WhatsApp AI Meeting Assistant

The problem

Scheduling a meeting usually means a long back-and-forth: “Are you free Tuesday?”, “What about 3pm?”, then someone has to check the calendar, create the event and send the invite. People also send requests the way they actually talk — a voice note, a screenshot of an email, or a quick text.

The solution

A WhatsApp assistant powered by an n8n AI agent that handles the whole thing in the chat:

• WhatsApp Trigger receives every message; a Switch routes it by type.
• Voice notes are downloaded through the WhatsApp media API and transcribed with OpenAI.
• Images are downloaded and analyzed by a vision model, so a screenshot of an invitation becomes structured text.
• Everything is normalized into one text field and passed to the AI agent, which keeps conversation memory so follow-ups like “make it 30 minutes later” just work.
• The agent decides which tools it needs:
  • Google Sheets — read existing meetings and check for scheduling conflicts, then log the new one
  • Google Calendar — create the event
  • Gmail — send the notification and calendar invitation
  • Google Search (SerpApi) — only when the request needs current, external information
• The final answer goes back to the user on WhatsApp.

Technical details

One agent, many inputs

Instead of three separate bots, the media branches (audio → transcription, image → vision) converge into the same text field. The agent never needs to know how the request arrived — which keeps the prompt and the tool logic simple.

Tool-calling instead of hard-coded paths

The agent chooses tools per request. “What do I have tomorrow?” only reads Sheets; “Book a call with Ahmed Thursday at 2” checks conflicts, creates the event and emails the invite.

The result

Scheduling becomes one message — typed, spoken or photographed — instead of a thread of back-and-forth, with conflicts caught before anything is booked.