WhatsApp Business AI Agent

A WhatsApp AI chatbot for customer service

WhatsApp AI chatbot for businesses: an AI conversational agent that answers inquiries using verified business knowledge and hands off to your team.

In short

A WhatsApp AI chatbot in StoreChart answers customer inquiries on your official WhatsApp Business number, drawing from your verified business knowledge, website content and approved chat answers. In supervised mode it yields instantly when a human agent joins the conversation.

Answers from your knowledge, not a menu

Replies in free text from your business's information and policies

  • Builds knowledge from past chats: questions and answers are extracted from your history and wait for your approval before the agent uses them.
  • Auto mode: keeps answering throughout the conversation on questions like shipping times, return policy and business hours.
  • Supervised mode: the agent answers until a team member replies in the conversation, then steps aside so your team takes it from there.
  • When a customer asks for a person, the agent confirms the chat is going to your team. When the knowledge has no answer, it doesn't guess: it sends your fallback message or leaves the chat to your team.
  • Runs inside the shared WhatsApp inbox in StoreChart, where your team sees every AI reply and can take over at any time.

WhatsApp AI agent capabilities

Control over when it replies and what it knows

Answers without waiting for staff
Customers get an answer even when nobody on your team is online, without waiting for someone to become free.
Control how it answers
Choose how closely the agent sticks to your knowledge: Exact, Balanced or Expansive, which also sets how confident it must be to send on its own. Decide what it does without an answer: send a fallback message, stay silent or try anyway.
Review what the agent learned
Answers extracted from past chats are saved as pending. You approve or reject each one, alongside the knowledge you add or import yourself.
Overview screen
See how much knowledge is approved, how much is waiting for review, how many replies the agent sent and how many questions went unanswered, so you know at a glance where to work.
Knowledge gaps
When the agent can't answer a question, it records it as a knowledge gap, so you can see what's missing and add the answer.
Replies in the customer's language
The agent replies in the language the customer writes in, including Hebrew, English and Arabic, so a business with customers in several languages doesn't need an agent per language.

Set up the agent in 3 steps

From your existing WhatsApp connection to a live agent you keep improving

1
Connect business WhatsApp

The agent runs on top of the WhatsApp Business API connection you already have in StoreChart, no extra phone number or complex technical setup required.

2
Choose an operating mode

Decide whether the agent runs in auto mode, or in supervised mode where it steps aside as soon as a team member replies, and you can switch later without losing history.

3
Review and extend knowledge

Review the answers extracted from past chats, fill in the knowledge gaps the agent records, and keep your content up to date.

How a WhatsApp AI agent works

Modes: off, supervised, auto

Every chat in the inbox has its own mode. You change it from the chat itself, or with an automation that switches the agent on for every incoming message.

  • Supervised: the agent replies on its own and stops in that chat as soon as someone on your team writes in it, even mid-draft. The ready-made automation starter uses this mode.
  • Auto: it keeps answering even after a team member has joined.
  • Off: your team answers alone.
  • Several agents: you can run several agents on one number, such as sales and service, with a default agent. An automation can assign a chat to an agent by conditions, such as existing customer versus new enquirer.

When it replies and when it hands over

The agent only replies inside the 24-hour window opened by the customer's message. It never messages a customer first, so it doesn't replace templates.

  • Text and voice notes: a voice note is transcribed before the agent answers.
  • Message bursts: several short messages in a row get one reply, after a few seconds' wait.
  • Asks for a person: the agent tells the customer the team will pick it up.
  • Existing order or live stock: the agent says honestly that the team needs to check and flags the chat. It isn't connected to order status or stock.
  • No answer in the knowledge: your fallback message or silence, as you chose. The question goes onto the unanswered-questions list.
  • Everything else: an answer from the knowledge, with a product link and photos you set up.

A thank-you, or "I've ordered, thanks", doesn't count as a request and doesn't hand the chat to anyone.

Where the agent's knowledge comes from

All the knowledge sits in the agent's Knowledge tab. That's also where you set its voice: name, role, goal, style, welcome message and sample answers.

  • Site scan: a product list and documents by topic, such as shipping and pickup, payment, returns, installation and guides.
  • Past chats: questions and answers from the number's history are saved as pending until you approve them.
  • Manual knowledge: text you write and photos the agent may send.
  • Background re-reads: StoreChart re-reads the site from time to time, so a changed price or policy gets updated.
  • Coverage card: flags topics with no document at all, such as returns or payment.
  • Fix from the inbox: a correction to an agent reply is saved as knowledge and used in later chats too.

Testing the agent before going live

The agent's settings let you check answers before customers see them.

  • Test chat: it also uses changes you haven't saved yet, and shows at once how the answer changes.
  • Knowledge adherence: Exact, Balanced or Expansive. The choice also sets how confident the agent must be to send on its own.
  • Reply length and fallback: change them and check in the same chat.
  • Example conversations: build a set from real chats you answered well, and run a check that shows which messages the agent passes and which it fails.

Example: a furniture store

Illustrative only. This is how a furniture store moves to automated replies step by step:

  1. It runs the agent in supervised mode on all messages.
  2. In the first week, the unanswered-questions list shows six enquiries about home assembly.
  3. It writes one document on assembly and runs the test chat again.
  4. Only once the answers look right does it move some chats to auto mode.

A WhatsApp chatbot without menus

A button-based WhatsApp bot means building and maintaining flows. Here, the ongoing work is keeping knowledge current.

  • Free text: customers write normally, and the agent answers from your business's knowledge.
  • Customer's language: it replies in the language the customer writes in, including Hebrew, English and Arabic.
  • Website chat: managed separately. A conversation that starts there and gets pointed to WhatsApp arrives in the inbox where this agent works.

Who it's for and what's not included

  • Good fit: a store with many repeat questions about products, sizes, shipping and returns that wants automated WhatsApp replies in the evening and at weekends too.
  • Weaker fit: a business where most messages are "where's my order", because the agent hands those to the team.
  • Actions: the agent won't change an order, cancel one or create a payment link.
  • Allowance: usage counts toward your plan's AI allowance.

Runs on your existing WhatsApp connection

Frequently asked questions about the WhatsApp AI agent

Does the agent reply on its own?

Yes. In both modes the agent sends its replies itself. In auto mode it keeps answering for the whole conversation; in supervised mode it stops as soon as someone from your team replies, so a human takes it from there.

What does the agent do when it has no answer?

If the agent isn't confident enough, it follows the rule you set, send your fallback message or stay silent, and records the question as a knowledge gap. Questions about a specific order or live stock are handed to your team with an honest message instead of a guess.

Do I need to build conversation flows?

No. The agent replies in free text from your business's knowledge: your site scan, past-chat answers you approved and content you added. There's no menu tree to build and maintain.

Is this the same as the website chatbot?

No. This is a separate module that replies on WhatsApp and is connected to your WhatsApp inbox, while the website chatbot is a chat window installed on your store's site. Each has its own knowledge and settings, and you can run both.

How do I start a WhatsApp AI agent safely?

The agent runs on your existing WhatsApp connection, so no extra number is needed. Run a scan of your website first, review the answers extracted from past chats, and start in supervised mode to check the replies before switching to auto mode.

Can I limit the agent to certain topics?

Not by a topic list. It answers from the knowledge you gave it, and in Exact mode it replies word for word from the stored answer. A customer who asks for a person, or a question about a specific order or live stock, goes to your team.

Does the agent know which language to reply in?

Yes. It replies in the language the customer is writing in, rather than one fixed language for every conversation.

A WhatsApp AI agent on your existing number

The agent runs on your existing WhatsApp Business connection