For an agency that sets up Instagram booking agents for med spas · ManyChat · 7 n8n workflows · OpenAI gpt-4o-mini · GoHighLevel · Google Calendar · Gmail
An Instagram agent built to book consults in the DMs.
I built this agent for an agency that sets up Instagram booking agents for med spas. It was made to answer DMs from each spa’s own knowledge base and book consults inside the thread. The engagement ended in July 2026.
Swipe sideways to see the whole thread.
Who it was for
An agency that sets up Instagram booking agents for med spas. I worked with them as a freelancer from April to July 2026.
Each spa had its own knowledge base, a Google Doc that the spa’s team edited. I built the full agent to read the intent of each Instagram DM, answer from that knowledge base and book consults inside the DM thread, with an OpenAI model (gpt-4o-mini).
How booking worked in the thread
Booking took five steps, all inside the DM.
Service
It asked which service the person wanted.
Preferred time
Then it asked what time worked best, such as a weekday morning.
Two open slots
It offered two open times from the spa’s calendar.
The intake form
It asked how the person wanted the intake form sent.
Contact details
It took the contact details for the form, such as an email address, and confirmed the booking in the thread.
Swipe sideways to see the whole thread.
Contacts and bookings
I set it up so every DM created or updated one GoHighLevel contact, with no duplicates. When a booking was confirmed, four things followed.
- A calendar event was added.
- The pipeline opportunity in GoHighLevel moved to Booked.
- The intake form was sent.
- The team got an alert.
The design goal
Built so every text DM became a GoHighLevel contact within 60 seconds.
Returning people and the handoff to a person
It remembered people, and it knew when a person should take over.
A returning client was welcomed back by name. In testing, it mentioned their last Botox visit and asked what they would like to book.
Some messages were never answered by the AI: medical, refund, legal or “talk to a person” messages. For those, it took a callback time and a way to reach the person, then alerted the team.
Someone unsure about a date also got a callback, flagged for the team.
Swipe sideways to see the whole thread.
Follow-ups when someone went quiet
If someone went quiet, it followed up after 1 hour and again after 20 hours, then stopped at Meta’s 24-hour limit.
What it was built with
ManyChat for Instagram, 7 n8n workflows, an OpenAI model (gpt-4o-mini), GoHighLevel, Google Calendar and Gmail.
I built the first version on Instagram’s own webhooks (the verification handshake and inbound message events), with replies sent through ManyChat.
One of the 7 workflows was the brain. It read the message, checked the knowledge base, ran the AI agent, decided the next step and replied in under 2 seconds.
Swipe sideways to see the whole workflow.
How it was tested
219 automated checks across 7 test suites, all passing on 27 April 2026.
The rule: no workflow change went back into n8n until the checks passed.
Status
Client build, delivered. The engagement ended in July 2026.
More of my agent work
Want one for your clinic or agency?
Tell me how people book with you today and which DMs your team answers by hand. I’ll walk you through what an agent like this could handle for you.