WINDOW 01UNCOVERED
The peak-hour rush
Lunch and dinner service hit and every hand is full. The phone rings against a wall of noise, and nobody can reach it.
[PLATFORM: ONITA] // VOICE AGENT // SG NODE // ACTION-BASED CORE
Singapore · Private beta
ONITA
The AI that answers every call.
ONITA picks up the phone for service businesses, in English, Singlish, Mandarin, Malay and Tamil. She books the table, takes the order, collects the deposit and saves the lead while the caller is still on the line. At the lunch rush. At three in the morning. Every time.
01 — The problem
30–40%
of calls to a busy service business go unanswered —
not from neglect, but because nobody is free to pick up.
A missed call is
a missed customer.
Not a missed message — a missed booking, a missed order, a missed patient. The caller does not leave a voicemail and try again tomorrow. They dial the business next door.
Phones fail in three predictable windows. ONITA was built to cover all three.
WINDOW 02UNCOVERED
After hours
The kitchen closes at ten. Callers do not. Every booking attempt between close and open goes straight to nobody.
WINDOW 03UNCOVERED
The language gap
One caller speaks English, the next Mandarin, the next Singlish — sometimes all three inside one sentence.
02 — When ONITA answers
Anything the
business needs.
ONITA is not a menu-reading bot. She is an action platform: the owner switches on the actions they want, and she only ever offers those. Seven of them ship today, inbound and outbound.
Answer questions
answer_questions
Take bookings
take_bookings
Take orders
take_orders
Take messages
take_messages
Transfer to staff
transfer_to_staff
Capture enquiries
collect_enquiries
Call back out
make_calls
Inbound · 22:47 · closed
FIG. 01 — LIVE CALLHi, tonight got table for four anot?
We’re closed now, but I can hold 7pm tomorrow for four. Shall I book it?
可以，帮我订。Also can add takeaway laksa?
订好了 — 明天七点，四位。And one laksa for pickup. I’ll text the confirmation now.
Table booked Order captured Confirmation sent
She fits into what the business already runs.
ONITA does not ask anyone to migrate. She plugs into the reservation book, the ordering system, WhatsApp, the point of sale, the payment rails and the calendar that are already in place. The owner changes nothing.
ChopeOddleWhatsAppPoint of salePayNowCalendar
Anything missing is an adapter, not a rewrite — the connector SDK is open to any business management system.
03 — The engine
One brain.
Any service business.
Most voice AI is built for one industry and dies there. ONITA’s core never mentions a menu or a table — it deals in bookings, orders, messages and enquiries. A vertical is only a set of labels and defaults on top. Flip the switch and watch the same engine become a clinic.
"Tonight got table for four anot?"take_bookingsTable bookings
take_ordersTakeaway orders
answer_questionsHours, menu, parking
transfer_to_staffTransfer to owner
take_messagesTake a messageUNCHANGED
collect_enquiriesCatering enquiry
make_callsConfirm & win back no-shows
Same brain, different switches. Nothing above is hard-wired to a vertical — the owner toggles the actions they want, and ONITA only ever offers what is switched on.
04 — Four surfaces
However they reach
you, she’s there.
ONITA is a platform, not a phone bot. She arrives as four surfaces — and the fourth matters more every quarter, as AI agents start making bookings on behalf of the people who used to dial.
SURFACE 01
Voice Agent
Answers the phone over a real Singapore number. Handles barge-in, turn-taking and mid-sentence language switching the way people actually talk here.
SURFACE 02
Five-screen dashboard
Home, Calls, Knowledge, Bookings, Settings. That is the whole product. Mobile-first, no settings labyrinth, and a returning monthly user finds everything instantly.
SURFACE 03
REST API
A direct line into outlets, bookings, call records and analytics. Keyed, rate-limited, and consistent — every response carries a request ID.
SURFACE 04
MCP Server
ONITA speaks Model Context Protocol, so other AI agents can use her as a tool — searching businesses, checking availability and booking on a customer’s behalf.
05 — Proof, monthly
The number that
renews itself.
Every call ONITA answers, she asks one question: would this have been missed without me? After closing time? While the line was busy? During the Friday rush?
Then she adds up what those saved calls were worth and shows the owner a single figure. Renewal stops being a feeling and becomes arithmetic.
Revenue recovered · this month
$4,120
After-hours calls$1,860
Busy-line calls$1,410
Peak-rush calls$850
Subscription$230
06 — No-shows
She protects the
revenue she books.
A party of ten that never turns up is a ruined night. So on large bookings ONITA texts a PayNow deposit link and holds the slot. Paid, the booking confirms. Unpaid, the slot releases itself.
No awkward phone chase, no staff time spent confirming, no dead tables. And when a no-show does happen, she calls back out to rebook it.
PayNow depositsAuto-release holdsNo-show win-back calls
STEP 01
Booking taken
Party of eight, Saturday 7pm. Slot held.
STEP 02
Deposit link sent
A PayNow link goes out by text within seconds of the call ending.
STEP 03
Paid — confirmed
The hold becomes a confirmed booking and syncs to the reservation system.
STEP 04
Unpaid — released
The slot returns to inventory automatically. Nobody has to chase it.
07 — Why it holds
Built for here,
not ported to here.
The voice AI bet has already been settled in the United States. What has not been built is a platform that speaks the way Southeast Asia speaks and plugs into what Southeast Asia runs on.
MOAT 01
Code-switching, not translation
English, Singlish, Mandarin, Malay and Tamil — switched mid-sentence. That behaviour sits in how ONITA listens, not in a translation layer bolted on afterwards.
MOAT 02
The local stack
Connectors into the systems Southeast Asian businesses already run on — reservations, ordering, WhatsApp, point of sale, PayNow. No US platform touches these rails.
MOAT 03
Action platform, not a vertical bot
The core never mentions menus or tables. Verticals are setup templates, so a clinic or a salon is a configuration — not a rebuild.
MOAT 04
MCP-first
As AI agents start booking on behalf of consumers, they need an endpoint. ONITA is built to be that endpoint for the region from day one.
Singapore-region infrastructurePer-tenant isolationSigned telephony webhooksRecorded, transcribed, auditable calls
08 — Pricing
Per outlet.
Per month.
One price per location, minutes included, and a per-minute rate after that. Going over is not a penalty — it is still cheaper than a missed booking.
Starter
$149SGD / outlet / mo
300 minutes included · then $0.20 / min
- Answer questions
- Take messages
- Transfer to staff
- Call log & transcripts
Pro
$230SGD / outlet / mo
600 minutes included · then $0.18 / min
- Everything in Starter
- Bookings & orders
- Outbound calls & reminders
- PayNow deposits
- Connectors
Business
$370SGD / outlet / mo
1,200 minutes included · then $0.14 / min
- Everything in Pro
- High-volume locations
- REST API access
- MCP server access
Onboarding Singapore F&B now
The fastest demo is
to let her answer.
Setup is self-serve and takes about fifteen minutes — point your number at ONITA, tell her about the business, switch on the actions you want. Then call your own line and hear it.