How to build a WhatsApp AI Agent in AiSensy

Most WhatsApp chatbots work on fixed keywords. A customer types "refund" and gets a refund policy. That works for simple queries, but customers don't always talk in keywords. They say things like "I ordered something last week and haven't received it" or "Can I change my plan mid-month?"

AiSensy's AI Agent Builder is built for exactly these situations. It understands what your customer is actually asking, finds the right answer from your business knowledgebase (help center), and can even take real actions, like checking an order status, booking an appointment, or fetching a quote - all without writing a single line of code.

In this tutorial, we’ll walk you through how you can build your own AI Agent on WhatsApp using AiSensy.

By the end of this, you’ll have a fully functional AI Agent that can understand user queries, respond intelligently, and even take actions automatically.

Note: This is a big guide, so stock up on your coffee & snacks & let's begin!

Before we start building, it’s important to understand how the system works.

Every AI Agent on AiSensy runs on a combination of three core components:

AI Routing (optional) — This lives above all your flows, at the account level. If you have multiple flows, AI Routing reads incoming messages, figures out what the customer wants, and sends them to the right flow — without any keyword triggers. You only need this when you have more than one flow.

Don't worry. Setup for each of these has been explained in detail and you will learn how to set all these up very easily.

So, let's make your WhatsApp AI agent.

1. Click on Flows in the left sidebar

Click the Flows icon in the left navigation rail. This opens the Flow Builder playground, your home for all chatbot flows and the new AI Agent capabilities.

On this page you'll see the Active Flows card (tracks how many flows are live and how many are AI-Enabled), the AI Messages card (where you purchase the AI Messages addon that fuels intelligent responses), and a + Create Flow button.

2. Click "+ Create Flow"

Click on + Create Flow to start a new flow.

4. Open the Knowledge Base tab

Once a flow is open, four tabs appear at the top:

• BUILDER: the familiar drag-and-drop canvas where you design conversation flows.

• KNOWLEDGE BASE: train this flow's AI on your business data.

• TOOL CALLING: configure API integrations the AI can trigger.

• AI ORCHESTRATOR: tell the AI how to behave with structured instructions.

Inside the flow editor, click the KNOWLEDGE BASE tab at the top.

5. Create a new Knowledge Base

Click + Knowledge Base in the top right. Give it a clear name that's easy to identify.

Something like "Product FAQs" or "Shipping & Returns" works well.

3. Add your content

Click the arrow to expand your Knowledge Base. You'll see three ways to add content:

4. Wait for indexing to complete

Once you add content, it needs to be processed before your agent can use it. Each source will show one of three statuses:

5. Enable the Knowledge Base

Use the toggle next to each KB to turn it on or off.

The pencil icon lets you rename it, the trash icon deletes it, and the arrow lets you add more sources to it anytime.

Tool Calling is what turns your agent from a question-answering bot into one that can actually do things. With Tool Calling, your agent connects to your tools (CRMs, ecommerce platforms like shopify etc) and brings back live data, like order status, appointment slots, customer account details, or a live price from your system, then sends the result straight to the customer on WhatsApp.

IMPT NOTE: If your agent only needs to answer questions from the Knowledge Base, you can skip this step entirely.

1. Open the Tool Calling tab

Inside the flow editor, click the TOOL CALLING tab. Click + Add New Tool to get started.

2. Name your tool and describe what it does

Give your tool a clear name, like "Check Order Status" or "Fetch Policy Details". Then fill in the Tool Description. This is important: the AI reads this description to decide when to call the tool. Write it in plain English, like: "Fetches the current status of a customer's order using their order ID."

3. Set up your API request

Under REQUEST CONFIGURATION, fill in your API details:

4. Define the input parameters

Under REQUEST PARAMETER CONFIGURATION, add the inputs your API needs. For each one, fill in:

5. Add Request Fields

Under REQUEST FIELDS, click 🪄 Auto-detect to let AiSensy auto-fill the fields from your endpoint, or click + Add Field to add them manually.

Specify each field's name and type. These are the values your tool sends to your API. The AI will automatically extract them from the conversation when the tool is called.

6. Map the API response to attributes

Under RESPONSE ATTRIBUTE MAPPING, click + Add Mapping to capture values from the API response and save them for use later in the flow. For example, map response.policy_number to a policy_number attribute to use it further down the conversation.

7. Configure the WhatsApp response

Under RESPONSE CONFIGURATION, choose how your agent replies after getting the API data:

Use {{field}} syntax in the message body to pull in values from the API response.

The live WhatsApp preview on the right shows exactly how your message will look.

8. Test and Save

Click the ▶ Test button to validate your API call works. Then hit Save to activate the tool. Your agent will now call this API whenever a relevant query comes in.

Note: Tool Calling works with any REST API. If it works in Postman, it works with your AI Agent.

The AI Orchestrator is where your agent really comes to life. It reads each conversation, weighs up the available knowledge and tools, and picks the best action, whether that's pulling an answer from the KB, calling a tool, escalating to a human, or writing a response from scratch.

This is also where you define your agent's personality, scope, and rules. Think of it as your agent's briefing document.

1. Open the AI Orchestrator tab

Click the AI ORCHESTRATOR tab inside your flow editor. You'll land on the Agent Instructions screen.

2. Choose Sections or Raw mode

For better customization, we recommend that you use the raw mode to insert your prompt and guide the agent behaviour.

3. Fill in "About your business"

Tell the agent what your business does, who your customers are, and what they typically reach out for. Be specific. The more context you give, the better your agent will perform.

Example: "You are the customer support assistant for AiSensy's WhatsApp Marketing Lite Plan. Customers reach out with questions about the plan, setup, QR Kit delivery, automations, broadcasting, API, KYC, and general usage. Your job is to help them get started and resolve their doubts quickly.

Bot Persona & Tone"

4. Fill in "Tone & personality"

How should your agent sound? Think about your brand voice.

Example: "Friendly, warm, and solution-oriented. Keep replies short and conversational — no long paragraphs. Use simple language, avoid technical jargon unless the customer uses it first. Acknowledge frustration before offering solutions. Always confirm whether the user is satisfied before ending the conversation."

5. Fill in "How the bot should work"

This is your step-by-step playbook for the agent: what it should do in different situations and which tools to call or how to react to complex questions.

Example: "1. For renewals, look up the user's existing policy first. 2. Always confirm vehicle details before quoting. 3. Use the integrated crm tool to fetch the latest price. 4. Never quote without the vehicle registration number."

6. Fill in "What to avoid"

Topics to refuse or hand off to a human.

Example: "Only handle motor insurance queries. Don't answer questions about health, life, or travel insurance. Escalate complex disputes to a human agent immediately."

7. Fill in "Anything else"

Extra rules, like pricing display formats, legal disclosures, escalation triggers, response length. These are basically the custom instructions for your WhatsApp AI agent, extra rules that ensure fine responses.

Example: "Always show prices in INR with GST included. If a claim amount exceeds ₹5 lakhs, escalate immediately."

8. Use Generate or Templates

Not sure what to write? Click Generate and AiSensy will auto-draft instructions based on your Knowledge base and tools. Or browse Templates for ready-made starting points built for common business types.

Pro Tip: Keep your instructions focused on your business domain. AiSensy already handles general conversational rules, formatting, and content safety. You don't need to cover those yourself.

9. Save

Hit Save. Your AI Orchestrator is now configured. Use the Undo button at any time if you want to roll back a change.

If you have multiple flows for different business use cases, like one for product queries, one for complaints, one for bookings. AI Routing is what automatically sends each customer to the right flow without them needing to type a specific keyword.

Example: A customer types "Hi, I'm looking for a 2BHK in Gurgaon." AI Routing reads that message, identifies the intent as Property Search, and routes the conversation to that flow, no keyword needed.

AI Routing lives at the Flow Builder index level, not inside individual flow editors. If you only have one flow, skip this step.

1. Open the AI Routing tab

Go to the Flow Builder index page and click the AI Routing tab. You'll see the AI Router toggle at the top.

2. Turn on the AI Router

Switch the toggle on to enable intent-based routing across all your flows.

3. Auto-generate your routing prompt/ Add your routing prompt manually

Click Auto-Generate from Flows. AiSensy drafts a routing prompt based on your active flows. You'll see five editable sections:

Note: You can draft all this manually as well.

4. Name your flows clearly

If you see a warning about unnamed flows, rename them from the Your Flows page. Vague names like "Flow 1" reduce routing accuracy.

The clearer your flow names and descriptions, the better AI Routing performs.

5. Save

Click Save Prompt to lock in your routing logic.

Before you go live with real customers, test your agent properly. A few minutes of testing can save you from a lot of awkward customer conversations later.

How testing works

Every flow comes with 500 free AI messages for testing on up to 3 designated Testing Numbers. In testing mode, your agent only responds to those numbers. Real customers won't see any AI responses until you explicitly go live.

Test checklist

Go through these before going live:

Once you're satisfied with testing:

Your agent will now respond to all incoming messages from real customers.

AI Agent Builder runs on an AI Messages addon. Here's how it's structured:

Plans

Pack sizes (number of messages, not pricing)

Note: This is the number of messages you get in each pack. For pricing, you can login to your dashboard & see the pricing based on your country/ region.

Billing

Monthly, Quarterly (discounted), or Yearly (best value). Activates immediately after payment. Login to your AiSensy dashboard.

For exact pricing, check your AiSensy dashboard.

How AI messages are counted

Every intelligent response, whether a KB answer, a tool call, or Orchestrator reasoning, counts as one AI message from your monthly quota. Usage resets at the start of each month.

Important: If you go over your quota, extra messages are billed as excess usage.

Your agent does NOT stop responding. Customers are never left hanging. Unused messages don't carry over to the next month.

Note: Remaining AI messages in the month get lapsed. They do not transfer to the next month.

Free testing allowance

500 free AI messages per flow for testing on up to 3 Testing Numbers. Test as much as you need before purchasing.

No AI handles every situation perfectly, and sometimes a customer just needs a real person. AiSensy handles this automatically.

The agent escalates in four situations:

What happens during escalation:

You can also build explicit escalation paths using the Human Handover node inside any flow, useful when you want to escalate based on specific topics or user actions.

What is WhatsApp AI Agent Builder?

AiSensy's next-generation chatbot system for WhatsApp. Unlike keyword chatbots that only respond to pre-set triggers, AI Agent Builder understands what customers actually mean, learns from your business content, and can take real API-driven actions, all without any coding.

Do I need coding skills?

Not at all. If you can type, you can build an AI Agent. AI Agent Builder is fully no-code. You set everything up through the drag-and-drop Flow Builder and a text-based Orchestrator instructions form.

How is it different from the classic Flow Builder?

The classic Flow Builder works on exact keyword matching and fixed conversation paths. AI Agent Builder adds natural language understanding, Knowledge Base retrieval, live API Tool Calling, and dynamic routing, so your agent handles open-ended conversations instead of just scripted ones.

Can I use both?

Yes. You can have some flows on the classic builder and others on AI Agent Builder within the same account. AI Routing works across both.

What happens if the AI can't answer?

It escalates to a human agent and stops responding to that user until your team replies or the chat auto-closes after 24 hours.

How do I add more KB content later?

Open the Knowledge Base tab inside your flow editor, expand the relevant KB, and click + Article, + URL, or + Document. You can add, edit, or remove content at any time.

Can I update KB content after going live?

Yes. Edit or delete any source at any time. AiSensy re-indexes automatically so your agent always uses the latest content.

What APIs can Tool Calling connect to?

Any REST API that accepts standard HTTP requests. If it works in Postman, it works with Tool Calling.

Is there a free trial?

Your Flow builder now comes with 500 free AI messages for testing on up to 3 Testing Numbers. Purchase an AI Messages addon only when you're ready to go live with real customers.

Can I have multiple Knowledge Bases per flow?

Yes. Create as many KBs as you need and toggle each on or off independently. Useful for enabling different content sets for different conversation types.

Does the AI know which KB to search?

Yes. The AI Orchestrator searches all active KBs simultaneously and combines results across semantic, keyword, and topic search to surface the most relevant answer.

What languages are supported?

Our WhatsApp AI agent currently only replies in english. However, it easily understands questions asked in multiple languages like Hindi, Tamil, Telugu etc.

How does AI Routing handle unrecognised intent?

If confidence falls below 0.75, the router doesn't route. The customer receives a fallback message, and you can configure what that says.

What counts as an AI message?

Any response involving intelligence, like a KB answer, a tool call, or Orchestrator reasoning, counts as one AI message from your quota.