfer the call).
Post Call Tools	Multi-select	Empty	LLM tools that run automatically after the call ends (e.g. update CRM, send an SMS follow-up).

Agent Security Settings

Field	Type	Default	Description
Allow List (Allowed Callers)	Text area	Empty	Regex patterns for phone numbers allowed to interact with this agent, separated by commas. Leave empty to allow all callers.
Block List (Blocked Callers)	Text area	Empty	Regex patterns for phone numbers that are blocked, separated by commas. Takes priority over Allow List.

Post Call Data Analysis

After each call, the agent can automatically analyze the transcript and extract structured data.

Field	Type	Description
LLM Model	Dropdown	The AI model used to run the post-call analysis. Can differ from the main agent model.
Parameters	Form / JSON	Define the data points to extract from each call. Switch between Form (add parameters one by one) and JSON (paste a raw JSON schema).

Parameter fields:

Field	Description
Parameter Name	The key name for this data point (e.g. `summary`, `user_sentiment`).
Details	Description of what this parameter represents and how to extract it.
Type	Data type: `string`, `boolean`, `number`, etc.
Required	Whether this parameter must be present in every analysis output.