Autonomous agents with memory and context — patent filed with the USPTO

Virtual salesperson

A virtual salesperson on your store's WhatsApp

It serves your customers by text, voice and image, shows them what the store sells, and follows up with the ones who went quiet mid-conversation — and on that same number, it answers you, the owner, about how the day went.

This isn't a chatbot running a flow. It's an operational agent trained to work.

Optical shops, niche fashion, pet, supplements, home decor, small electronics and e-commerce that lives on WhatsApp.

Interior of a retail store with tidy shelves and a salesperson helping at the counter.
The problem

Your store has WhatsApp. WhatsApp talks, but it doesn't sell.

A traditional chatbot depends on a flow somebody drew in advance. Step outside the script and it breaks. It answers questions, but it doesn't help you sell or run the store.

Chatbot running a flow

  • Depends on a flow someone had to design first
  • Breaks the moment a customer asks something off-script
  • Doesn't really check inventory or the catalog
  • Never follows up with the people who stopped replying
  • Leaves the owner opening a dashboard to find out what happened

Hal-AI virtual salesperson

  • Holds context instead of walking a decision tree
  • Understands the question however it arrives — text, voice or photo
  • Reads the catalog and, once connected, your store systems
  • Follows up with customers who stalled partway through a purchase
  • Reports back to the owner on WhatsApp, with no dashboard in between

Specialty retail feels this first: mid-ticket stores with variant-heavy catalogs, where customers ask about size, color, compatibility and delivery time before they buy.

Two audiences, one agent

It serves your customer. And it answers to you.

Same agent, same number. All that changes is who's on the other side.

What it does with the customer

  • Answers product questions in a natural tone, drawing on the catalog
  • Shows items and categories, presenting the options with photos and the details available
  • Checks inventory and systems: connects to your ERP or platform so the answer is accurate
  • Follows up and re-engages customers who were left waiting or stalled partway through a purchase
  • Hands off to a human: it recognizes when to escalate and passes the context straight to your rep
  • Works 24/7 on Meta's official WhatsApp, by text, voice and image

In the demo, the agent knows the public catalog on your store's website. Connecting it to your systems happens at activation.

What it answers for you, the owner

  • Sums up the day's activity and tells you what happened, without you opening a dashboard
  • Flags bottlenecks and complaints: it spots patterns and points out what needs attention now
  • Answers factually about the catalog on your site: products, terms and policies
  • Points out customers who almost bought and stalled, so you can act on it today
  • Takes operational reminders like "let me know when item X is running low"
  • When a figure doesn't come from a connected system yet, it says it's simulating — it never claims to be connected
Sample conversation

The questions that actually come in

From the customer's side and from the owner's side. Switch tabs to see both.

Store's virtual salesperson online
Hi! Do you carry this product?2:02 PM
We do. It's one of the items in the line that's in our catalog. Want me to show you the versions we have?2:02 PM
Can you show me more options?2:03 PM
Of course. I pulled three similar options, each with a photo and its description. Tell me which one caught your eye.2:03 PM
0:07 Voice note from the customer asking whether it comes in their size.
I listened to your voice note: you want it in your size. Let me confirm availability in stock and I'll get right back to you — and if you'd rather, I can place the order right here.2:05 PM
Can I buy over WhatsApp?2:06 PM
You can. I'll stay with you here through checkout. If at any point you'd rather talk to someone at the store, I'll hand the conversation over with the whole history.2:06 PM

What the agent says depends on your store's catalog and systems.

How it's set up

The conversation above comes out of this screen — and out of no flowchart

You don't draw conversation paths. You fill in who the salesperson is, what it knows, what it's allowed to do, and when it acts on its own.

Below is the setup screen for the virtual salesperson of a fictional store. The first half holds the personality — name, tone, voice, how much it may improvise, whether it remembers the customer between conversations — and the knowledge base built from the website catalog. The second half holds the tools that are switched on, which are the actions it has permission to perform, and the task schedule: the routines the agent carries out by itself, on time, with nobody pressing anything. All of it fits on one screen because the intelligence lives in the agent, not in a decision map somebody has to keep up to date.

Hal-AI · Virtual salesperson — Aurora Store Omega
Virtual salesperson setup Personality, knowledge base, tools and schedule — no flow to draw. Test conversation
Connected channel Official WhatsApp · Aurora Store Active
Name the customer sees Helen — store support
Internal description Answers the store's WhatsApp, walks the customer through the catalog, confirms availability and stays with them through checkout.
Voice profile Helen — clear voice, shop-floor pace Play sample
Temperature and creativity (0.25) Deliberately low: in a store, making up a size, a color or a delivery date is expensive. The tone stays natural; what shrinks is the room to guess.
Long-term memory On — a returning customer doesn't start over With memory on, the agent remembers what that customer already asked about and already bought, and picks the conversation up where it stopped. Off, every conversation starts clean.
Catalog knowledge base website catalog — 412 items size chart exchange and return policy delivery times and areas internal supplier list The base is built from the public content the store already publishes, and it is reindexed whenever the catalog changes. Anything internal stays switched off until someone decides to switch it on.
Tools switched on api_get_catalog api_get_inventory api_post_order GET /orders/status api_del_order Each chip is an action the salesperson has permission to perform. Switched off, it simply doesn't exist for the agent — deleting an order is not a salesperson's job.
Owner phone numbers (admins) +1 415 555-0180 +1 415 555-0132 Add This list is what separates the person asking about a size from the person asking for the day's summary. Same number, different answer.
System prompt You are Helen, a salesperson at Aurora Store. Speak American English, in short sentences and a shop-floor tone. Before stating a size, a color, a compatibility or a delivery date, check the catalog and the inventory tool. If the data doesn't come from a connected system yet, say you're simulating — never claim to be connected. When the customer asks for a person, hand off with the full conversation history.
Agent task schedule 4 routines · store time zone New task the agent runs these routines by itself, on time, with nobody pressing anything
Virtual salesperson task schedule: routine, timing, recipient and state.
TaskWhen it runsWho it reachesState
Follow up with customers who stalled mid-purchase every day · 10:00 AM conversations idle for 24h Active
Daily summary for the owner Mon–Sat · 7:30 PM owner phone numbers Active
Flag an item that's running low when it crosses the minimum owner phone numbers Simulated
Check-in after delivery 3 days after delivery the order's customer Paused
Until inventory is integrated, the task states that the number is simulated Save setup
What you're looking at

Four decisions, one screen

Top to bottom: who the salesperson is (name, tone, voice, how much it may improvise, whether it remembers the customer), what it knows (the website catalog, the size chart, the return policy, the delivery times), what it's allowed to do (the chips that are on — check the catalog, check inventory, open an order) and when it acts on its own (the schedule: follow up with people who stalled at 10:00 AM, summarize the day for the owner at 7:30 PM).

Note the chip that's off and the amber badge: deleting an order is switched off, and the inventory task states that the number is simulated for as long as the store system isn't connected.

The advantage

Changing a rule takes minutes

When the store changes its return policy, you swap one document in the knowledge base. When you decide the salesperson may open an order but not cancel one, you switch off a chip. When you want the daily summary earlier, you change a time on the schedule.

In a flow-based chatbot, every one of those changes is a new branch on the decision tree — and somebody has to draw, test and maintain that branch.

Only at Hal-AI

The schedule belongs to the same agent

The routine that follows up with a customer at 10:00 AM is run by the same salesperson who talked to them yesterday: same memory, same catalog base, same tools. It doesn't open the conversation by asking who you are.

And the owner phone numbers in the setup are what let a single number serve the customer and answer the shop owner, with no second channel and no second agent to keep in sync.

On official WhatsApp

The same number your store already advertises

Mockup of a retail conversation on WhatsApp: the customer asks about a product and the Hal-AI agent replies with options from the store catalog, staying with the customer through checkout.

Hal-AI is a Meta Business Partner and runs on the Official WhatsApp Business Cloud API. Anyone on your team can take over the conversation at any moment.

How it starts

Four steps, and your store's website is the starting point

The agent is pre-trained on the content your store has already published.

1

You give us your store's website

The website address is all Hal-AI needs to get started. No long project, no flow to draw.

2

Hal-AI pre-trains the agent

The public content on the site becomes the knowledge base: catalog, categories, terms and store policies.

3

You test the conversation

Talk to the agent as if you were your own customer. Adjust the tone and correct anything it answered off the mark.

4

It goes live on official WhatsApp

The agent moves onto your store's official WhatsApp Business number. Activation is also when the ERP and platform integrations come in.

Questions

What store owners usually ask

Can the virtual salesperson check the store's inventory?

Yes, once it is connected to your ERP or e-commerce platform. That integration happens at activation. Before then, in the demo, it answers from the public catalog on your website.

Can customers send voice notes and photos?

They can. The agent understands text, voice and images, around the clock, on Meta's official WhatsApp.

What happens when a customer wants to talk to a person?

The agent recognizes when to escalate and passes the context of the conversation to a rep. Your team can also trigger a human takeover at any moment.

Does the owner talk to the same agent that serves customers?

They do. It's one agent: it serves customers and answers the owner about the day's activity, bottlenecks, complaints and customers who stalled partway through a purchase.

Does it make up numbers when it doesn't have the data?

No. When an operational figure doesn't come from a connected system yet, the agent says explicitly that it is a simulation and never claims to be connected to your store's system.

Put a virtual salesperson on your store's WhatsApp

Give us your store's website and start talking to an agent pre-trained on your own content.