Route WhatsApp support replies through AI and Telegram

An n8n flow drafts WhatsApp answers from a knowledge base and sends uncertain cases to a human on Telegram.

Done with n8nยท Intermediateยท Jul 2026

What they did
They built an n8n workflow that reads each new WhatsApp message, adds the conversation history and company knowledge base, and asks OpenAI for a grounded reply. A confidence check then decides whether the agent should answer directly or hand the case off. When the score is not strong enough, the full thread is sent to the human team in Telegram.
How it went
The setup was meant to handle routine support questions, provide coverage at all hours, and keep people involved when the answer looked unreliable. There were no published usage numbers, and the exported n8n workflow file was still being added.
Tips
You will need to design your own confidence test, since the repo does not explain how it is calculated. Make sure escalations include the full chat history so a person can take over without asking the customer to repeat themselves.

Try it yourself with n8n

Set up an n8n workflow that watches incoming WhatsApp messages for [company name]. Pull in the current conversation plus [knowledge base source], then draft a reply only from that information. Before sending anything, check whether the answer is confident enough; if it is not, or if the customer asks for a person, pause and send the full thread to me on Telegram for approval and reply. If the answer is approved, post it back to WhatsApp. Test it with one simple question and one tricky question, and confirm the tricky one is escalated with context before any message is sent.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on github.com โ†—

1 more report of this

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.