Update, October 9: we retired the connector preview described below. AI shopping agents now reach a store's digital twin through the store's "Ask our expert" page and through site tools next to Shopify's own (changelog).
Short answer: The sqrly MVP now runs end to end on our test store. One sales agent knows every product, color and stock level, answers shoppers on the storefront, and, in a working preview, answers AI assistants like Claude with product cards and a one-click Shopify checkout. In an honest test, an AI reading a well-written website picked the same products. The shop's own agent won on what the website doesn't say: shipping deadlines, deals, the merchant's field knowledge, and finishing the sale. We're now looking for pilot shops.
The MVP gives an online store the expert it never had time to hire: someone who knows every product, checks live stock before recommending anything, and closes the sale. That expert now works on the store's own website, and in a working preview inside the AI assistants shoppers are starting to use.
sqrly is a Shopify app. It runs 24/7 on our test store, Breakline Fishing, a lure brand built for pike, striped bass and barracuda. Breakline doesn't have customers yet; it exists so we can test sqrly on a real catalog with real products. The orders in this post are test orders. Everything else (the advice, the stock checks, the product cards, the checkout links) runs exactly as a shopper would see it.
This post covers what shipped, an honest test of where a shop's own expert beats an AI agent reading the website, the full changelog, and what's next.
What shipped: one expert, two channels
Picture a shopper planning a barracuda trip to the Florida Keys. They can meet sqrly in two ways.
On the store's website. They open the twin on the storefront and ask what to buy for barracuda. The expert recommends the Needle 180 in Green Needle, explaining that it's the lure Breakline's Keys customers catch the most barracuda on, burned fast so it skips across the surface. It adds the one thing a beginner forgets: barracuda teeth cut fluorocarbon line, so they need a titanium leader. Each recommendation comes as a product card with the right color's photo, the exact price, and an Add to cart button.
If the shopper asks about a color that's sold out, the expert says so and suggests the closest one in stock. If they ask whether a freshwater lure will survive a beach session, it says no (the hooks aren't rated for salt) and points them to the saltwater version.
Inside an AI assistant. The same shopper might never visit the website. They ask Claude: "I'm going after barracuda in the Keys and leave Monday afternoon. What should I get from Breakline Fishing, will it arrive in time, and is there a deal on the whole setup?" In our working preview, Claude consults Breakline's expert through sqrly and gets back three things:
- the same grounded recommendation
- a delivery answer worked out from today's date and Breakline's shipping rules ("it's past today's cutoff, so it ships Friday; only Overnight arrives Monday")
- the shop's kit deal, applied automatically
The products appear as interactive cards right in the conversation. One click on Checkout opens Shopify's checkout with the chosen colors and the discount already in the cart.
To be clear about where this stands: today we connect Claude to sqrly by hand for this preview. Shoppers won't set up anything to talk to a store, and they shouldn't have to. The goal is zero setup, so a shop's expert shows up wherever shoppers ask AI for advice. More on that under What's next.
Checkout always stays Shopify's own. sqrly never touches payment.
The honest test: reading the website vs. asking the shop
Before claiming that a shop needs its own expert for AI shoppers, we tested the obvious alternative: an AI assistant that just reads the store's website.
We gave Claude Breakline's full storefront (catalog page plus every product page, sold-out colors greyed out exactly as shoppers see them), turned off web search, and asked the same question in two windows. One window read the website. The other asked the shop's expert through sqrly.
The product picks were the same. On a 15-product store with detailed product pages, Claude reading the website chose the same hollow frog and titanium leader for pike that the expert did. It even noticed that one frog color was greyed out. Good product pages plus a strong AI assistant get you good advice.
We think that's worth saying out loud. If a shop's whole value to an AI agent were "pick the right product", a well-written website would be enough.
The difference showed up everywhere the website is silent. A website reader has no way to answer:
| The shopper asks | Reading the website | Asking the shop's expert |
|---|---|---|
| Will it arrive before I leave on Monday? | "The site doesn't say; check at checkout." | Works it out from today's date and the shop's shipping cutoff, and names the one option that makes it. |
| Is there a deal on the whole setup? | "No bundle pricing is mentioned." | Knows the kit deal and its conditions, and applies the code to the checkout link. |
| What actually works for barracuda in the Keys? | A reasonable guess from product descriptions. | The merchant's field note: what their Keys customers really catch fish on. |
| Can I buy it now? | A list of product links. | The exact colors in a cart, discount applied, one click to Shopify checkout. |
The shop's expert wins on knowledge only the merchant has, live facts like stock and dates, and finishing the sale. That's what we're building sqrly around. The merchant also sees every question shoppers and AI agents ask, which a static website never shows them.
Changelog: what's in the MVP
The expert
- Grounded in the store's live catalog: titles, descriptions, every color and size, prices and stock. It never recommends a sold-out option and never invents a product.
- Honest trade-offs. It says when a product is the wrong choice ("freshwater hooks rust in salt") and points to the right one.
- Merchant knowledge: shipping rules, policies and field notes the merchant teaches it, treated as authoritative.
- Date-aware: it knows the current date and time in the shop's time zone, so it can reason about shipping cutoffs and delivery deadlines.
- Typical answers in about 6 to 13 seconds for a store of this size.
The twin on the storefront
- Product cards with the photo of the selected color, the exact price, an option picker and Add to cart.
- Checkout link that applies any discount the cart qualifies for.
- Aware of the page the shopper is on and what's already in their cart.
- Conversations survive page changes; a start-over button begins a new conversation; the panel can be resized.
AI agent channel (preview)
- A sqrly connector that lets AI assistants such as Claude consult the shop's expert. Today it's set up by hand; zero-setup access is next. (Since retired; see the note at the top.)
- Interactive product cards inside the assistant's conversation, with a one-click Checkout.
- Checkout links built only from in-stock options, with qualifying discounts applied.
- Shop-agnostic: the connector describes each shop from its own catalog.
For the merchant
- Automatic catalog sync: product, price, image and stock changes reach the expert within seconds.
- Offers: define a deal once (code, what the shopper gets, which products qualify). The expert mentions it when it fits, and checkout links apply it. sqrly checks that the code exists in Shopify.
- Conversations: every question from shoppers and AI agents, with transcripts.
- Revenue attribution: orders that came from a sqrly conversation, split into storefront and AI agents.
Under the hood
- Built as a Shopify app with a theme app embed for the storefront and a server AI agents could connect to (since retired; see the note at the top).
- Runs 24/7 in the cloud, with spending caps per shop and per visitor.
What we learned building it
The catalog is the easy part. The merchant's head is the moat. Any AI can read product pages. What it can't read is the shipping cutoff, the deal for a full kit, or which lure actually catches fish in the Keys. The value of a shop's own expert grows with how much of that knowledge it holds, so that's where we're investing next.
Agents shop differently from people. An AI assistant describes the shopper's whole situation in one long request (destination, dates, budget, gear) and expects a decision, not a conversation. We built the expert for that: make the best call from what you're given, state the assumption, and hand back exact options the assistant can put in a cart.
Agent platforms are still young. Connecting a shop to an AI assistant works today, including product cards inside the conversation. But the platforms are changing fast: how tools are discovered, cached and offered from one message to the next. We designed for that, keeping requests self-contained and doing validation on our side.
Honest beats pushy. Every guardrail we kept pays off in trust: never recommend sold-out items, say when a product is wrong for the job, and never invent a deal. An expert that tells you not to buy something is one you believe when it says "this is the one".
Measure before you optimize. Our first version answered in about 14 seconds because it searched the catalog step by step. Measuring showed the time went into back-and-forth lookups, not the agent itself. Giving a small store's whole catalog to the expert at once cut answers to about 6 seconds, with the same picks.
What's next
Knowledge that grows by itself. Today a merchant teaches the expert by typing into a text field. Next, the expert will notice what it couldn't answer and ask the merchant, one line at a time: "Three shoppers asked whether the frog comes in pink this week. What should I tell them?" It will also start from what's already in the store (policies, shipping rates, FAQ pages, existing discounts), so a new shop gets a useful expert on day one without writing anything.
Pilot shops. We're looking for a handful of Shopify merchants with products that need explaining: gear where fit, compatibility or use case decides the sale. Fishing tackle, outdoor equipment, sporting goods and specialty hardware are good examples.
Reaching AI shoppers with zero setup. A connector each shopper adds by hand works for a demo, not for distribution. We're working on two paths that need nothing from the shopper:
- Through the channels agents already use. Commerce platforms like Shopify are building the bridge between stores and AI assistants, so agents can find, compare and buy from merchants directly. When an agent looks at a store that way, it reads the store's data. sqrly can make that data smarter: the merchant's field notes, compatibility advice and shipping logic, available to every agent automatically.
- One sqrly app, many shops. Instead of one connection per store, a single sqrly app in the assistants' app directories that knows every participating shop. Shoppers can find it like any other shopping app, and assistants can suggest it when a question calls for expert advice. (Update: we ruled this path out. See the note at the top.)
Under the hood, the agent channel will also move to proper authentication and serve every installed store, rather than one connection per shop.
If you run a Shopify store where good advice sells, or you're thinking about how your store shows up when shoppers ask an AI instead of searching, we'd love to talk: email florian@sqrly.ai with your store's URL and what you sell.