Voice Calling

Setup

1. Create a bot with type: "inbound_call"
2. Configure Twilio credentials in Settings
3. Set your Twilio phone number webhook to:

https://api.callmissed.com/api/v1/webhooks/twilio/voice

Call Flow

Incoming call → Twilio
  → POST /api/v1/webhooks/twilio/voice
  → Returns TwiML to open WebSocket stream
  → WebSocket /ws/call/{call_id} receives audio
  → Audio chunks → STT → text
  → Text → LLM → response
  → Response → TTS → audio
  → Audio streamed back to caller

WebSocket Streaming

Connect to the voice WebSocket for real-time audio. Requires an API key:

wss://api.callmissed.com/ws/call/{call_id}?api_key=cm_your_key

The WebSocket implements a full STT → LLM → TTS pipeline:

1. Audio in — Twilio sends 8kHz mulaw audio chunks
2. STT — saaras:v3 transcribes in real-time
3. LLM — Generates response using bot's system prompt + conversation history
4. TTS — bulbul:v3 synthesizes speech (MP3 at 24kHz)
5. Audio out — Streamed back to the caller

LLM and TTS run concurrently for minimum latency — audio playback begins while the LLM is still generating.

Outbound Calling

The outbound_call bot type exists, but a public API to initiate outbound calls is not yet available — today the voice pipeline is driven by inbound Twilio calls (and LiveKit voice sessions). Programmatic outbound dialing is on the roadmap. Talk to us if you need it.