AI Voice — Attendants & Response Speed
Jeff and the six department attendants (Brian, Tim, Elizabeth, Matilda, Eric, Bella) — whether they're
on the line, and the setting that controls how consistently fast they answer.
The AI voice line is switched OFF (support_voice_enabled=0). Nothing below is doing anything until it's on.
Response speed mode
Every attendant's instructions are cached with Anthropic so replies are fast —
but that cache expires after 5 minutes of silence. The first caller to reach a quiet department pays for it: a
turn that's normally 1.5–3 seconds can take 8–13 seconds instead. This setting controls whether we spend a
little to prevent that from ever happening.
Gated (recommended)
Near-zero added cost
Keeps a department's cache warm only while it's actually being called (within the last 30
minutes). Departments nobody's calling right now cost nothing. Trade-off: the very first call after a long
quiet spell can still be slow, once.
Always on — the faster service
~$85–90/month, running continuously
Every attendant stays warm 24/7, so even someone testing the line at 3am on a Tuesday with
nobody having called in hours gets the fast experience, every time. Costs real money whether or not anyone's
actually calling.
Cost estimate: Anthropic Claude Haiku 4.5 pricing, cache-read
$0.10 per million tokens (claude.com/pricing#api), 7 attendants pinged every 4 minutes, 24/7. Takes effect on the
very next check (within 4 minutes) — no restart needed either way.
Attendants
| Ext | Name | Enabled | Registered | Last real call | Warm right now |
|---|
| Loading... |