overview
The infrastructure where AI assistants and local commerce meet.
AI assistants are moving off the screen and into the world. Online, they already search, compare, and check out. When a person walks into a local business with an assistant, the business often has no good entrance for that request: nowhere clear to receive it, negotiate terms, personalize the arrival, or make sure staff know what was agreed.
We build one system in three parts. The AI Front Door for the business: one entrance, in the owner's voice, connected to the tools they already run. On the person's side: context that travels with them, an assistant that works quietly, and trust built over time. In the middle: a real match first, both sides say yes, and a deal that holds end to end.
02 · the shift
Assistants are learning to navigate the real world.
Online, assistants already search, compare, and check out. Off the screen, they get quiet. A person shows up at a local business with an assistant that knows what they want, and the business often has no good entrance for that request: nowhere clear to receive it, answer it, or shape the visit around it.
The customer has an assistant working for them. We give the business the AI Front Door: one entrance for outside inquiries, connected to the tools the business already runs. Connect your tools, tell it what matters, and it answers in your voice. On the person's side, their assistant can check nearby and act only with consent. We also build the middle where the two meet.
The customer arrives with context. The business needs a way to react to it. That's what we build on both sides, and where they meet.
03 · where things stand today
Three fronts, converging now.
Agents can act on other systems. Agents are gaining the authority and secure means to buy on a person's behalf. And consumers now expect their context to travel with them, from one app to the next. Each shift is real, and each was built for the screen. None of them reach the moment a person walks into a local business.
front · 01
Agents are learning to act on other systems.
who's driving it
Assistants used to answer inside a chat window. Now they open apps, call tools, and hand work to each other. The plumbing for that is being built in the open, and it's ratcheting up what an agent can actually do on a person's behalf.
front · 02
Agents are being trusted to buy.
who's driving it
Agents are starting to check out on their own, with signed proof that a person really approved the purchase. Payments infrastructure is catching up to a world where the buyer isn't always the person tapping a screen — an assistant is acting for them.
front · 03
Context is starting to travel with the person.
who's driving it
People are starting to expect that their preferences, history, and permissions come with them, instead of being re-entered in every app. Portable memory and user-governed context are moving from research into working code.
Three fronts, all real, all still on the screen. None of them yet meet the moment a person walks into a local business. That's the opening we're building for: the AI Front Door on the merchant side, quiet checks and consented action on the person's side, and a middle ground where the two speak the same language.
04 · the gap
What happens when a person walks in with an assistant.
Booking a slot and checking out a known item is increasingly handled. What isn't: the part where a business actually reacts to who's arriving, what they want, what they've been offered, and what staff need to know. That's where we work, then we hand off to the tools the business already runs. On the business's side, we call this the AI Front Door: one entrance that receives assistants, apps, and walk-ins, in the owner's voice.
Generic transactions
Reserving a time slot, checking out a known item, booking a standard service. Widely handled across restaurants, hotels, services, retail. Already handled.
Assistant-driven catalog checkout
You know what you want; the assistant puts it in a cart and pays. UCP, AP2, ACP shipped this year. Online, close to working. Also handled elsewhere.
Personalized in-person action
A person arrives at a local business with context, preferences, budget, party, non-negotiables. Today the business often has no clear entrance to receive that, negotiate before arrival, personalize the visit, or make sure staff know what was agreed. Standards don't cover it. This is where we work.
For the business, we're the AI Front Door: one place that receives outside agents, apps, and walk-in requests, connected to the tools the business already runs. Connect your tools, tell it what matters; it learns your place and the people coming through your door, makes the right call in the moment on pricing, availability, personalization, and permissions, and acts in your voice.
We speak the online standards — MCP, UCP, AP2 — and carry them into the last hundred feet. Where the business already has a booking tool or a register, the agreement lands there. We don't replace either.
Where this lands: restaurants, health & wellness, retailers, venues. Anywhere a person walks in with intent.
Your assistant is your front door out into the world. We're the business's AI Front Door in.
05 · what we're building
One system. Three parts.
Assistants already work online. We build for the visit: context that travels with the person, a Front Door on the house, and the layer where they meet. Three parts on each side. That's the whole picture.
For the person
with their assistant
The person's side. Context stays with them. Their assistant checks around without constantly pinging them. How well a guest and a place know each other shapes what's offered, and what gets through afterward. The person stays in charge.
Context that travels with you
Who they are travels with them: preferences, party, non-negotiables, and what prior visits already taught. When they are near a place or walk in, the house can react to real context instead of a blank guest.
Works quietly, rarely interrupts
Their assistant tests what is true nearby and keeps a high bar for interruption, including after the visit. Most options end in silence. Houses can reach them only inside consent; the assistant still decides what deserves attention.
Trust built over time
How a house is known on the network, and how this guest and this house know each other over time. That signal decides what the assistant trusts, proposes, refuses, and keeps alive after they leave.
For everyone
the layer between
Where the two meet. A person's assistant proposes. The house answers. Fit first, then both sides say yes, then the visit carries through the tools, the door, and settling up, and only then any opted-in follow-up.
A real match first
A request is weighed against what this business will actually do tonight. The same fit test applies when a house goes looking for customers on the network: only people the house can actually serve.
Both sides say yes
Both sides say yes before anything moves. That is when a message becomes a visit. Following up after the visit stays on the same rule: scoped permission, not an open inbox.
The deal holds, end to end
Once the visit is on, arrival does not reopen the deal. Agreed terms are honored in the booking and payment tools the house already runs. Context rides through settle, and when both sides opted in, through a closed post-visit channel that ends when the window ends.
For the business
on the floor
The business's side. One entrance for assistants, apps, and walk-ins. It decides the terms. Reach out on the network to bring in the right customers and stay in touch after, still in the owner's voice. The house keeps control.
AI Front Door
Connect the tools you already run. Tell it what this house stands for. Outside demand lands in one place, so staff know who is coming and what is already agreed.
Offers made for the moment
Price, availability, and terms for this party, from this demand, right now. Tailored to the individual visit, not one coupon for the whole city.
Known across the network
How this house is known across the network, and the history with each guest. That standing is how the house goes out to secure fit customers, how the Front Door answers them, and how post-visit follow-up still sounds like the house, without a second marketing stack.
Context that travels needs a house that can react. Offers made for the moment need both sides to say yes. Reach across the network needs reputation on both sides, and a person's side that's genuinely willing to be reached. The parts only work together.
06 · what could go wrong
What could go wrong.
The things that could sink this, and where we've landed on each.
◐ market risk
"Isn't this a booking product?"
Booking tools find slots. CityOS runs the visit layer: what the business will agree to, the moment both sides say yes and make it real, presence at arrival, and settling up when it ends. We hand off to the booking tool the business already uses when a slot needs to be reserved.
Assistants aren't in enough hands yet.
We build the middle before they're everywhere. If adoption is slower than expected, the consumer side and the business tools have to stand on their own without a wider agent ecosystem.
Businesses are slow to change.
The tool has to disappear into an existing routine. If a manager has to think about us on a busy night, we've lost.
A larger platform copies what we've built.
They could. What's harder to copy is a set of businesses that trust what we've built because they helped shape it.
◐ execution risk
Register integrations are brittle.
We treat each integration as ongoing maintenance, not a one-off.
Privacy expectations are moving fast.
Consumer-side preferences on device. Business-side rules on the side of the business. Identity exchanged only after both sides have agreed. Harder to build than the alternative; we think it's the right constraint.
A relay only works if both sides trust it.
Merchants need enough reach to make the channel worth using. People need it to actually stop when they leave town or when a window closes. If either side loses trust in the scoping, the relay isn't useful to anyone.
07 · who we are
The three of us building this.
Product lead, two founding engineers. Marius is also an AI researcher. All of us work on everything together.
Anthony Carter
Product Lead
Product and partnerships. Lives inside the merchant conversation.
◐ atlanta
Andrei Tîltu
Founding engineer
Systems and integrations. Builds the connections between the tools businesses already run.
◐ europe
Marius Boitor
Founding engineer & AI researcher
Agent behavior and judgment. Shapes how the door decides.
◐ europe
08 · get in touch
Get in touch.
No one has the full shape yet. If you're building in the same direction, we'd rather work it out together.
◐ who we want to build with
We're already building the infrastructure. What we need are partners who see the same future and can help us test it, then expand it.
- 01A partner or region that sees the shift and wants to prove it in one dense, walkable market first.
- 02Local businesses ready to be the first real-world proof points for the AI Front Door.
- 03Assistant platforms and agent companies that need a physical-world layer to reach local commerce.
- 04Time and space for the infrastructure build: testing environments, expansion tooling, and merchant-side systems.