The pieces of the workflow
- WhatsApp trigger: receives every incoming message.
- AI agent node: calls the model (Claude, GPT, Gemini…) with your instructions.
- Memory: that contact's history, stored in a database, so the agent isn't replying blind.
- Tools: what the agent can look up (prices in a spreadsheet, availability in a calendar).
- Send: the node that posts the reply to WhatsApp.
Two ways to connect WhatsApp
The official API (WhatsApp Business Cloud). n8n has nodes for it. You need a Meta developer account, business verification and a dedicated number. Replying to people who message you within 24 hours is straightforward; starting a conversation yourself requires approved templates and Meta charges per message. More in our guide to the WhatsApp Business API.
An unofficial QR bridge. Open-source projects such as Evolution API or WAHA run on your server, link your number like WhatsApp Web, and n8n talks to them. More flexible, and you keep your usual number, but it isn't official: Meta can restrict the number if it's misused.
What you'll need to run it
- A server running around the clock (self-hosted n8n, or n8n Cloud plus somewhere to host the bridge if you use one).
- An AI model account, paid per use.
- A database for memory and for tracking who needs a follow-up.
- Someone who notices when something breaks, including at weekends.
What shows up afterwards
- The customer sends five messages in a row and the bot replies five times.
- Voice notes, photos and documents that need handling separately.
- The bot replying to your mum on your personal number.
- Knowing when to stop and hand over to a person.
- Following up two days later with people who went quiet: another workflow, with its own database.
- The Saturday the WhatsApp session drops and nobody notices.
Each point is an afternoon's work. None of it is impossible; it's ongoing maintenance.
When it's worth it
Build it in n8n if you enjoy it, have time to maintain it or your case is very specific. If what you want is an agent selling this week, MultiChats already handles everything above (waits for the customer to finish typing, transcribes voice notes, only touches sales conversations, hands over to a person, follows up and watches the health of each number) from €150 a month excl. VAT, with no contract. It also connects by QR rather than the official API, so the same caution about Meta restrictions applies. More in how to build a WhatsApp AI agent.
Frequently asked questions
Is n8n free?
The open-source edition can be self-hosted for free; n8n Cloud is paid. Either way, you pay separately for the AI model and, with the official API, for Meta's template messages.
Which AI model should I use in n8n?
Any of the major ones works. What changes the result most is the instructions and examples you give it.
n8n or Make?
Both work. n8n can be self-hosted and is widely used for agents; Make is simpler to start with. The pieces are the same.
Can an n8n WhatsApp bot get my number banned?
With the official API, the risk comes from low quality ratings and blocks. With an unofficial QR bridge, Meta can restrict a number that sends in bulk or gets reported. Keep volumes human and only message people who contacted you.