Skip to content
Giallo

Custom Shopify chatbot: agents that know your catalog, your orders and your rules.

What's the difference between a Shopify chatbot from an app and a custom-built one?

A custom Shopify chatbot reads your catalog, metafields and orders through Shopify's APIs and answers in chat or on WhatsApp following your rules. Giallo Studio builds these agents with Claude, GPT and Gemini as a custom app for your store; you approve returns, discounts and order changes.

Frequently asked questions, straight answers.

Answer using the store's live data. A custom-built Shopify AI chatbot reads products, variants, metafields and orders through the APIs and answers on availability, sizes, shipping and order status, passing anything that needs a decision to the team.

It depends on how specific your catalog is. If FAQs and order status are enough, an off-the-shelf app can work; if you need your own rules, metafields, WhatsApp or handoffs to the warehouse, a custom Shopify virtual assistant serves you better.

Yes. We connect Shopify order events to the WhatsApp Business Platform with a custom app for your store: notifications with approved templates and answers to order questions, only for customers who chose WhatsApp.

No, not in the projects we build. The agent reads orders and prepares the request, but address changes, cancellations and refunds are confirmed by a person in the Shopify admin.

We keep the impact as low as possible. We add it as an app embed block of an app extension, which can be turned on or off from the theme editor without touching the theme code.

For fixed rules, yes; for free text, no. Shopify Flow is Shopify's automation tool that reacts to store events; we use it as a trigger and add the agent where a note or message needs to be understood.

Not right away. We keep your current app active until the agent has passed testing on the development store; then you turn it off from the theme editor and, once you're convinced, you uninstall it yourselves. Before that, it's worth exporting the conversation history if the app allows it: it's the best material for writing the new assistant's rules.

Sources of answers: Does the chat widget slow down the store?Isn't Shopify Flow already enough for automation?

The AI agents we design, one by one.

The order notes agent at work on Shopify: it reads the customer's free-form note, turns the requests into instructions on the parcel (gift ribbon, labels for the concierge desk and delivery time), leaves the discount request to the team and writes a clear internal note for the warehouse.

Store chat agent

Answers questions about products, variants, shipping and orders by reading the Shopify store's live data.

When a visitor opens the chat from a product page, a collection or the cart.

Connects to: Theme app extension · GraphQL Admin API · Store pages

  1. It's loaded into the theme as an app embed block of an app extension, which can be turned on from the theme editor.
  2. Reads the product, variants, inventory and metafields of the page the visitor is on.
  3. For order questions it asks for email and order number and reads the status from the Admin API.
  4. Answers by citing the reference page, such as the size guide or shipping terms.
  5. If the question falls outside the rules, opens a conversation for the team with the summary already written.
Claude Sonnet
It sticks to the data it receives from the store and says clearly when a piece of information isn't there, instead of making up availability.
Claude Haiku
Triages simple incoming questions, such as shipping times or order status, and leaves only the more complex conversations to Claude Sonnet.

Who approves. You approve the pages and fields the agent can use; discount codes, custom prices and order changes stay with the team.

WhatsApp orders agent

Brings order confirmations and updates to WhatsApp and answers the questions that follow.

When Shopify reports an order as created, fulfilled or delivered, and when the customer replies on WhatsApp.

Connects to: Shopify webhooks · WhatsApp Business Platform · Team inbox

  1. Receives order events through Shopify webhooks.
  2. Sends the update with an approved WhatsApp message template, only to customers who chose that channel.
  3. Reads the customer's reply and, if it's about the order, answers with the updated status.
  4. Passes requests for address changes, cancellations or returns to the team, with the order already linked.
Claude Haiku
Customer replies to shipping notifications are short and very frequent: you need speed and a low cost per message.
Codex
Writes and maintains the custom app that receives the webhooks and talks to WhatsApp, with the code reviewed by us; it doesn't read conversations.

Who approves. You approve the message templates before they're submitted to Meta; address changes and cancellations are confirmed by a person in the admin.

Order notes agent

Reads customers' free-form notes and turns them into tags and clear instructions for whoever packs the parcel.

As soon as a new order arrives with a customer note.

Connects to: Shopify Flow · Order tags · Internal notes

  1. Triggered by a Shopify Flow workflow or a webhook when the order contains a note.
  2. Classifies the request: gift wrapping, delivery, replacement, question to handle.
  3. Adds tags to the order and an internal note in a fixed format the warehouse can read.
  4. Flags to the team the notes that ask for a decision, such as a discount or a product swap.
GPT
Returns a classification in a structured format that the app validates before writing tags and notes to the order.

Who approves. The agent only adds tags and internal notes; it doesn't change items, amounts or addresses.

Metafields and catalog agent

Fills in, as drafts, the metafields defined in the store, starting from the free text of the descriptions, ready for your review.

When you upload a new collection or import products from a supplier.

Connects to: Metafield definitions · Product descriptions · GraphQL Admin API

  1. Reads the store's metafield definitions: field type, unit of measure, allowed values.
  2. Looks in existing descriptions for material, dimensions, compatibility and care instructions.
  3. Converts each value to the type the field requires, for example a measurement with its unit or an entry from a closed list.
  4. Leaves blank the fields the text says nothing about and prepares a review table; nothing is written to Shopify until you confirm.
GPT
Returns structured output bound to a schema: we use it to respect each metafield's type and allowed values, so the import doesn't fail.

Who approves. No data is written to the store until you approve the review table.

Your Shopify store in three lines

Send us your Shopify store link and the questions customers ask you most often: we'll tell you what the agent can read, on which channels and where to start.

Prefer the full form? Go to the contact page.

No spam or sales pitches. Response within 1 business day.

We use this data to contact you and, if we work together, to manage the project. We store it in Giallo's internal management system, on servers in the European Union, where only our team can see it; by email we only receive a notice, without your data. If you do not become a client, we delete it 3 months after our last contact. Details and your rights: privacy policy.

Do you prefer to write or call us?

Where time gets lost today.

These are the points we hear about most often. Each one is a possible first agent.

The chat app answers, but knows nothing about the store

The widget installed from the App Store answers with generic phrases or manually uploaded FAQs. It can't tell a size from a variant, doesn't read metafields, and every catalog change has to be copied over.

The agent queries the store's products, variants, inventory and metafields in real time: if you change a price or a product page, the answer changes without reloading anything.

Customers write on WhatsApp, orders are in Shopify

Whoever manages the business number opens the admin, looks up the customer, reads the order and replies from the phone. Shipping updates go out by email and the same questions arrive on WhatsApp.

An agent connected to store events answers on WhatsApp with the order status and sends shipping notifications with approved templates, on the channel the customer actually uses.

Customer notes on orders get lost

“It's a gift, don't include the receipt,” “deliver after six in the evening,” “I'm returning the second size”: requests written in order notes that the warehouse reads only if someone notices them.

The agent reads every note as soon as the order comes in, turns it into tags and instructions for the warehouse and flags the ones that need a decision.

Empty metafields, filters that don't filter

Material, dimensions, compatibility and product care sit in the description as free text. The store's filters stay half-working and no one has time to fill in the fields product by product.

An agent extracts attributes from description text, converts them to each field's type, fills in the metafields as drafts and asks you to confirm before writing to the store.

Which AI model we use, and for which task.

We don't tie you to a single provider: we work with Claude by Anthropic, GPT and Codex by OpenAI, and Gemini by Google, and we choose the version during testing, on your own cases.

Claude Sonnet

  • Store chat agent. It sticks to the data it receives from the store and says clearly when a piece of information isn't there, instead of making up availability.

Claude Haiku

  • Store chat agent. Triages simple incoming questions, such as shipping times or order status, and leaves only the more complex conversations to Claude Sonnet.
  • WhatsApp orders agent. Customer replies to shipping notifications are short and very frequent: you need speed and a low cost per message.

Codex

  • WhatsApp orders agent. Writes and maintains the custom app that receives the webhooks and talks to WhatsApp, with the code reviewed by us; it doesn't read conversations.

GPT

  • Order notes agent. Returns a classification in a structured format that the app validates before writing tags and notes to the order.
  • Metafields and catalog agent. Returns structured output bound to a schema: we use it to respect each metafield's type and allowed values, so the import doesn't fail.

How we choose between models: the best AI model for agents.

How we work with you.

  1. Read-only access to the store

    We start with a custom app that only reads: catalog, metafields and recent orders, to understand how the data is really organized.

  2. Your customers' real questions

    We collect chat, email and WhatsApp conversations, anonymized, and write the response rules starting from them.

  3. Minimal permissions, scope by scope

    Each agent gets only the API access scopes it needs; write access to the store is added later, where needed.

  4. Testing on the development store

    We test the agent on a copy of the theme or on a development store before activating it on the published theme.

AI agents for Shopify stores

What should your store's chatbot know?

Send us your Shopify store link and the questions customers ask you most often: we'll tell you what the agent can read, on which channels and where to start.