Your AI assistant is your mouth.

Be understood by any business — in any language, with any voice, or with no voice at all.

A communication layer for everywhere commerce happens. Ordering is just the first thing we pointed it at.

An animation shows a request changing form, over and over — and it is a different request every time it runs. First it is sound: a voice waveform. Then the sound scatters, the way a speaker in a noisy place loses it. Then the particles reassemble into the words someone actually used, in one of more than fifty languages: a cheeseburger in Spanish, shawarma in Arabic, pad thai in Thai, momos in Nepali, an accessible hotel room in French, a prescription pickup at a pharmacy. Then those same particles resolve into the one exact line that business receives — “1× Cheeseburger · No Onion”, “Accessible Room · Late Arrival” — and finally into a pickup code. Any language, any counter, one exact transaction out. Nothing is lost on the way, because the sound was never the carrier.

Here is what a transaction looks like, and it changes every few seconds to a different language and a different kind of counter. In the first, a customer writes to their assistant in Hungarian: “Rendelnél nekem egy nagy lattét zabtejjel?” — would you order me a large latte with oat milk? The assistant replies: “Done — sent to Fern Café.” The café’s register receives it in English as exact menu entries: one large latte with oat milk, $6.25, paid by secure checkout, pickup code S P L dash 4 B 2 F. Later ones show a cheeseburger ordered in Spanish, pad thai in Thai, an accessible hotel room booked in French and a prescription collected in Spanish at a pharmacy — each arriving as an exact structured transaction the business already understands.

That’s the whole product. You wrote in your language. The kitchen got exact menu items in theirs. No one spoke.

Typed, not spoken — the same order from an AAC device or a keyboard.

A second exchange, and nothing about it was said out loud. The customer types four words on a keyboard or composes them on an AAC device: “large latte, oat milk”. The assistant replies: “Done — sent to Fern Café.” The receipt that arrives is identical to the first one — Fern Café, one large latte with oat milk, $6.25, paid by secure checkout, pickup code S P L dash 9 C 7 D. Same order, same accuracy, no voice at any point.

Identical receipt. The restaurant cannot tell which of these two orders was spoken, because neither of them was.

The counter is everywhere.

Every one of these is the same transaction wearing different clothes: a person has something specific in mind, and a business needs it exactly. The only thing standing between them is communication.

• Drive-throughsNoise, a low-fidelity speaker, and a queue behind you.
• Counters & cafésA busy room, an accent either way, and a line that makes repeating yourself expensive.
• HotelsRequests that matter — accessibility, allergies, a late arrival — made in a second language at midnight.
• AirportsYou are, by definition, somewhere your language may not be spoken.
• RetailSpecifics: the size, the variant, the exact thing rather than the nearest thing.
• Hospitals & pharmaciesWhere being misunderstood carries the highest cost of anywhere on this list.
• Stadiums & venuesLoud by design, timed by an interval, and impossible to negotiate through.
• Government service desksWhere the person least able to be heard is often the one who most needs to be.

Spling is not a food-ordering app that might expand. It is a communication layer, and ordering is simply the first thing we pointed it at — because it is the highest-volume, most-measurable version of the same problem.

Every hop loses something.

An order isn’t one event, it’s a relay. Each hand-off is a chance to lose a word — and the analogue ones lose the most, because sound is the least reliable thing in the chain. Fewer hops, clearer order. That’s the whole engineering thesis.

The middle isn’t a language. That’s the point.

Spling doesn’t translate your words into the cashier’s words — translation is just another lossy hop. It converts what you meant into item IDs, modifier IDs and quantities: a small, exact, wordless format that belongs to no language at all.

Every language on earth can enter it, and it comes out the other side as whatever that till speaks. Hungarian in, English out. Thai in, French out. Nothing is “nearly right”, because the middle has no room for nearly — an item either exists on that menu or the order doesn’t leave. A universal language for one narrow thing, which is exactly why it works.

Three things nobody else is building.

Ordering, payments, reservations and merchant integrations are becoming commodity infrastructure. We inherit them wherever they exist — they are not the advantage. These three are.

• 01
  Moat

  Language composition

  You communicate in whatever language you naturally use. Spling does not merely translate what you said — translation is another lossy hop, and it still leaves a sentence for someone to interpret.

  It composes a validated structured transaction that matches the business’s live system: real items, real modifiers, real quantities, checked against what they actually offer right now. They receive exactly what their system understands. It cannot invent something, mishear a size, or pass along free text — and that constraint is what makes accuracy possible where speech fails.

• 02
  Moat

  The communication profile

  Speech is optional. A speech difference, deafness, aphasia, autism, anxiety, a temporary injury, or simply being impossible to hear over bad audio — Spling carries that profile everywhere you go, along with your language, your allergens and what you actually like.

  You never have to explain yourself again. Not at a new counter, not in a new city, not to a new assistant. It holds the negatives too — what you sent back, what you never order twice — so “the usual” is a real instruction and a surprise stays inside your allergens. An anaphylaxis allergen never depends on you remembering to mention it: it is blocked before the transaction exists.

• 03
  Moat

  Accuracy intelligence

  The valuable asset was never the ordering. It is owning the whole chain:

  1. Intent
  2. Structured transaction
  3. What the business received
  4. What actually arrived
  5. The correction

  Today the payment processor sees the ticket, the business sees the complaint, and nobody joins them across merchants. Every completed interaction improves the system — and this is operational intelligence, not reviews. It answers questions no star rating can:

  • Which locations are consistently accurate?
  • Which modifiers fail most often?
  • Which translations create confusion?
  • Which merchants improve over time?
  • Does Spling objectively reduce errors?

  This dataset compounds forever, and nobody starting later can back-date it.

One connector, any assistant

Works with

Claude

and

ChatGPT

— and any assistant that speaks MCP.

Spling isn’t an app you open. It’s one connector, built on MCP — the open standard assistants use to reach outside tools. Add it once and it’s there in every conversation after that, on whichever assistant you happen to be using.

Your profile isn’t in the assistant. It’s yours.

This is the part that matters. Your language, how you communicate, your allergens, what “the usual” means, every order you’ve ever placed — none of it lives inside Claude or ChatGPT. It lives with you, and it arrives the moment you connect.

So when you switch assistants — to whatever is best next year, or whatever your school, workplace, or country puts in front of you — you don’t start over. You don’t re-explain your speech difference to a new machine. You don’t retype your allergens and hope. You connect, and it already knows.

And it’s exportable, in the open PAM format, any time you want it. Take it and leave. Nothing about Spling holds you here — that’s the whole point of a profile you own.

Payment

Checkout happens through a secure Square link on your own screen. Spling never sees your card.

The connector is in build. The waitlist gets it first.

What comes after the connector

None of this is live yet. It’s written down here because it’s what the architecture is being built to allow, and we’d rather be held to a published list than quietly ship less.

• The code on your wrist

  A pickup code is four characters and a tap — it belongs on a watch, not buried in a phone. Someone stages an order for you; it arrives on your wrist; you raise it at the counter and it’s verified. No app to find, no screen to unlock, no explaining yourself at the window. For a caregiver ordering remotely, that hand-off is the whole interaction.

• Context that loads before you ask

  Arriving somewhere known should be enough. The business, what they offer right now, your language, your profile, your requirements — all present before you open anything. The location is only the trigger; anyone can draw a circle on a map. What loads inside it is the part worth building.

• Beyond the menu

  The same structured composition applies to a hotel request, a pharmacy counter, an airline desk or a government service window. Wherever a business publishes what it offers, Spling can compose an exact request against it — ordering is just the first catalogue we pointed it at.

• Menus you’ve programmed already

  Your regular orders, saved and named, ready to send in one action — the same idea as an AAC board, applied to ordering. If you eat the same thing at the same place every week, you should never have to compose it twice, and the person behind the counter should get an identical ticket every time.

• Nothing that has to be heard

  For deaf and hard-of-hearing customers the failure isn’t only speaking — it’s the reply. “Was that a medium?” shouted through a speaker is the part that ends the transaction. Order status, substitutions, sold-out items and the ready signal all arrive as text you can read at your own pace, and anything the restaurant needs to ask comes back the same way.

• Caregiver staging, properly

  A parent, support worker or partner composes; the person the order belongs to confirms. Their allergens, their profile, their confirmation — a caregiver can prepare an order but never silently place one. Independence is the point, so the last tap is always theirs.

Want one of these sooner than the others? Tell us at hello@spling.org — it moves the order we build in.