Reference

Get PIM Sync on the Shopify App Store · PIM Sync features, pricing and FAQ · Pro is $10 a month for six months with code PRO10.

Thank you first

Every app we ship is built on someone else's good ideas. PIM Sync is no exception.

The same problem on Shopify

Both ran into the same problem on Shopify, and it is not a flaw in either tool. Shopify's own documentation says its storefront content delivery network is "backed by Cloudflare" (Shopify: performance on the platform). An app whose data lives on Google Cloud or Firebase works, but every Shopify call then crosses from one cloud to another, and moving e-commerce data that includes personal (PII) or payment (PCI) data between clouds gets harder each year.

Two Shopify changes make it sharper:

Keeping the app, its tokens and its data on Cloudflare and Xano, in the market the data belongs to, keeps that path short: one network that already sits in front of Shopify, and one data layer with a home region.

A year ago the same cross-cloud path carried less risk. AI has made development faster, and it has made attacks faster too: the same tools that write an integration in an afternoon can probe one just as quickly. So when data crosses channels, from Shopify to an app, from one cloud to another, from a store to an agent, encryption is mandatory, not a setting, and the token timeline needs a strategy of its own: how long each token lives, where it is stored, who refreshes it and what happens when it expires.

Many of the plugins we are used to adding will no longer be needed. When the data is modeled securely from the start, with every hop over TLS between Cloudflare and Xano, permission rules on the rows and keys kept server-side, the jobs those plugins did, such as moving data, guarding it and syncing it between tools, are already done by the stack.

Built just before AI

Those tools came just before AI could live next to the data. Since Cloudflare's Workers AI binding arrived in 2023, a model runs in the same Worker as the request, beside KV and R2, instead of being a call out to another cloud.

So we kept the methods and extended them, in partnership with AI:

OpenRouter, or a model bound to the data

OpenRouter is a great answer to a real problem: one API and one bill for hundreds of models from many providers, with routing and fallback between them. If you need the widest choice of frontier models, it is hard to beat.

PIM Sync answers a different question: which model should touch this merchant's data, and under which rules?

How does PIM Sync handle model weight and regional LLMs?

PIM Sync Translation models screen: Japanese and Korean each have four steps (reuse a reviewed translation, m2m100-1.2B, Qwen3 30B, keep English), a minimum bge-m3 quality score of 0.75 for each machine model, on switches, move up and down buttons, and a Save ladder button.

Knowledge base segments, with nested rules

Shopify admin, Content, Metaobjects, faq entries: each FAQ row published from CRM Sync carries its question, category, category key, slug and locale (en-US, en-GB), added by the UCP Purchase Agent app.

CRM Sync, not PIM Sync: FAQ rows published to Shopify as metaobjects, each with its category key, slug and locale.

A knowledge base for AI is only safe if each answer comes from rows the asker may see. CRM Sync segments its knowledge base the way it segments customers: identity, then consent, then entitlement, then segments. FAQ and article rows carry segments and a locale, and a question only searches rows in the asker's segment and language, so a French answer never bleeds into an English one and a partner-only answer never reaches a shopper.

One knowledge base, four type structures, one backend: the same support article (FAQ, category, tags and PDF) as a Webflow Collection, a Shopify metaobject definition, a Rust struct and the Xano record each type it, with how each checks shape, handles a missing value, resolves references and keeps a private article private. Xano holds the record and the Cloudflare Worker decides what each function returns to the search page and the chatbot.

See it working:

A global chatbot is where this gets hardest. One widget lives across several Shopify deployments, one per market, and a single conversation can cross a return (RMA), a currency change and a stock check. Each answer touches a different system of record: the ERP for the order and refund, the CRM for the customer and their consent, the WMS for stock and the return label, and the storefront for the price in the shopper's currency. Keeping that answer right, in the right language and currency, for the right market, is only manageable when every system is read through the same segments and rules, not wired to the widget one by one.

LLM automation is what makes the fallbacks practical. Each answer has three ladders, tried in order, and says plainly when it stepped down one:

The model does the wording and the choosing. The records supply every number.

The rules nest. A store's rules sit inside the platform's rules, a team's inside the store's, and an agent's inside the team's, so each level can narrow what the one above allows but never widen it. The same rules decide which tools an AI agent may call: a tool call is one more request, checked against the same entitlement claims before it reaches Xano.

Today PIM Sync's Rust crate, compiled to WebAssembly, enforces the product identity rules Google reads (product group ID, SKUs, check-digit-verified GTINs) at the edge. Rules compiled once in Rust and run as WebAssembly in the Worker are the pattern we are extending to these nested data permissions, so the same compiled rules run wherever the request does.

Your Xano, in your country

Bring your own Xano (Plus). A merchant can connect their own Xano workspace, so the catalog stays in their account. If they leave PIM Sync, the data stays with them.

The same pairing of Cloudflare and Xano meets local data-storage rules market by market, with a Xano instance and Cloudflare settings chosen for each:

Each market's rules are checked by tests, so a release cannot quietly break them. Today those cover Korea's prior-consent rule and the Korean knowledge base; China, Japan and Germany get theirs as each deployment goes live.

Knowledge base fallback, curated. When a question in Korean has no Korean answer, the knowledge base searches Korean content first, then falls back to English and says that it did, instead of guessing or mixing languages. Translated answers are curated per market, so each region's knowledge base carries only the answers reviewed for it.

Curated, not blanket AI

The easy path is AI applied across everything: one model, every row, every user. We take the opposite path. Content, security and permission rules are tuned on purpose, built from a curated structure:

All of it is tailored to stakeholder roles. A merchandiser, a compliance lead, a developer, a partner and an AI agent each see the content, tools and controls their job needs, and nothing more.

It is built for and with design-first principles of utility: every screen, rule and tool call has to make someone's job simpler, or it does not ship.

Here is what that utility looks like. A Korean company can reach Korean and American audiences, using AI as the intermediary and as the security and privacy gate. A small business in Korea can work effectively with a small creative agency in a small town in the USA. The agency writes the product copy and designs the pages in English; the owner reads the Korean translation, fixes any wording by hand and approves it before it reaches the store. Both work on the same catalog rows, each in their own language. Neither needs an enterprise team, an interpreter or a developer to stand between them, and each sees only the content, tools and controls their role needs.

A pathway to success: discover, build, deploy

Concept, review, publish: nothing destructive

Most developer tooling assumes a computer-science background: branches, migrations, build pipelines and rollback commands, each one a way to break something in production. This stack is built so the people who own the content can work safely without that training. Every change moves through three steps, and none of them overwrites what is live until someone says so:

Review has to test what actually runs, not only how it is configured. In our own consent audit work in September 2026, we withdrew a headline finding after runtime testing disproved what the configuration seemed to show. A setting read on its own is a hypothesis; a test of what the page actually loads, and when, is the evidence. That is why consent load order is tested on every release, not checked once.

Security and scaling come with the stack rather than being added afterwards. Keys stay server-side in Cloudflare and Xano, never in a browser or a content editor. Permission rules run before data reaches a page, a person or a model. Cloudflare runs the app wherever requests arrive, so there are no servers to size and no database migrations between a reviewed change and a published one.

Migrating from Gadget

Every Gadget part has a counterpart on Cloudflare and Xano, and your Shopify app keeps its app ID, installs and subscriptions. The feature-by-feature map and the six migration steps are on the PIM Sync page:

Migrating from Gadget

Try PIM Sync Pro: $10 a month for six months

Use code PRO10 on the Upgrade Plan page in PIM Sync. Pro is $10 a month for your first six months, after the 30-day free trial, then $69 a month. The code is for new monthly Pro subscriptions, once per store.

PIM Sync on the Shopify App Store