_
From 2¢ to a fraction of a cent.
Under 2¢ a minute today, under 1¢ within six months, a fraction of a cent in about a year. We own every model in the call path, so the price only goes one way.
Engineered for any compute.
Built to run on whatever compute you have, so concurrency and scale are never the problem. No queue, no scarce hardware, no cap. A server adds capacity. A chip adds a device.
Every model in the call path. Ours.
A domain-specific language model, our own speech-to-text and our own text-to-speech, trained on real business calls and tuned to your domain, menu, scripts and tone. Multiple languages, and any business that answers a phone.
Built for a billion devices.
Because it fits on commodity compute, voice can live where the customer is. Watch, phone, car, fridge, TV, speaker, POS, laptop.
Guardrails, not guesswork.
Cards on screen, never spoken. Child-safe, script-bound, no hallucinated answers. GDPR and local law built in, with consent per jurisdiction.
02 · The problem, five times
Voice AI was blocked five ways. We removed all five.
Three rented vendors. Three margins. Someone else sets your price.
Under 2¢ today, under 1¢ in six months, a fraction of a cent next year.
GPU queues ration concurrency. New customers wait for capacity.
Commodity compute. A server adds capacity, a chip adds a device.
Generic models do not know your menu, your scripts or your tone.
Domain-specific and tuned per business. Multiple languages, a training run away.
Voice lives in the data centre. Your customer is on a watch, in a car, at a till.
Fits on commodity compute, so voice becomes a layer on every device.
Hallucinations, cards read aloud, off-script, blind to local law.
Cards on screen. Read-back. Child-safe, script-bound. GDPR and local law per jurisdiction.
03 · Product · AgentBox
One voice agent that answers, acts and hands off.
Two nights, September
Your details
Guest“Checking in on the 23rd for two nights, two adults.”
Friday night, four guests
Your details
Guest“Table for four this Friday at eight.”
04 · Product · Where it lives
Put AgentBox where your customers already are.
A voice widget on the site you already have.
One script tag. It reads your pages, your menu and your rules, and answers out loud. Scan either code and talk to a live one.
A number that never rings out.
Point your existing line at AgentBox, or take a new one. Same agent, same knowledge, on the phone. Call either number now.
The site is the agent.
No forms, no menus. Customers open the page and talk. It books, orders, checks in and answers, and the screen keeps up with the conversation.
05 · Competition · who owns the call
Two kinds of competitor. Neither one owns the call.
Published list prices and standard-plan concurrency, Sep 2026; all-in ranges include speech, language and telephony pass-through. Direct-app effective cost is subscription ÷ typical monthly minutes. AgentBox runs all three models in-house on commodity compute, so a minute is under 2¢ all-in and concurrency is bounded only by servers.
06 · Roadmap · 36 months
Every six months: a new language, a cheaper minute, more venues. At thirty-six, our own data centres.
Hotels and restaurants in North America, on phone and web. English tuned on live calls; the minute drops under a cent.
Spanish live. Salons and clinics join the same stack. Inference moves to the edge: under a tenth of a cent a minute.
French and Portuguese open Europe and Brazil. PMS and POS marketplaces carry the close; the agent runs on watch, car and speaker.
Arabic and Hindi. The Gulf's multilingual desks and Asia's property counts, at a price only our minute can serve. Six languages, ten thousand venues.
Our own inference fleet in three regions. Every minute on commodity compute we own, and the same racks sell speech and language as a service.
Venues are paying properties across the in-stay and booking lines, at a blended $1.2K a year; ARR is subscription only, no overage. Cost per minute is the all-in cost to serve. The pre-seed funds the first 24 months; the data-centre build is a separate infrastructure raise.
07 · Market · Across the globe
A $35B category by 2033. We need 0.65% of the venues that already take calls.
A $35B category, 39% a year.
Voice agents are the fastest-growing slice of conversational AI. Gartner expects them to remove $80B of contact-centre labour cost by 2026. The spend exists; it moves to whoever owns the cheapest minute.
Who takes calls and misses them.
Sized from venues, not hype: 187K hotels and 15M restaurants worldwide, at the price a property already pays for front-desk software. 62% of calls to small businesses go unanswered.
Every region, in its own language.
North America pays first. Europe and the Gulf buy languages. Asia has the property counts only our cost per minute can serve. One stack, one deployment, any language.
25,000 venues. $30M ARR. 0.65% of SAM.
The year-five plan needs under one percent of the serviceable market, at a blended $1.2K a year, on 80%+ gross margin, because the minute is ours and costs under 2¢.
08 · The raise · Pre-seed
Raising our pre-seed. The terms open in the room.
The raise, the terms and the use of funds open when we talk.
Book 30 minutes to unlock them →
Book 30 minutes with Puru
calendly.com/roshan-purush · or email puru@quickdial.ai
