Skip to main content

Session Modes

Kallglot supports different session modes for various use cases. Each mode configures how audio is processed and how translation/AI features behave.

Available Modes

ModeUse CaseTranslationAI Agent
bidirectional_translationCustomer support across languagesBoth directionsNo
assistive_translationAgent assistance with translationOne directionNo
ai_agentAutomated voice agentOptionalYes
help_chat_voiceVoice-enabled help deskOptionalYes

Bidirectional Translation

The most common mode for multilingual customer support. Both parties hear translated speech in their preferred language.

Configuration

{
  "mode": "bidirectional_translation",
  "source_language": "de",
  "target_language": "en"
}

Use Cases

  • International customer support centers
  • Sales calls with foreign customers
  • Technical support across languages

Features

  • Real-time speech-to-speech translation
  • Dual-channel recording (original + translated)
  • Full transcript with translations

Assistive Translation

The agent hears the original audio while receiving real-time translated subtitles. The customer hears translated audio of the agent’s speech.

Configuration

{
  "mode": "assistive_translation",
  "source_language": "de",
  "target_language": "en"
}

Use Cases

  • Agents learning a new language
  • Compliance requirements for original audio
  • Quality assurance with human oversight

Features

  • Agent hears original customer audio
  • Real-time translated subtitles for agent
  • Customer hears translated agent speech

AI Agent

An AI-powered voice agent handles the conversation autonomously. The agent uses natural language understanding to respond contextually.

Configuration

{
  "mode": "ai_agent",
  "source_language": "en",
  "target_language": "en",
  "ai_agent": {
    "system_prompt": "You are a helpful customer service agent for Acme Corp...",
    "voice": "alloy",
    "knowledge_base_id": "kb_01ABC",
    "tools": [
      {
        "name": "lookup_order",
        "description": "Look up an order by order number",
        "parameters": {
          "type": "object",
          "properties": {
            "order_number": { "type": "string" }
          }
        }
      }
    ]
  }
}

Use Cases

  • 24/7 automated customer support
  • Appointment scheduling
  • Order status inquiries
  • FAQ handling

Features

  • Natural language understanding
  • Custom system prompts
  • Knowledge base integration
  • Function calling for external integrations
  • Handoff to human agents

Voice Options

Voice IDDescription
alloyNeutral, professional
echoWarm, friendly
fableExpressive, narrative
onyxDeep, authoritative
novaEnergetic, youthful
shimmerCalm, soothing

Help Chat Voice

A voice-enabled help desk assistant that combines AI chat with voice capabilities. Ideal for embedded help widgets.

Configuration

{
  "mode": "help_chat_voice",
  "source_language": "en",
  "target_language": "en",
  "help_chat": {
    "knowledge_base_id": "kb_01ABC",
    "escalation_enabled": true,
    "escalation_phone": "+14155551234"
  }
}

Use Cases

  • In-app help widgets
  • Self-service portals
  • After-hours support

Features

  • AI-powered answers from knowledge base
  • Voice input and output
  • Seamless escalation to human agents
  • Chat history context

Comparing Modes

FeatureBidirectionalAssistiveAI AgentHelp Chat
Real-time translation✅ Both ways✅ One wayOptionalOptional
Human agent requiredOptional
AI responses
Knowledge base
Function calling
Escalation support

Switching Modes

You cannot change the mode of an active session. To switch modes:
  1. End the current session
  2. Create a new session with the desired mode
  3. Reconnect the audio stream
// End current session
await endKallglotSession(currentSessionId);

// Create new session with different mode
const newSession = await createKallglotSession({
  mode: 'ai_agent',
  // ... other config
});

Best Practices

Determine the session mode based on your use case before creating the session. Mode changes require creating a new session.
For ai_agent mode, test your system prompts and tools extensively in a staging environment before going live.
For ai_agent and help_chat_voice modes, always configure an escalation path to human agents for complex issues.
Use assistive_translation mode to train agents on new languages while maintaining quality.