OpenClaw Phone Assistant with Semantic Turn Detection

A phone-based voice interface for OpenClaw that uses Deepgram's semantic turn detection to understand when callers are finished speaking, not just when they stop making noise. Supports inbound calls via Twilio or Telnyx with low-latency responses and natural barge-in capabilities.

No demo yetView source Fork template

The numbers

latency90ms TTFB (TTS), ~5s LLM buffering

cost / min$0.030/1K chars TTS, $0.085/min Twilio

frameworkCustom

The stack

telephonyTwilio Voice

speech-to-textDeepgram Nova-3

llmClaude Haiku 4.5

text-to-speechDeepgram Aura-2

System prompt

No prompt published.

Config

config.json

{
  "voice_id": "aura-2-thalia-en",
  "barge_in": true,
  "default_model": "anthropic/claude-haiku-4-5-20251001",
  "turn_detection": "semantic_flux",
  "voice_agent_api": "deepgram",
  "supported_voices": 80,
  "telephony_options": [
    "twilio",
    "telnyx"
  ],
  "supported_languages": 7
}

Voice AI recipes, picks, and analysis.

Get the useful new templates plus the occasional teardown of what’s working in production voice AI.