Voice API Documentation

• Authorization: Bearer [token] or API key auth
• Content-Type: application/json
• X-Scope-Type: personal | workspace
• X-Workspace-ID: required when scope is workspace

{
  "message": "Book a follow-up call tomorrow",
  "session_id": "sess_123",
  "stream": true,
  "voice": {
    "input_format": "wav",
    "output_format": "wav",
    "sample_rate": 16000
  },
  "metadata": {
    "lang": "en",
    "client": "web"
  }
}

{
  "success": true,
  "session_id": "sess_123",
  "content": "Sure — I scheduled a follow-up for tomorrow.",
  "provider": "openai",
  "usage": {
    "input_tokens": 210,
    "output_tokens": 96
  },
  "metadata": {
    "latency_ms": 842
  }
}

• session.started — voice session initialized
• stt.partial — partial transcription chunk
• llm.delta — incremental model text tokens
• tts.chunk — audio chunk produced
• session.completed — final response + usage

Interrupt stops in-flight voice generation for a specific session and releases the active execution path safely.

POST /api/v1/voice/interrupt/{session_id}

{
  "reason": "user_barge_in"
}

• Recommended input: WAV (PCM16, mono, 16kHz).
• Non-WAV input can be converted before STT.
• Output format depends on provider and runtime settings.
• Large payloads may return 413.

curl -X POST "/api/v1/voice/chat" \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -H "X-Scope-Type: personal" \
  -d '{
    "message":"Summarize this call",
    "session_id":"sess_123",
    "stream":false
  }'