Native Magento 2 integration, not an embed
41 customer intents out of the box
Answers only from your catalog, prices and policies
Why native
Why an AI Chatbot for Magento 2 Has to Be Native
41 customer intents out of the boxDays from the OAuth connection to live on the storefront
Most AI chatbots for ecommerce were built for Shopify first and reach Magento through a generic embed: a script tag, a nightly export of the catalog, and answers assembled from whatever the export contained. On a Magento 2 store that approach fails in the places that matter. The catalog is larger and more structured, with configurable products, custom attributes, multiple store views and customer-group prices, so a flat copy goes stale and gets the price wrong. Orders, stock and customer accounts live in Magento, so a chatbot without a real integration can describe a product but cannot add it to the cart, check the stock or say where an order is. And a Hyvä storefront, which more Magento stores adopt every month, breaks the widgets written for Luma's jQuery. The ZestIQ Sales Agent was built for Magento 2 from the start: it connects through OAuth from the Magento admin, syncs the catalog through a resumable queue and keeps it current with webhooks, lets you choose which attributes drive search, ships a widget for Hyvä and Luma, and attributes every cart addition and order back to the conversation that produced it. The AI sits on top of that integration, not in place of it.
In the conversation
What the Sales Agent Does in a Magento Conversation
Everything a good salesperson does at the counter, inside the chat, with your catalog as the only source of truth.
Finds the Right Product, Even From a Vague Question
Exact SKU first, then SKU prefix, then semantic search across the attributes you choose. A customer can type a part number, a plain description or a vehicle, and the agent shortlists what fits and says why.
Checks Fitment Before It Recommends
The customer states the vehicle or the equipment once. Every result after that is filtered to parts that fit, and compatibility is confirmed in the conversation rather than in a separate widget.
Adds to Cart, Validated
The agent adds the item from the chat, checks stock and options before it goes in, and the cart addition is attributed to the conversation, so you can see exactly what the AI sells.
Tracks Orders for the Verified Owner Only
Order status, carrier and tracking number are shown only once the customer has been verified, and never to anyone else. The rest of the conversation stays available to them.
Answers Policy Questions From Your Own Pages
Shipping, returns, warranty and freight answers come from the CMS pages you publish, indexed as knowledge, so the agent cannot invent a policy or quote one you retired.
Hands Off to a Human With the Context
When a request is complex or high value, the agent opens a ticket, a live chat, a callback or a contact form and passes the conversation so far. Your team can also take over live from the admin panel.
What generic chatbots get wrong
Six Ways a Generic Chatbot Fails on a Magento Store
Each of these is a reason Magento merchants switch an assistant off within a month. The Sales Agent was designed around them.
Invented Products and Prices
A model answering from memory will describe a product you do not sell and quote a price you never set. On a store with customer-group pricing, that is a complaint and a refund.
A Stale Copy of the Catalog
A nightly export misses the price change, the stock-out and the new SKU. The customer is told a product is available because it was, yesterday.
No Cart, No Orders, No Account
Without a real Magento integration the chatbot can talk about products but cannot add one to the cart, check an order or see the customer's account. It becomes a search box with manners.
Breaks on Hyvä
Widgets written for Luma assume jQuery and RequireJS. On a Hyvä storefront they fail to load or fight the theme, and the store that invested in speed gets a slow, broken chat.
Scripts That Slow the Store
A heavy third-party script on every page costs Core Web Vitals on every page. The chatbot that was meant to lift conversion lowers it before anyone types a word.
Answers Nobody Can Audit
If nobody can see what the assistant said, why it said it and what it cost, nobody can tune it, and the first bad answer a customer screenshots is the last day it runs.
Connect, sync, tune, go live
How the Sales Agent Goes Live on Magento 2
A pilot takes days, not months. Nothing is installed by hand, no credentials are copied, and every conversation is measured from the first one.
OAuth Handshake
Scoped Access
No Credentials to Copy
No Developer Time
Products and Prices
Stock Levels
Product Attributes
Store Content as Knowledge
AI Tuning With Version Control
Product Rules Engine
Content Management
Roles and Permissions
Widget on the Storefront
The Agent Answers
Every Conversation Measured
Live in Days
Accuracy and control
How a Magento 2 Chatbot Stays Accurate, and Who Controls It
3 tiers of product search: exact SKU, SKU prefix, then semanticYour model OpenAI, Anthropic or self-hosted, switchable per workspace
Every message runs through a pipeline built for correctness before it is built for charm. The agent first works out what the customer wants, across 41 intents covering commerce, fitment, orders, information, technical help and escalation, each with its own definition, examples and confidence threshold. Product search is tiered: an exact SKU match, then a SKU prefix match, then semantic search over the attributes you selected, so the cheap, certain steps run first and the model is used only where it adds value. Prices, stock, orders and policies are read from Magento at the moment of the answer, never from the model's memory, and the response is checked for leaked secrets before it is sent. A conversation state machine keeps context across turns, so a follow-up does not restart the search. The AI provider is your choice: OpenAI, Anthropic or a self-hosted model, switchable per workspace without a redeployment, which also settles where your data lives. Customer emails, phone numbers, card numbers and identifiers are masked before anything is stored, prompt-injection attempts are blocked before they reach the model, and every security event and configuration change is logged. None of that is an enterprise tier; it is how the product works.
Verified reviews
What Our Clients Say
2.1s
faster largest contentful paint
This team consistently goes above and beyond in terms of helping to ascertain the most viable and feasible outputs and outcomes for our evergrowing business needs.
Anthony Perrotta
E-Commerce Operations Manager
4 yrs
as their sole technical partner
They designed for where the product was going, not just what we asked for. Early on we described a single-company setup; they built the data model and permission layer so multi-tenancy wouldn’t require a rewrite later.
Alex Tee
CTO
Inside your Magento store
What Native Magento 2 Integration Means in Practice
Connected through OAuth in minutes, then kept in sync automatically. This is the part a generic embed cannot do.
Your products, attributes and published content flow in and stay current automatically
Automated Catalog Sync
Products, attributes, and CMS content sync via a resumable queue with progress tracking. Real-time webhooks push updates and deletions instantly.
Store Content as Knowledge
FAQ and policy pages are indexed so the AI answers from your actual published content.
You decide which attributes drive search, and every order traces back to the conversation
Attribute-Level Search Control
Choose exactly which product attributes power semantic search and which are exposed as filters, tuned per store, per catalog.
Conversion Attribution
Cart additions and completed orders are tracked back to chat conversations, so you see exactly what the AI sells.
Trusted by 100+ companies
Keep reading
Reading Before the Demo
FAQ
Questions, Answered
Still have questions?
Tell us about your store and we will come back with straight answers.
Book a Demo
Does the AI chatbot work on a Hyvä storefront?
Yes. The widget ships for Hyvä as well as Luma, and it does not depend on jQuery or RequireJS, so it loads on a Hyvä theme without fighting it. A store that moves from Luma to Hyvä keeps the agent, its catalog sync and its conversation history.
Does it work on Luma and on PWA storefronts?
Luma, yes, out of the box. PWA Studio and custom headless storefronts are connected through the same Magento integration, with the widget placed in the frontend application; that setup is scoped in the demo.
Which Magento versions does it support?
Magento 2 and Adobe Commerce, Open Source and Commerce editions, on-premise and on Adobe Commerce Cloud. The integration uses Magento's own OAuth and APIs, so it follows the versions Adobe supports.
How does the chatbot avoid making things up?
It is grounded. Prices, stock, orders and policies are read from your Magento data and your published pages at the moment of the answer, never generated from the model's memory. When a request matches too many products the agent asks a clarifying question instead of guessing, and every response is checked before it is sent.
Can it see customer orders?
Only for the verified owner. Order status, carrier and tracking number are shown after the customer is verified, and never to anyone else. Staff see conversations in the admin panel according to their role.
How is the Sales Agent priced for a Magento store?
Per store, agreed after a demo, based on the catalog size, the storefronts and the conversation volume you expect. There is no price list on this site; the demo on your own catalog is where the conversation about price starts.
Can we use our own OpenAI or Anthropic key, or a self-hosted model?
Yes. The AI provider is configured per workspace and can be switched without a redeployment: OpenAI, Anthropic or a self-hosted model. That choice also settles where your conversation data is processed.
How long does it take to go live?
Days. The OAuth connection takes minutes, the catalog sync runs on its own with progress you can watch, and the widget goes onto the storefront once the intents are tuned to your store. Most pilots are answering customers within the first week.
Does it replace our support team?
No. It handles the questions that have an answer in your catalog and your policies, around the clock, and hands the rest to your team with the conversation attached: a ticket, a live chat, a callback or a contact form. Your team can also step into any conversation live from the admin panel.
What customer data does the chatbot store?
Conversations, with personal data masked before storage: emails, phone numbers, card numbers and identifiers are detected and replaced with placeholders, so the model only ever sees placeholders. Security events, logins and configuration changes are logged and reviewable.
Can we control what the agent says?
Yes, without a developer. Personas, boundaries and confidence thresholds are set per intent in the admin panel, product rules decide what it may recommend, and every change is drafted before it goes live and can be rolled back. Access to each part of the panel is granted by role.
Does it handle fitment for auto parts and equipment?
Yes, and it is one of the reasons the agent was built. The customer states the vehicle or equipment once, every result after that is filtered to what fits, and compatibility is confirmed inside the conversation rather than in a separate year-make-model widget.
Let’s build something great together!
Ready to See It on Your Magento Store?
Book a demo. We connect the agent to a staging copy of your catalog and show it answering your customers' real questions, fitment included.