Voice Agent Configuration Guide & Settings

Configure prompts, workflows, voice settings, integrations, behaviors, and AI agent parameters for production deployments.

Configuration Object

Each agent has a nested configuration object that controls its runtime behaviour. Configuration is versioned alongside the agent and can be updated independently.

{
  "llm": {
    "model": "gpt-4o",
    "temperature": 0.7,
    "max_tokens": 1024,
    "system_prompt": "You are a helpful assistant..."
  },
  "stt": {
    "provider": "deepgram",
    "model": "nova-2",
    "language": "en-US",
    "interim_results": true
  },
  "tts": {
    "provider": "elevenlabs",
    "voice_id": "pNInz6obpgDQGcFmaJgB",
    "stability": 0.5,
    "similarity_boost": 0.75
  },
  "turn_taking": {
    "silence_threshold_ms": 700,
    "interruption_enabled": true,
    "endpointing_model": "auto"
  },
  "tools": ["mcp_filesystem_read", "api_crm_lookup"],
  "greeting": "Hello! How can I help you today?",
  "farewell": "Goodbye, have a great day!",
  "max_call_duration_sec": 600
}

LLM Configuration

Name	Type	Required	Description
llm.model	string	âœ“	Model identifier: gpt-4o, gpt-4o-mini, claude-3-sonnet, etc.
llm.temperature	number		Creativity 0-2 (default: 0.7). Lower = more deterministic.
llm.max_tokens	integer		Max response tokens (default: 1024)
llm.system_prompt	string	âœ“	System prompt defining the agent personality and instructions
llm.top_p	number		Nucleus sampling threshold (default: 1.0)
llm.frequency_penalty	number		Penalize repeated tokens (default: 0)

Supported Models: Voxentis supports OpenAI (GPT-4o, GPT-4o-mini), Anthropic (Claude 3.5 Sonnet, Claude 3 Haiku), Google (Gemini Pro), and custom fine-tuned models via API.

Speech-to-Text Configuration

Name	Type	Required	Description
stt.provider	string	âœ“	"deepgram", "google", "azure", or "whisper"
stt.model	string		Provider-specific model (e.g. "nova-2" for Deepgram)
stt.language	string	âœ“	Language code for transcription
stt.interim_results	boolean		Stream partial transcripts (default: true)
stt.smart_format	boolean		Apply auto-punctuation and formatting
stt.keywords	string[]		Boost recognition of specific words

Text-to-Speech Configuration

Name	Type	Required	Description
tts.provider	string	âœ“	"elevenlabs", "azure", "google", or "playht"
tts.voice_id	string	âœ“	Provider-specific voice identifier
tts.stability	number		Voice stability 0-1 (ElevenLabs)
tts.similarity_boost	number		Voice clarity 0-1 (ElevenLabs)
tts.speed	number		Playback speed multiplier (default: 1.0)

Turn-Taking Settings

Name	Type	Description
turn_taking.silence_threshold_ms	integer	Milliseconds of silence before the agent responds (default: 700)
turn_taking.interruption_enabled	boolean	Allow callers to interrupt the agent mid-sentence (default: true)
turn_taking.endpointing_model	string	"auto", "strict", or "relaxed"

Update Configuration

curl -X PATCH https://api.voxentis.com/v1/agents/agt_9f2k4h1a/config \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "llm": {
      "model": "gpt-4o",
      "temperature": 0.5
    },
    "tts": {
      "voice_id": "21m00Tcm4TlvDq8ikWAM"
    }
  }'

Configuration changes on active agents are applied to new calls immediately â€” no redeployment needed (Hot Reload).