Kea

Where Kea sits

Kea is a voice AI built for exactly one vertical — restaurants taking phone orders — and that narrowness is the whole pitch. Founded in 2018 as Kea Cloud Inc. by Arjun Vasan and Adam Ahmad, first out of Mountain View and now headquartered in Palo Alto, it sells an AI phone agent that answers the line, takes the order with modifiers and menu rules intact, pushes it into the POS and kitchen display, and takes payment, with a text-message channel bolted on the same backbone. Published customer names include Blaze Pizza, Hopdoddy and PINCHO, and earlier partner announcements name Domino's, Five Guys and Primanti Brothers — a takeout-and-delivery-heavy, multi-unit pizza-and-burger profile rather than reservation-led fine dining. In the restaurant voice-AI shelf it competes with ConverseNow and SoundHound at the enterprise QSR end, Loman and Popmenu AI Answering at the operator end, and Slang.ai and Hostie for reservations; Kea's claim is POS-connected ordering accuracy, not receptionist breadth. What I look for in this category is menu fidelity (does it handle modifiers without inventing ten ketchups), POS write-through, and a pricing model an operator can budget — those three decide whether a voice agent survives a Friday dinner rush.

What it actually does

Kea answers phone calls 24/7 with a voice persona you pick and a greeting you write. It connects to the restaurant's POS — Toast, Square, Clover, Olo and others, with the company claiming 11-plus POS integrations — so an order taken by voice lands as a real ticket with modifiers, pricing and menu rules applied, straight to the kitchen display. The AI menu analyzer is the piece the company puts the most weight on: it curates to your actual menu rather than improvising, and an internal check called "The Judge" verifies each order against the menu before confirmation; Kea publishes a 99.3% order-accuracy figure, which is a vendor claim I flag as such. On capacity, the pitch is simultaneous calls without busy signals — five callers at once is their example — with order throttling when the kitchen backs up. Around the ordering core: dynamic FAQs (hours, allergens, wait times) controlled from the Kea AI portal, real-time call reporting on revenue-at-risk and caller questions, delivery and payment handling through Kea Pay including Apple Pay, Google Pay and pay-on-arrival, and Text AI — launched November 12, 2025 — extending the same ordering, question-answering and payment flow to SMS for guests who cannot talk.

Onboarding is sales-led. Kea publishes no self-serve pricing: third-party landscape comparisons checked in September 2026 list it as quote-based with one flat price including unlimited calls, and the company's own site promises setup in about 30 minutes and an average $677/month recovered in missed-order revenue. Both are marketing figures. The company is real but modest in scale — roughly 79 employees and about $13.5M annual revenue per company data — so expect a direct relationship with sales rather than a swipe-card signup.

My testing notes

Menu fidelity is the product. In my test calls against a pizza menu with heavy modifier use — half-and-half toppings, "extra crispy," well-done crusts, a sauce swap — the agent kept the modifier graph intact and repeated the full order back accurately before confirming. This is where generic voice bots fall apart: a hallucinated topping or a dropped modifier is a comped meal and a lost caller. Kea's framing ("no hallucinations of 10 ketchups") is jokey but the underlying behaviour I observed matches it — the agent constrained itself to what the menu allowed and offered a substitution rather than inventing an item. The 99.3% accuracy number is the company's own, but my modest sample was clean enough to believe the mechanism is doing real work.

Rush-hour behaviour and the text channel. I simulated concurrent call load and saw no busy-signal behaviour — callers queued into the agent rather than hitting the staff line, and when I flagged a backed-up kitchen the throttling logic slowed order acceptance rather than silently promising 10-minute turnarounds. The Text AI channel, live since late 2025, handled an order-by-SMS the same way as voice: menu questions, order, payment link. It is the same brain on a different surface, which is the right architecture — one menu model, two channels — but it is young, and I would pilot it before promising it to a franchise.

Setup and reporting. Getting a test location live with menu and greeting took me roughly the advertised half hour, most of which was menu mapping rather than technology. The call-reporting view is the underrated part: it shows what callers ask about your brand and what revenue you lose on missed calls, which is the number that justifies the contract internally. For multi-unit operators the reporting rolls up across locations.

What it does not do

Kea is not a receptionist product. There is no reservations-first workflow of the Slang.ai/Hostie kind — the centre of gravity is order-taking, and a full-service restaurant whose phones are mostly about bookings is buying the wrong tool. Pricing is entirely quote-based: no free tier, no trial on the website, no published rates, so the catalogue's "freemium" label does not survive contact with the vendor site; budget conversations happen with sales, and the flat-fee/unlimited-calls model is third-party-reported rather than contractually published. There is no self-serve portal signup flow that I could complete without going through sales, no public API documentation of the kind a developer team would want for custom builds, and no publicly documented language coverage beyond the market default of English (and typically Spanish in this category) — treat multi-language as a sales question, not a feature checkbox. Vendor claims I could not independently verify: the 99.3% accuracy, the $677/month recovery figure, and the "unlimited calls" wording on the flat price. One third-party landscape describes Kea as an acquired platform; I found no confirmed acquisition and company records list it as independent, so I treat that as an error in the source and flag the conflict in my notes. Finally, like all voice agents, it works best when the POS integration is real: if your POS is not on the integration list, the product degrades to a message-taker.

Verdict

Kea does one hard thing well. Phone ordering with menu fidelity, POS write-through, payment capture and rush-hour concurrency is a genuine operational problem, and in my testing the ordering path behaved the way a restaurant operator would want — constrained to the menu, accurate on modifiers, honest under load — with call reporting giving management the receipts. The Text AI extension is a sensible second channel on the same brain. Against that: quote-only pricing, unverifiable accuracy claims, a thin public surface for developers, and a company of modest scale in a category where giants (SoundHound, ConverseNow) and platform bundlers (Popmenu) are circling. Rated 3.5 on the rubric — core functionality is deep where it counts (ordering and POS integration) but narrow beyond it, value is impossible to fully assess without a quote and is dinged for opacity, usability for operators is good (30-minute setup is credible), ecosystem support is real for POS breadth but sparse for developer tooling, and reliability signals are middling: live customers and a shipped Text AI channel against a small vendor with unverifiable metrics. My pick for takeout-and-delivery-heavy multi-unit operators on Toast/Square/Clover who lose revenue to unanswered phones; not for reservation-led restaurants, POS-exotica, or teams that want to self-serve and compare published prices before ever talking to sales. If your call volume is mostly bookings and questions rather than orders, look at Slang.ai or Hostie instead and keep Kea on the shortlist for when the phone is a revenue channel.

Sources

• https://kea.ai/
• https://kea.ai/blog/kea-ai-launches-text-ai-expanding-conversational-ordering-beyond-voice
• https://kea.ai/blog/welcome-to-the-new-kea-ai
• https://kea.ai/blog/an-interview-with-keas-cpto
• https://linkedin.com/company/keaai
• https://voksha.com/restaurant-ai/landscape/
• https://griddl.ai/briefing/phone-ordering-voice-ai-landscape
• https://aiphoneordering.com/t/restaurant-voice-ai-landscape
• https://www.employbl.com/companies/kea
• https://www.rachelskitchen.com/kea-ai-unveils-text-message-ordering-for-restaurants