The AI that actually runs your front desk.

Wepickup is a full conversational operating system. Four integrated planes work together to answer calls, take action, and never miss a customer.

Architecture

Four planes, one seamless experience

Control Plane

Decides what to say and do
Orchestrator rules for every situation
Intent detection: booking, ordering, inquiry
Agent switching: intent-based or stage-based
Guardrails: silence recovery, loop detection

Runtime Plane

Holds the live conversation
Call session memory with rolling summary
Live tool execution (lookup menu, place order)
Barge-in with configurable sensitivity
State machine: receptionist → booking → payment

Intelligence Plane

Knows your business
Your real catalog — menu items, services, prices
Knowledge Base: docs, FAQs, website content
Caller memory — remembers returning customers
Business profile: hours, policies, location

Media Plane

Connects the phone line
Twilio PSTN — your business phone number
Meta Cloud API — WhatsApp messaging
Live streaming with Twilio Media Streams
Audio recording and transcription

Anti-hallucination built in.

Wepickup is grounded in your real data — your menu, services, hours, and policies. It can only say what's true about your business.

  • Catalog grounding — only your real items and prices
  • Knowledge priority: catalog → business profile → FAQ
  • Strict mode: zero deviation from your data
  • Scenario simulation to test changes safely
  • Full config versioning with one-click rollback

Live tool trace — Control Room

Caller: "Do you have lobster?"

AI tool: lookup_menu_item('lobster')

Result: not found

AI says: "Lobster is not on our menu."

✅ Grounded — never invents items

Under the hood

What happens when a call comes in

When a customer calls your business number, a chain reaction fires across four integrated systems — from the moment the first ring hits Twilio to the confirmation SMS landing in the customer's phone. Here's exactly how it works.

1

The phone rings

Your Twilio number forwards the inbound call to Wepickup. The Media Plane establishes a live audio stream. Within 300ms, the Runtime Plane has accepted the call and is ready to speak.

2

The AI listens and understands

Deepgram transcribes the caller's speech into text in real time. The transcript is sent to the Intelligence Plane, which runs it against your business profile — catalog, knowledge base, caller history, and business rules.

3

It decides what to do

The Control Plane's orchestrator determines the caller's intent: is this an order, a booking, an inquiry, or an out-of-hours call? It selects the right agent (ordering agent, booking agent, FAQ agent) and retrieves the relevant data.

4

It takes action

The Runtime Plane executes tool calls in real time: look up menu items, check table availability, create a booking in Google Calendar, calculate the order total. Each tool call is logged to the Control Room for live viewing.

5

It speaks back naturally

The AI formulates a response using your business data and OpenAI's GPT-4o. It speaks via the Media Plane with your chosen voice — natural, human-like, with appropriate tone and pacing.

6

The call ends with confirmation

When the call completes, the AI sends an SMS or WhatsApp confirmation with the order summary, booking details, or payment link. The conversation is logged to your CRM. The caller's memory is updated for next time.

Why it works

Reliability by design

Three architectural decisions make Wepickup different from generic AI chatbots.

Separation of concerns

The Control, Runtime, Intelligence, and Media planes are independently scalable. A media issue doesn't affect the intelligence layer. Each plane can be updated, rolled back, or scaled without affecting the others.

Catalog grounding

The AI can only reference what's in your catalog, knowledge base, and business profile. No hallucinations. No invented prices. No wrong items. Every fact is traceable to a source document.

Full observability

Every call is recorded, transcribed, and logged. The Control Room shows live transcripts, tool traces, and stage transitions. Scenario simulation lets you test changes before going live.

Configurability

Configure every detail

Your AI behaves exactly how you want. Every toggle, every rule — you control it all.

Caller Memory

Remembers returning callers, past orders

Agent Switching

Different AI for different situations

Silence Recovery

Configurable timeout and retry

Barge-in

Callers can interrupt naturally

After-Hours Mode

Voicemail, callback, or AI-only

Scenario Sim

Test changes against sample calls

Config Versions

Full history, one-click rollback

Voice Personality

Tone and pace customization