AI agent
Answers built on your store. Handoffs with context.
The Kanesh AI agent answers customers from your policies, catalogue and live Shopify orders, and hands off to your team when it shouldn't answer.
Illustration: a customer asks on WhatsApp where her order is and whether she can exchange it. Before replying, the AI verifies she is the customer on the order, reads the order status from Shopify, finds the exchange rule in the brand's returns policy and checks the draft against the brand's data and rules. Then it replies with the tracking link and the policy.
Answering fast shouldn't take another shift.
The same questions every day
Where's my order, which size fits, how long is shipping. Your team answers them again and again.
Sales slip away after hours
Whoever asks at night or at the weekend buys elsewhere if nobody replies.
A bot without data makes it worse
If it can't see the order or your policies, it guesses or hands everything over, and the customer still waits.
How a reply gets made.
Every reply goes through the same four steps. The demo above shows them on a real-looking question.
- 01
Read the conversation
It reads the customer's message and the conversation so far, in Spanish or English.
- 02
Verify, then look up
Before any order detail, it checks that the person writing is the customer on the order. Then it reads Shopify.
- 03
Find it in your knowledge
Policies, FAQs and website pages you've added. The answer has to be backed by them or by your store data.
- 04
Check before sending
The draft is checked against your data and rules. If it can't be backed up, it goes to a person.
It answers from what you've written.
Upload your policies and FAQs, and the AI answers from them. Add pages from your website too, and set the rules it has to respect.
- Upload policies, FAQs and guides as documents.
- Index pages from your website into the knowledge base, then check which pages made it in.
- Rules for what it may and may not say, checked on every reply.
Illustration of a knowledge base: two uploaded documents, two pages from the brand's website (one indexed, one still to review) and a brand rule.
It knows the order and the catalogue.
It answers “where is my order?” with live Shopify data after verifying the customer, and it recommends products from your catalogue with the price and a link.
- Reads order status, shipments and tracking, returns, refund status and gift cards.
- No verified match, no order details.
- Product suggestions from your Shopify catalogue, with a link to the product page.
Illustration: a customer asks for something warm for an October wedding in size S; the AI searches the live catalogue and suggests two products available in that size, with prices and links.
It knows when to stop.
When a request needs a decision, falls outside your policy or the data isn't there, the AI hands the conversation to your team with the reason and the context.
- The reason for the handoff and the context of the conversation.
- Situations you define, like exceptions, complaints or wholesale enquiries.
- The conversation stays in the shared inbox, so nobody asks the customer twice.
Illustration: on Instagram, a customer asks for the price difference on a coat that went on sale after she bought it. The AI finds her order and checks the returns policy, which doesn't cover price adjustments, so it hands the conversation to the customer care team with the reason and a summary. Marta from the team picks it up and replies.
Drafts your team approves or edits.
In the conversations your team handles, the AI drafts a reply from the same knowledge and store data. Approve it as it is, or edit it before it goes.
- Your team always has the last word on what's sent.
Illustration: in the inbox, the AI proposes a reply about shipping times; the agent can approve and send it or edit it first.
What the AI won't do.
No AI gets everything right. That's why the checks sit before sending, and why the way out is always a person.
Make up order data
Order answers come from Shopify, after verifying the customer. If it can't verify them or can't find the order, it doesn't share order details.
Promise what your policy doesn't say
Replies are checked against your documents and rules before they're sent. If a reply can't be backed up, it goes to a person.
Keep going without the data
When the answer isn't in your knowledge or your store, it hands the conversation to your team with the reason, instead of improvising.
Want the detail? We explain how the AI works, what it reads and where its limits are.
How our AI worksThe practical details.
Chat channels and email
WhatsApp, Instagram, Messenger, web chat and email, from the same knowledge.
Spanish and English
The AI replies in Spanish and English.
In your shared inbox
Every AI conversation is in the inbox, next to the ones your team handles.
Built on leading LLMs
We don't publish model names: they change, and every reply goes through the same checks.
Already answering for real brands.
0.9 min
median first reply on WhatsApp
Measured · Customer-started WhatsApp conversations, 1–27 September 2026. All got a reply; 90% within 1.7 minutes.
MonisquiRead the story: Monisqui97%
of chat conversations waiting for an answer got one within 2 hours
Measured · WhatsApp, Instagram and Messenger, in the 7 days to 30 September 2026.
MonisquiRead the story: Monisqui2.2 min
median first reply on WhatsApp
Measured · Customer-started WhatsApp conversations, 1–27 September 2026. 96.3% got a reply within 24 hours, from the AI or a person.
Alma ConceptRead the story: Alma Concept
Questions about the AI agent
How does the Kanesh AI agent answer customers?
Which channels does the AI answer on?
How does it verify the customer before talking about an order?
Can the AI cancel, refund or change orders?
When does the AI hand a conversation to a person?
What data does the AI use?
Which languages does it reply in?
Which AI model do you use?
See what your conversations are missing.
See how your channels, team and AI work together. Let’s talk about your operation and your plan.