AI agent for e-commerce

Your AI employee for e-commerce.

Answer customer questions, recommend products, capture sales leads and automate support with an AI agent that understands your store.

Connect your store knowledge, policies and catalogue. Let it handle customer conversations around the clock.

No coding required · Free to start · No card needed

AI shopping assistant

Customer

I need running shoes under ₹5,000.

AI agent

I can help with that. What size do you need?

Customer

Size 9.

AI agent

Here are three options that match — all in stock in size 9, all under budget.
View products

AI activity

Understood product requirements

Searched product knowledge

Filtered by price

Matched size

Generated recommendations

Illustration of a typical conversation. Product results come from recommendation items you add.

The daily reality

Your customers ask the same questions every day.

Your support team spends valuable time answering repetitive questions instead of helping customers with complex issues and growing revenue.

The answers already exist — in your shipping page, your returns policy, your product descriptions. The cost is not that nobody knows them. It is that somebody has to type them out again.

  • Where is my order?
  • Can I return this?
  • Do you have this in size M?
  • How long does delivery take?
  • Which product is right for me?
  • Can I change my order?
  • Do you ship to my location?
  • What is your refund policy?
  • … and the same again tomorrow

The difference

One of these ends with an answer. The other ends with a ticket.

Same question, same store. What changes is how many steps the customer has to take before anything happens.

Today

Traditional support

  1. Customer question
  2. Search the FAQ
  3. Read a help article
  4. Contact support
  5. Wait for a response

Five steps, and the last one happens on your team’s schedule rather than the customer’s. Most people abandon somewhere in the middle.

With an AI agent

AI e-commerce agent

  1. Customer question
  2. The agent works out what they meant
  3. Checks your store knowledge
  4. Recommends, or retrieves the information
  5. Uses a connected tool where one is set upOnly the actions you have configured and permitted.
  6. Resolves the request

One step for the customer. Everything after it happens in the same conversation, in seconds, at three in the morning if that is when they asked.

More than a chatbot

A chatbot answers. An agent gets something done.

Four capabilities that build on each other. Most store widgets stop at the first one.

01

Answer

Answer product, shipping, return, payment and policy questions from your store's own material — and cite which page the answer came from.

02

Recommend

Help shoppers find the right product for their needs, preferences and budget. Matched by meaning, so “something for muddy trails” finds the trail shoe.

03

Convert

Spot buying intent, guide the shopper toward the right product, and capture their details as a scored lead when they want a person.

04

Act

Connect your store or back office over its API and the agent can perform the actions you allow — an order lookup, a support request, an availability check.

On the roadmapThrough your own API today. Packaged store apps are still being built.

Product discovery

Turn product questions into product recommendations.

Your shopper does not know your catalogue. They know what they need it for. The agent works from the second thing to the first.

Shopping conversation

Customer

I need a laptop for video editing under ₹80,000.

AI agent

Sure. What matters most to you — performance, portability, or battery life?

Customer

Performance.

AI agent

Based on that, these three are worth a look. All under budget, and I have put the strongest one for editing in the middle.

One clarifying question is usually the difference between a recommendation the shopper acts on and a list they scroll past. The agent asks it, then narrows on the answer.

Best value

Studio 14 Pro

8-core CPU, 16 GB, 512 GB SSD

₹69,999
View product

Best match

Creator 15 Max

12-core CPU, 32 GB, discrete GPU

₹74,999
View product

Most powerful

Render 16 Ultra

12-core CPU, 32 GB, colour-calibrated

₹79,999
View product

Example products. Recommendation items carry the title, description, price and link you give them; a shopper's stated budget filters them. Prices are the ones you enter — there is no live catalogue sync yet, so keeping them current is still your job.

On the roadmapAutomatic catalogue import from a storefront platform.

Customer support

Handle support questions around the clock.

The same six subjects account for most of what your inbox receives. Every one of them is answerable from material you already have.

Orders

Where is my order?

Shipping

When will it arrive?

Returns

Can I return this?

Product

Does this come in black?

Policies

What is your refund policy?

General

Do you ship internationally?

It answers from your material, not the internet.

Train your agent on your website, FAQs, policies, product information and documents. Answers are drawn from those and cited, so you can check any of them. Where nothing covers the question, it says so rather than filling the gap.

  • Your website, crawled
  • Individual pages
  • PDFs
  • FAQ lists
  • Pasted text

From answering to acting

Don't just tell customers where to look. Help them get it done.

“Track your order on the orders page” is a redirect, not an answer. An agent with access to your order endpoint can simply tell them.

Order support

Customer

Where is my order #48291?

AI activity

Customer identified

Order found

Shipping status retrieved

AI agent

Your order has shipped and is expected to arrive tomorrow.Tracking: TRK-928374
Available with a connected APIExample workflow

An order lookup runs against an endpoint you connect. There is no packaged storefront integration yet, so nothing here happens until you point the agent at your own order API.

What connecting involves

You describe the call once

A URL, the fields it takes, an authentication header, and a sentence saying what it does. The agent works out when it applies.

Credentials are encrypted before storage

Written with AES-256-GCM and never shown again, including to you. Rotating one means replacing it, not reading it back.

Each action gets its own rule

Run automatically, ask the customer to confirm, wait for your team, or switch it off. A lookup and a cancellation do not need the same setting.

Every call is logged

What was sent, what came back, how long it took, and whether it succeeded. Visible in the dashboard, not just in a server log.

On the roadmapOne-click Shopify and WooCommerce apps that skip this setup.

Connect your first API

Automation

Automate the support work that repeats.

A return request is the same seven steps every time. Give the agent the ones it can do safely, and keep the one that needs a decision.

  1. Customer asks about a return
  2. The agent checks your return policy
  3. It identifies the order
  4. It works out whether the item is eligible
  5. It asks the customer to confirmOr waits for your team, depending on the rule you set.
  6. The return request is created

The part that matters

Sensitive actions can require confirmation or human approval.

Every connected action carries one of four settings: run automatically, ask the customer to confirm, wait for your team to approve, or switched off. A tracking lookup and a cancellation are both “actions”, and almost nobody wants them treated the same way.

Nothing is automatic by default — a newly connected action asks the customer to confirm until you change it. Moving one to automatic is a deliberate choice, made one action at a time.

Workflows start from

  • A customer message
  • A detected intent
  • A new lead
  • A handoff being raised
  • A schedule
  • An inbound webhook

And are built from

  • Condition
  • Tool call
  • Delay
  • Notification
  • Data transform

Escalation

It knows when to involve a person.

An agent that guesses is worse than no agent. This one is instructed to stop, say so, and pass the conversation on with everything it already learned.

  1. The agent cannot answer confidently
  2. It says so, plainly
  3. It offers to bring in a person
  4. It captures contact details
  5. A handoff is created
  6. Your team takes overWith the full thread attached, so nobody asks the customer to repeat themselves.

When it does not know

Customer

Do you offer an international warranty on this?

AI agent

I don’t have enough information to answer that accurately. Would you like me to connect you with our support team?

That answer is a feature, not a failure. The question is also recorded as a knowledge gap, so you can see what to write next — and the second customer who asks it gets a real answer.

Human handoff

View conversation
CustomerSarah
ReasonRefund request
PriorityHigh
StatusPending

Revenue

Turn conversations into sales opportunities.

Somebody asking which of two products suits them is further along than anyone who filled in your contact form. The agent notices, and captures them.

  1. A visitor arrives
  2. They ask about a product
  3. The agent gauges how serious the interest is
  4. It recommends something specific
  5. They ask for help, or leave their details
  6. A scored lead lands in your dashboardDe-duplicated, so the same person from two devices is one lead.

New leads

12 this week
John SmithRunning shoes, size 9
87High intent
Sarah PatelPremium plan comparison
64Medium intent
Mike AlvarezRequested a product demo
91High intent

The score explains itself.

Each score keeps the signals that produced it — how specific the question was, whether they named a budget, whether they left contact details, how long they stayed in the conversation. So when your sales team asks why this one is an 87, there is an answer.

Knowledge

Teach it everything about your store.

The answers are already written down somewhere on your site. This is the part where they stop being your team's job to repeat.

What to give it

  • Your website
  • Product information
  • Size and fit guides
  • FAQs
  • Shipping policy
  • Return policy
  • Refund policy
  • PDF documents
  • Anything you paste in

Uploaded as a website crawl, a single page, a PDF, an FAQ list or plain text. Indexing starts immediately and usually finishes in minutes.

  1. Your store material

    Crawled, uploaded or pasted. Split into passages and indexed by meaning rather than by keyword.

  2. The shopper asks something

    In their own words. “Do you deliver to Spain” finds a shipping page that never uses the word deliver.

  3. Only relevant passages are kept

    Anything below the relevance floor is dropped rather than padded into the prompt.

  4. The answer is written from those

    With the source cited. Where nothing relevant was found, it says it does not know.

Train your AI

Integrations

Connect your store. Connect your tools.

One connector is available today, and it is the general one — any HTTP API. The packaged apps are marked with where they actually are.

How a connection is used

  1. Your store
  2. A connected API
  3. Your AI agent
  4. Something the customer asked for
Available

Your own API

Any HTTP endpoint your business already runs. Describe the call once and the agent decides when it applies.

Available

Website widget

One line of script on your storefront. Themed to your brand, and the fastest way to have something answering today.

On the roadmap

Shopify

A packaged app that connects your store without setup. Reachable through your own API in the meantime.

On the roadmap

WooCommerce

Same as Shopify — the REST API works today via a custom connector, the one-click app does not exist yet.

Coming soon

WhatsApp

Built against your own Meta business account. Not yet verified end to end on a live number.

Coming soon

Email

Inbound email as a conversation channel. Written and tested, not yet run against a live mailbox.

On the roadmap

CRM

Push leads into HubSpot, Salesforce or Zoho. All three have public APIs, so a custom connector works now.

On the roadmap

Helpdesk

Create tickets in Zendesk, Freshdesk or Intercom. Same position as CRM — API today, packaged app later.

On the roadmap

Calendar

Booking and availability for stores that also take appointments.

“Coming soon” means the code is written and tested but has not been verified against a live account. “On the roadmap” means it has not been built. We keep the two apart because they are very different things to plan a launch around.

Store platforms

Running on Shopify or WooCommerce?

Both work today through their own APIs. Neither has a one-click app from us yet, and we would rather you knew that before signing up than after.

On the roadmap

Shopify

The Admin API exposes orders, customers, products and fulfilment. Point a connector at the endpoints you want the agent to reach and it can look up an order or check stock in a conversation.

Reachable today via a custom connector

  • Order status lookup
  • Stock and variant checks
  • Customer order history
  • Fulfilment and tracking
On the roadmap

WooCommerce

The WooCommerce REST API covers the same ground with a key pair you generate in WordPress. The setup is one connector per endpoint, and the credentials are encrypted before they are stored.

Reachable today via a custom connector

  • Order status lookup
  • Product and stock queries
  • Customer lookup
  • Coupon and shipping rules

Connecting either one is a job for whoever looks after your store — roughly twenty minutes, once, per endpoint. If you would rather wait for the packaged app, start on your website widget and add the store connection when it lands.

Control

You decide what it is allowed to do.

An agent that can act needs a reason it will not act wrongly. Not a promise about the model — a setting, on every action, that you chose.

Agent settings

Active

Knowledge

  • Product information
  • Return policy
  • Shipping policy
  • Internal pricing sheet

Actions

  • Check an orderRuns automatically
  • Check stockRuns automatically
  • Create a support requestCustomer confirms
  • Start a returnYour team approves
  • Cancel an orderSwitched off

Conditional gate

refund_amount > ₹5,000requires human approval

Your AI works inside the rules you set.

Nothing runs unasked. A newly connected action defaults to asking the customer to confirm, and moving it to automatic is a deliberate choice you make. A tracking lookup is fine on its own; a cancellation probably should not be.

  • Four modes per action

    Automatic, customer confirmation, team approval, or off. Set per action, not per agent.

  • Roles for your team

    Someone can handle conversations without being able to approve a refund or change what the agent says.

  • An audit trail

    Who changed which setting, when, and what it was before. Sensitive values are redacted in the log.

  • A boundary on knowledge

    It is instructed to answer from your material and to say so when it cannot, rather than filling the gap.

Analytics

See how much work your AI is handling.

And, more usefully, where it is falling short. Both are on the same screen because you cannot act on one without the other.

12,482

Conversations

9,840

AI resolved

1,284

Leads captured

312

Human handoffs

%79%

Resolution rate

Most asked

Last 30 days
Order tracking24%
Return policy18%
Shipping time15%
Product availability11%
Product recommendations9%

Knowledge gaps

“Do you offer an international warranty?”

Asked 23 times · never answered

The most valuable list in the product.

Every question the agent could not answer is recorded and counted. Write one paragraph about international warranties, and twenty-three customers a month stop being handed to a person.

Resolution rate climbs as the gaps close.

Example figures. Your own numbers appear per agent, over any date range, alongside a conversation funnel, tool success rates and response times.

See it on your own store

The full journey

Built for the whole customer journey.

The same agent, the same knowledge, from the first product question to the third reorder. Nobody has to hand anything over.

  1. Discover
  2. Recommend
  3. Purchase
  4. Order
  5. Track
  6. Return
  7. Retain

01

Pre-sale

Product discovery and recommendations. The shopper describes what they need; the agent finds what matches.

02

Purchase

Checkout, payment and delivery questions answered in the moment, where an unanswered one becomes an abandoned basket.

03

Post-purchase

Order status and tracking, against your own order endpoint where one is connected.

04

Returns

Explain the policy, check eligibility, and start the request — with the approval gate you configured.

05

Retention

Follow-up conversations, scheduled check-ins and ongoing support, driven by workflows rather than by someone remembering.

Store types

Works for the way you sell.

The agent is the same everywhere. What differs is the catalogue it learns and the questions your customers actually ask.

Fashion

Sizing, fit and stock in a colour.

Beauty

Ingredients, skin type and shade matching.

Electronics

Specifications, compatibility and warranty.

Furniture

Dimensions, materials and delivery lead times.

Home & living

Care instructions and room-by-room suggestions.

Health & wellness

Product information and usage questions.

Food & beverage

Allergens, storage and delivery windows.

D2C brands

One catalogue, deeply understood.

Subscriptions

Plan changes, pauses and billing questions.

For agencies

An AI platform you can run for your clients.

Every workspace is isolated at the database level, which is what makes running four brands from one account safe rather than merely convenient.

Your agency

  • Fashion brand

    AI sales agent

  • Beauty brand

    AI support agent

  • Electronics store

    AI shopping agent

  • D2C brand

    AI customer agent

Four workspaces, four sets of knowledge, one login. A query scoped to one workspace cannot return another’s rows — there are tests written specifically to try.

  • A separate workspace per client

    Its own knowledge, agents, tools, customers and usage limits. Nothing crosses between them.

  • Several agents per client

    A sales agent and a support agent on the same store, each with its own instructions and its own knowledge.

  • Usage tracked per workspace

    Conversations, messages and tokens counted against that client's plan, so you can bill from a real number.

  • Your team in every workspace

    With a role per workspace, so a junior can handle conversations for one client without touching another's settings.

Not built yet

White-label branding

On the roadmap

Your logo and domain in place of ours, for the dashboard your clients log into.

A single cross-client console

On the roadmap

One screen across every workspace, and consolidated billing behind it. Today you switch workspaces.

Security and control

Your customers' data stays yours.

Five mechanisms, all of them in the product today. What follows is deliberately a list of things that work rather than a list of logos.

Tenant isolation

Every database query is scoped to one workspace, and the workspace id is carried on every record rather than inferred from a join. Isolation has its own test suite whose only job is to try to break it.

Permissions

Four roles and ten capabilities, checked on the server for every request. A support agent can work a conversation without being able to approve an action or change what the agent says.

Approval gates

Sensitive actions can require the customer to confirm, or your team to approve, before anything happens. Off is also a setting, and it is the default.

Audit log

Configuration changes and agent actions are recorded with who, when, and what changed. Secrets and personal fields are redacted before the entry is written, not after.

Knowledge boundaries

The agent is instructed to answer from your material and to say when it cannot. Credentials for connected systems are encrypted before storage and never shown again.

What we do not claim

No compliance certification is claimed. We are not SOC 2 audited, and we do not describe the platform as GDPR or HIPAA compliant, because neither has been assessed by anyone qualified to say so. If your procurement process needs a security review, ask for one and we will go through it properly.

Pricing

Start with your store. Scale as your AI handles more.

The free plan covers one agent with enough messages to judge whether it works on your own catalogue. No card, and nothing to cancel if it does not.

Questions

What store owners ask before signing up.

Including the four where the honest answer is “not yet”.

What can an AI agent actually do for my online store?

It answers the product, shipping, return and policy questions that repeat all day, recommends products to shoppers who describe what they need rather than searching for it, captures interested visitors as scored leads, and hands anything it cannot answer to a person with the conversation attached. Where you have connected one of your own APIs, it can also look things up in your systems and act on them within the limits you set.

Can it answer product questions?

Yes, from your own material. Point it at your product pages, upload a spec sheet or a size guide, or paste text in directly. It answers from those and cites which one it used, so you can check it — and where nothing covers the question it says so instead of guessing.

Can it recommend products?

Yes. You add your products as recommendation items with a title, description, category and link. They are indexed by meaning, so a shopper asking for “something for muddy trails” finds the trail shoe without either of you using the word matching. The agent only recommends when the question actually calls for it — nobody asking how to reset a password wants a product pitch.

Is there automatic catalogue sync from my storefront?On the roadmap

Not yet. Recommendation items are added and updated through the dashboard or your own script against our API today. Automatic import from a storefront platform is being built.

Can it check orders?On the roadmap

Yes, once you connect the endpoint. There is no packaged store integration, so an order lookup works by pointing a connector at your own order API — a URL, an auth header and a description of what it returns. That takes a developer about twenty minutes, once. After that the agent decides when a conversation calls for it.

Can it handle returns?

It can explain your return policy immediately, and where you have connected a returns endpoint it can check eligibility and start a request. What it cannot do is process a refund on its own — money movement is not an action the platform performs. Every connected action carries an approval setting, and returns are the obvious case for requiring your team to confirm.

Can I train it on my store?

Yes, and this is the part that takes minutes rather than days. Give it your website to crawl, individual pages, PDFs, FAQ lists, or text you paste in. Indexing starts immediately, and most stores have something answering real questions the same afternoon.

Can I connect Shopify?On the roadmap

Through the Shopify Admin API today, using a custom connector. A one-click Shopify app that skips the setup is on the roadmap and is not built, so if a packaged install is what you need, this is not ready for you yet.

Can I connect WooCommerce?On the roadmap

Same position as Shopify. The WooCommerce REST API works today with a key pair you generate in WordPress and a connector per endpoint. The packaged plugin does not exist yet.

Can customers reach it on WhatsApp?Coming soon

WhatsApp is built and works against your own Meta business account, but it has not been verified end to end on a live number. Email as a channel is in the same position. We would rather tell you that now than have you discover it during a launch.

Will the AI replace my support team?

No, and a store that treats it that way will regret it. It handles the repetitive half of the inbox — order status, delivery times, the returns policy — which is the half your team least wants. Anything complex, angry or unusual is escalated to a person, with everything the agent already learned attached. Most teams end up doing the same amount of work on far more interesting problems.

What happens when it does not know the answer?

It says so and offers to bring in a person. It is instructed not to fill gaps with guesses, and every unanswered question is recorded with a count of how many people asked it — so the gap becomes a to-do list rather than an invisible failure.

Can I control what the AI is allowed to do?

Yes, per action. Each one is set to run automatically, ask the customer to confirm, wait for your team to approve, or switched off. A newly connected action starts at customer confirmation, so nothing runs automatically until you say it may. Separately, your staff have roles, so someone can handle conversations without being able to approve an action or change the agent's instructions. Every configuration change is recorded.

What does it cost to try?

Nothing. The free plan covers one agent with enough messages to tell whether it works on your own catalogue, and no card is needed to start.

Get started

Turn your store into an AI-powered customer experience.

Let it answer questions, recommend products, capture leads and handle the requests that repeat — around the clock, on your own material.

Point it at one product page and ask it something awkward. Twenty minutes is enough to tell whether it is any good on your catalogue.

Start with your website · Add integrations as you grow · No card required