Building an AI Agent for WhatsApp, Instagram, and More

An eGrow AI Agent is an automated assistant that chats with your customers directly inside your inbox — on WhatsApp, Instagram, and every other channel you connect. It understands the customer’s language, answers questions using your own knowledge base, searches your product catalog, and can even take and manage orders end to end — create an order, confirm it, modify it, track it, or cancel it — and it can answer WhatsApp voice calls. You stay in control: you decide its personality, which channels it handles, which tools it can use, when it is available, and what it is allowed to do.

This guide explains every part of building an AI Agent, step by step. You will create your agent at https://v2.egrow.com/ai-agents/inbox.

What your AI Agent can do

Reply automatically to customers on WhatsApp, Instagram, and other connected channels, in the customer’s own language.
Access your products — search your catalog, describe items, quote live prices and stock, and send store links so customers can buy.
Take and manage orders — create, confirm, update, add/replace items, add notes, track, and cancel orders.
Look up customers — order history, last order, and saved details.
Answer voice calls on WhatsApp with a natural AI voice.
Reply to voice messages with text or audio.
Follow your rules — a custom persona/prompt, a private knowledge base, working-hours availability, and escalation to your human team when needed.

Before you begin

At least one connected channel (a WhatsApp or Instagram account). Connect one under https://v2.egrow.com/integrations — for WhatsApp specifically see https://v2.egrow.com/integrations/whatsapp, and review linked accounts at https://v2.egrow.com/integrations/connections.
If you want the agent to sell and answer product questions, add your items to the catalog at https://v2.egrow.com/products and keep stock accurate at https://v2.egrow.com/products/inventory.
Access to the AI Agents page: https://v2.egrow.com/ai-agents/inbox.

Step 1: Open the AI Agents page

Go to https://v2.egrow.com/ai-agents/inbox. This lists your existing agents. Click Create (or open an existing agent to edit it) to open the agent builder. The builder has a General section at the top, followed by five tabs: Connections, Tools, Datasources, Availability, and Calls. We’ll cover each one.

Step 2: Configure the General settings

These fields define who your agent is and how it speaks.

Agent Name (required) — an internal name, e.g. Sales Assistant.
AI Model (required) — the AI engine that powers the agent. Each model shows its own pricing (per text and audio usage); pick one and you can change it later.
Gender — affects the text-to-speech voice used for audio replies and calls.
Language — leave as Auto detect to have the agent reply in whatever language the customer writes in, or set a fixed language.
Products Languages — the language(s) your product information is written in, e.g. Arabic, French, English. This helps the agent match products correctly when customers ask in different languages.
Audio Language — the locale used when the agent replies with audio, e.g. ar-MA for Moroccan Arabic.
Preset prompt — turn this on to use eGrow’s built-in, sales-ready instructions. Turn it off to write your own from scratch.
Answer audio messages — when on, the agent can reply with a voice message instead of text.
Custom Prompt — free-text instructions that shape the agent’s personality, tone, and rules (for example: “You are a friendly assistant for our store. Always greet the customer, never promise discounts, and confirm the delivery city before creating an order.”). This is where you make the agent yours.

Step 3: Connect your channels (Connections tab)

Open the Connections tab. Here you see every channel you’ve linked to eGrow — each WhatsApp and Instagram account. Toggle on the connections this agent should handle. The agent will automatically reply to conversations that arrive on the connections you enable, and ignore the rest.

If a channel is missing, connect it first at https://v2.egrow.com/integrations/connections, then return here.

A single agent can serve several numbers at once, and you can run different agents for different connections (for example, one for sales, one for support).

Step 4: Give your agent tools (Tools tab)

Tools are the actions your agent can take. On the Tools tab you enable the built-in tools you want, adjust their settings, and optionally add your own custom tools. The agent reads each tool’s description and decides on its own when to use it during a conversation.

Product tools — how your agent uses your catalog

Because the agent is connected to your product catalog at https://v2.egrow.com/products, it can find items, quote real prices and stock, and guide customers to buy. The product tools are:

Important — the agent only reads products that have AI enabled. A product is completely invisible to the agent until you switch on AI for it. Open https://v2.egrow.com/products, select the products you want the agent to sell and choose Enable AI — or use Enable AI for all to turn it on for your whole catalog at once. You can enable or disable AI per product, or in bulk, at any time. If the agent can’t find an item, the first thing to check is that AI is enabled on that product.

Search products — used when a customer asks for a specific item, name, or category. In its settings you can choose:
- Search method — AI (LLM), where the agent understands intent and finds the best matches, or Query, a direct keyword lookup.
- Max products — return All matches or Limit to a set number (for the AI method).
- Include out of stock — whether out-of-stock items may appear in results.
Available products — returns a general catalog when a customer asks broadly, e.g. “what do you have?”. In its settings:
- Max to user — how many products the agent may show the customer.
- Source — AI-chosen (the agent picks what to show, up to a Max to AI limit) or Specific products (you hand-pick the exact items using the Browse picker).
Get product information — detailed pricing, stock status, and variant options for one specific product.
Send product link — sends the customer a direct store link so they can order the product from your website.

Keep your catalog and stock up to date at https://v2.egrow.com/products and https://v2.egrow.com/products/inventory. The agent quotes whatever is in your catalog, so accurate prices, variants, and stock levels mean accurate answers.

Order tools — a full ordering assistant

With these enabled, the agent can handle the entire order lifecycle inside the chat:

Create new order — builds a new order from the items the customer wants and calculates the total.
Confirm order — finalizes an order only after the customer explicitly says “yes” to the summary.
Get order information — retrieves an order’s details by order number.
Track order — returns tracking status, shipping location, and delivery details.
Update order info — changes the delivery address and city.
Replace order products — swaps items or updates quantities and recalculates the total.
Add products to order — adds items to an existing order.
Add order note — attaches a note or special instruction.
Cancel order — cancels an order with a reason.

Customer tools

Save customer information — stores the customer’s name and details when they share them.
Get customer orders history — the customer’s last 10 orders.
Get customer last order — their most recent order.
Search customer orders — find a customer’s orders within a date range.

Conversation & operations tools

Handle customer request — when a customer asks for something the agent can’t do, it logs and escalates the request (with a note) to your team instead of guessing.
Add label to conversation — tags the conversation so you can categorize and filter it in the inbox.
Send email notification — sends an internal email alert to your team when something noteworthy happens. Set the Recipient emails in the tool’s settings.
Schedule self prompt — lets the agent schedule a follow-up for itself at a future time; when that time arrives it is re-invoked with the queued instruction (useful for “check back tomorrow” behavior).

Custom tools — connect your own systems

Need the agent to call your own API (for example, check a warranty, look up a loyalty balance, or trigger an external workflow)? Click Add custom tool and define:

Name — in snake_case, e.g. check_warranty. The AI calls the tool by this name.
Description — what the tool does. The AI reads this to decide when to call it, so be clear.
Authentication — API key (header name + value), Bearer token, Basic (username + password), or none.
Parameters — the inputs the AI must collect and pass to your endpoint.
Communication — the HTTP method (GET/POST…) and endpoint URL.
Timeout — how long to wait for your API to respond.

Step 5: Add knowledge (Datasources tab)

On the Datasources tab, click Add to give the agent private knowledge it can use to answer questions. Each data source has a Title and a Content body, for example:

Shipping Policy — delivery times, fees, and covered regions.
Return & Refund Policy — conditions and timeframes.
FAQs — common questions and their answers.

The agent draws on these entries when replying, so it answers with your policies and facts instead of guessing. Add as many as you need, and edit or remove them anytime.

Never put product information in Data Sources or in the Custom Prompt. Prices, stock, and variants change constantly — hardcoding them here makes the agent quote stale or conflicting details and leads to wrong answers and mis-priced orders. Keep all product data in your catalog at https://v2.egrow.com/products so the agent always reads it live through the product tools. Use Data Sources and the Custom Prompt only for things that are not products — shipping and return policies, FAQs, tone, and rules.

Step 6: Set availability (Availability tab)

Control when the agent replies. On the Availability tab:

Schedule Timezone (required) — e.g. Africa/Casablanca or Europe/Paris.
Weekly windows — click Add window to set active hours per day. Days you don’t list are closed all day; use 24:00 for end-of-day. If you add no windows at all, the agent replies 24/7.
Holidays — optional entries that replace a specific day’s normal window (for example, to close on a public holiday).

Outside its available hours, the agent stays quiet so conversations can wait for your team.

Step 7: Enable voice calls (Calls tab)

Your agent can also answer WhatsApp voice calls. On the Calls tab, enable call handling and configure:

Language (required) — the language the AI speaks on calls.
Voice (required) — the voice the AI uses.
Max Call Duration (seconds) — calls longer than this end automatically (between 60 and 3600).
Permission Wait Limit (seconds) — how long to wait for the customer to accept the call before giving up (leave blank to wait indefinitely).
Schedule — a timezone and optional weekly windows, exactly like Availability. With no windows, the AI answers calls 24/7.

The AI answers calls on every WhatsApp connection linked to this agent in the Connections tab.

Step 8: Save and activate

Click Save. Set the agent’s status to Active to put it to work; set it to Inactive to pause it at any time. Once active, it immediately begins replying to new conversations on its connected channels, within its availability hours.

Proactive follow-ups: Scheduled Prompts

Beyond replying, your agent can reach out on a schedule. Open https://v2.egrow.com/ai-agents/inbox/scheduled-prompts to create scheduled instructions — for example, a follow-up message to customers who didn’t confirm their order, sent at a chosen time. You can review and filter scheduled prompts by status from that page.

How it works in the inbox

Once your agent is active, it handles conversations automatically in your inbox at https://v2.egrow.com/inbox. For each incoming message the agent reads the conversation, decides which tools to call (search a product, create an order, look up an order status, and so on), and replies — in text or audio. Your team can always jump in and take over a conversation manually. Every response is recorded as a run with the exact steps and tool calls the agent made, so you can review its decisions and fine-tune its prompt, tools, and knowledge.

Best practices

Write a clear Custom Prompt. Spell out the tone, what the agent may and may not do, and the steps to follow (e.g. always confirm the city before creating an order).
Only enable the tools you need. Fewer, well-chosen tools make the agent’s behavior more predictable.
Enable AI on the products you want it to sell at https://v2.egrow.com/products, and keep stock accurate at https://v2.egrow.com/products/inventory — the agent only sees AI-enabled products, and quotes whatever the catalog says.
Keep product details out of Data Sources and the Custom Prompt. Product info belongs only in the catalog, so the agent always reads it live and never quotes stale prices or stock.
Feed it real policies as Data Sources (shipping, returns, FAQs) so non-product answers match your business.
Set availability to match your operating hours, and use Handle customer request so anything the agent can’t do is escalated to your team rather than mishandled.
Test in a real conversation and review the run history, then refine the prompt, tools, and knowledge.

That’s everything you need to build a capable AI Agent. Start now at https://v2.egrow.com/ai-agents/inbox.