Skip to main content

​
Agent Configuration

Agents are the core of Arkenos. Each agent defines a voice persona with its own system prompt, voice, STT provider, functions, and webhooks.

​
Agent Modes

Arkenos supports two agent modes:

​
Standard Agents

Dashboard-configured agents that use the shared voice engine. No code required — configure everything through the UI:
  • System prompt and greeting message
  • STT provider selection
  • Voice selection (Resemble AI)
  • Function calling definitions
  • Pre/post-call webhooks

​
Custom Agents (Preview)

Write your own Python agent code for full control over the voice pipeline. Custom agents:
  • Have a built-in code editor in the dashboard
  • Include an AI coding assistant (powered by Gemini) to help write and edit code
  • Are built into Docker containers and deployed automatically
  • Track file versions for rollback
  • Can use any libraries or custom logic beyond what standard agents offer

​
Creating an Agent

Navigate to Dashboard → Agents → Create Agent or use the API: 
curl -X POST http://localhost:8000/api/agents/ \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Customer Support",
    "type": "PIPELINE",
    "user_id": "your-user-id",
    "config": {
      "system_prompt": "You are a friendly customer support agent for Acme Corp...",
      "first_message": "Hello! Welcome to Acme Corp. How can I help you today?",
      "stt_provider": "assemblyai",
      "voice_id": "resemble-voice-uuid"
    }
  }'

​
Configuration Options

FieldTypeDescription
system_promptStringInstructions for the LLM defining agent behavior
first_messageStringGreeting the agent speaks when the call starts
stt_providerStringassemblyai, elevenlabs, or deepgram
voice_idStringResemble AI voice UUID
functionsArrayFunction calling definitions (see Function Calling)
webhooks.pre_callStringURL called before the voice session starts
webhooks.post_callStringURL called after the session ends

​
STT Provider Selection

Each agent can use a different speech-to-text provider:
ProviderBest For
AssemblyAI (default)High accuracy, general purpose
ElevenLabsLow latency streaming
DeepgramReal-time streaming, enterprise
Set the provider in the agent config and ensure the corresponding API key is configured at API Keys in the dashboard.

​
Voice Selection

Voices are pulled from your Resemble AI account. Browse available voices at Dashboard → Agents → [Agent] → Voice or via the API: 
curl http://localhost:8000/api/resemble/voices?page=1&page_size=10

​
Testing

Use the Preview button on any agent to start a browser-based voice session without a phone call. This connects directly via LiveKit WebRTC — no phone number or Twilio setup needed.