Build a WhatsApp FAQ Bot with n8n and DeepSeek
Your customers already ask the same ten questions on WhatsApp every day — opening hours, delivery, prices, payment. This tutorial wires n8n to the DeepSeek API so those questions answer themselves, for about USD 0.33 per 1,000 replies. The workflow is three nodes, and you can import it ready-made below.
What you need
- n8n — self-hosted or n8n Cloud; the free self-hosted version is fine.
- A WhatsApp Business number — via the Meta WhatsApp Cloud API, which n8n's WhatsApp nodes connect to. (A test number from Meta works for this tutorial.)
- An AkiliNode API key — sign up free, USD 0.10 test credit included; top up later with M-Pesa if you are in Kenya (see DeepSeek API in Kenya). AkiliNode is an OpenAI-compatible gateway, which is why the setup below is just a URL and a key.
How it works
Three nodes, in a line:
- WhatsApp Trigger — fires whenever a customer messages your business number.
- HTTP Request → DeepSeek — sends the customer's message, plus a system prompt containing your FAQ, to
https://api.akilinode.com/v1/chat/completionswith modeldeepseek-flash. - WhatsApp (Send Message) — sends the model's answer back to the same customer.
Step 1 — Get your AkiliNode API key
Create a free account, verify your email, then open the console's token page and create a token. Keep it handy — it goes into the workflow in Step 3. Full walkthrough in the quickstart.
Step 2 — Import the workflow
Download the workflow file from the button at the top of this page. In n8n, open Workflows → ⋯ → Import from File and select it. You will see the three nodes plus a sticky note with the same setup steps.
Step 3 — Connect WhatsApp and paste your key
- In the WhatsApp Trigger node, create a credential with your Meta WhatsApp app details. n8n shows a webhook URL for the trigger — paste it into your Meta app's webhook settings so incoming messages reach n8n.
- In the Ask DeepSeek node, replace
YOUR_AKILINODE_API_KEYin the Authorization header with your token from Step 1. - In the Send WhatsApp Reply node, connect the same WhatsApp credential. The recipient and text are already wired by expressions: the customer who just messaged you, and the model's answer (
choices[0].message.content).
This is the exact request the middle node sends — note the custom base URL is the whole trick, and it works because AkiliNode is OpenAI-compatible:
POST https://api.akilinode.com/v1/chat/completions
Authorization: Bearer YOUR_AKILINODE_API_KEY
Content-Type: application/json
{
"model": "deepseek-flash",
"messages": [
{ "role": "system", "content": "You are the WhatsApp assistant for ... Answer using ONLY the FAQ below. ..." },
{ "role": "user", "content": "<the customer's WhatsApp message>" }
]
}
Step 4 — Write your FAQ into the system prompt
The system prompt in the workflow ships with sample questions (hours, delivery, payment, returns). Replace them with yours. Two rules that matter:
- Tell the model to answer only from the FAQ and to hand off to a human (with a real contact) when the answer is not there. This keeps a small model from inventing prices or policies.
- Keep the FAQ short. It is sent with every single request, so a bloated FAQ is a recurring cost on every reply. Ten crisp Q&As beat two pages of prose.
Step 5 — Test, then go live
Send your business number a WhatsApp message ("What time do you close?") and watch the execution in n8n. Check the field paths in the expressions against your trigger's output — WhatsApp payload shapes vary slightly between setups. When the reply lands, switch the workflow to Active and swap the test webhook for the production one.
What it costs
With deepseek-flash at AkiliNode prices (USD 0.33 input / USD 1.32 output per 1M tokens), a typical FAQ exchange — about 400 tokens in (FAQ + question) and 150 tokens out — costs roughly USD 0.00033:
| Volume | Model cost on AkiliNode |
|---|---|
| 1,000 replies | ≈ USD 0.33 |
| 10,000 replies | ≈ USD 3.30 |
| One USD 5 top-up | ≈ 15,000 replies |
WhatsApp conversation fees and n8n hosting are separate and depend on your setup. If some questions are too hard for Flash, route just those to deepseek-v4-pro (USD 1.45 / 4.36 per 1M) — same workflow, one word changed.