(Theo · Ext. 109 · knowledge base)

Inbound Receptionist.

Answers every call 24/7 — qualifies, books, takes messages, and warm-transfers emergencies — so voicemail never loses you another customer.

Launch this role

Theo listening to an inbound caller from a quiet reception station in the Stavatar workspace — **Theo**Ext. 109 · Inbound Receptionist

What it does from day one

(no ramp-up, no aspiration)

Answer 100% of forwarded calls 24/7 with a custom greeting — primary, after-hours, or overflow, no voicemail leakage
FAQ answers grounded in the client's knowledge base: hours, pricing ranges, services, insurance
Scripted lead qualification with scoring and spam/robocall screening
Live appointment booking mid-call against real Cal.com/Google Calendar availability, with confirmation SMS
Structured message-taking delivered to Slack/SMS/email with transcript and recording link
Warm-transfer of emergency/VIP calls to the on-call human with a spoken whisper summary, plus sub-minute text-back on abandoned calls

How the work flows

(brief to logged output)

Onboard — client intake (business facts, FAQ doc, qualification script, booking rules, escalation list) compiles into a versioned agent prompt + KB; client approves a test call
Wire the phone — provision a Twilio/Retell number; client sets conditional forwarding (no-answer/busy/after-hours), zero change to their published number
Live call handling — STT→LLM→TTS pipeline greets, qualifies, answers FAQs, checks the calendar via function call, books or takes a message, transfers on escalation triggers
Post-call processing — webhook fires with recording + transcript; agent writes summary, CRM record, appointment log, and sends confirmations and client notifications
Review loop — daily digest of calls/bookings/flags; weekly QA pass proposes prompt/KB updates for client approval

The stack it plugs into

(real tools, official APIs)

Retell AI or Vapi — the hosted voice pipeline: telephony, STT, LLM orchestration, TTS, barge-in, transfer, webhooks
Twilio — numbers, conditional forwarding flows, and outbound SMS for confirmations/text-back
Cal.com / Calendly / Google Calendar API — live availability and mid-call event creation
HubSpot / GoHighLevel or the client's CRM — contact, lead score, disposition, transcript link
ServiceTitan / Jobber / Housecall Pro APIs — book jobs straight onto the dispatch board for home-services clients
Slack API / email — instant message delivery and daily digests

Its machine

(8GB RAM · 4 vCores, dedicated)

With a hosted pipeline (Retell/Vapi) the machine runs only orchestration: webhook server, CRM-sync workers, SMS/digest schedulers, prompt/KB management — under 2GB. A self-hosted LiveKit/Pipecat worker also fits (LiveKit's own sizing: 4 cores/8GB handles 10-25 concurrent sessions with hosted STT/TTS/LLM). Rule: keep speech models in the cloud; occasional Playwright for legacy booking systems without APIs.

Seat economics

light token load

Tokens are trivial (5-15k per 2.5-min call with caching); the real meter is voice minutes at $0.10-0.30/min all-in — a typical SMB at 300-600 min/month costs $30-180, so a 1,000-1,500 min/mo fair-use cap keeps worst case under the $499 seat, exactly the cold-caller's metering design.

Flat seat from $499/agent/mo — one agent, one dedicated machine. See pricing.

Judged by

(the numbers that matter)

Answer rate / missed-call recovery — vs the 62% of SMB calls that go unanswered and the 85% of callers who never call back
Call-to-booking conversion — ~40% of qualifying calls booked is standard, 60%+ is excellent
Containment vs escalation quality — 72-81% containment for appointment scheduling, with a healthy 30-45% escalation band (too-low escalation is itself a red flag)

Tasks in this role

(each documented in the task library)

24/7 AI Call Answering
Every forwarded call answered around the clock with a custom greeting — primary line, after-hours, or overflow — so voicemail never loses you another customer.
Phone Appointment Booking
Appointments booked live, mid-call, against real calendar availability — with a confirmation SMS before the caller hangs up.
Missed Call Text Back
When a caller hangs up before connecting, a text goes out in under a minute to pull them back before they dial your competitor.
AI Message Taking
Structured messages instead of voicemail mush: caller, reason, urgency, and callback details delivered to Slack, SMS, or email with the transcript and recording link attached.
Emergency Call Escalation
Emergencies are never handled solo.

Never alone

(guardrails — how trust works here)

Never handle emergencies solo: defined keywords (gas leak, chest pain, 'attorney now') trigger immediate warm transfer or a 911 instruction — never booked as routine appointments
Never take payment data, SSNs, or full medical/legal details — minimum-necessary contact info only, PHI behind secure links, no advice ever, intake only
Never quote binding prices or commit outside the approved script — KB ranges only; anything off-script becomes 'I'll have the team confirm' plus a message ticket
Never cancel/modify existing appointments or issue refunds without human sign-off; always disclose it's an AI assistant and honor call-recording consent per the client's configured policy

Full policy: our AI disclosure.

Try it

(alpha)

"Miss a call, watch it get saved": split-screen browser demo for a fictional business — visitor plays the 9:47 PM caller through a web-call widget (mic via WebRTC, no phone number), while a live ops dashboard populates in real time: streaming transcript, qualification checklist ticking, a calendar slot getting grabbed, a CRM card written, a confirmation SMS rendered as a phone mockup. A 'trigger an emergency' button ('I smell gas') fires the escalation path live — proving the guardrails is the impressive part for skeptical buyers, and it reuses the cold-call demo's web-call plumbing. Cold calling ships first — try that demo today.