Two ways to talk. One room.
Prayer Pew’s live conversation can run on GPT Live 1 or on LiveKit, Deepgram Flux and a TTS provider. Production uses GPT Live first and keeps the older stack as a fallback. Open one room at a time. Each test hangs up after a minute and only answers what you say; it does not turn the call into a prayer.
Measured 15 September 2026
| GPT Live 1 | LiveKit + Deepgram + TTS provider | |
|---|---|---|
| Typical reply after you stop | 200–400 ms | 2,100–3,100 ms |
| Conversation layer, published | $0.05 / min | ~$0.018 / min floor |
| Minimum billed connect | 15 s | 10 s |
These public rooms are talk-back tests. They hang up after 60 seconds and do not prepare a prayer. Plan, write, and TTS for a finished prayer still cost the same after a real product call.
These rooms only answer what you say. Each call is capped at 60 seconds and hangs up on its own.