I will build a production ai voice agent with vapi, retell, twilio, elevenlabs

Both. Vapi and Retell are excellent for straightforward agents and I'll use them when they fit. When you hit their ceiling — cost at scale, latency, custom telephony, compliance, on-prem — I build directly on LiveKit and your own SIP. I'll tell you honestly which one you actually need.

Depends on your stack. Optimized (streaming STT, fast TTS, direct SIP): ~500-700ms. Typical production with RAG and tool calls: ~700-1000ms. On-prem or compliance routing: 1s+. I measure every hop, show you where the time goes, and tell you what's recoverable and what's physics.

Model provider and tier, STT/TTS choice, whether your VM sits near your telephony route, cloud vs on-prem, and whether tool calls block the critical path. Anyone quoting you one number without asking these questions is guessing. I'll ask them.

Yes. Reliable tool calling is the part most agents get wrong. Calendar, CRM, database lookups, payments — built and tested against real phone calls, not just the demo. That's where most agents break.

Yes. Full source code and deployment on Standard and Premium. It runs on your infrastructure, your accounts, your keys.