Hey ​@davidle

So yes, there are several options to consider:

AI Step: this is the easiest and fastest to setup, but you’re giving up a bit of control over what happens inside.  In your case that you want to store the user inputs, the AI Step won’t let you do that.

Direct ChatGPT/OpenAI Completion: this alternative gives you more control, and you can do it through Manychat’s native integration or through External Requests directly through the API.

In this case you have more control, but you’re limited to the prompts you send. And adding memory to it is possible, but it’s a ton of work and unreliable in my opinion (memory would be that the AI remembers the previous interactions and can answer the question ‘what was my first question?’ correctly).

Assistants API: this gives you the most flexibility, allows you to train the Assistants with your business’ information, and comes with built-in memory. It’s a more complicated setup with several steps - but it’s doable, and if you don’t want to go through all the hassle of reading the docs and figuring it out, I have a template that I sell that solves all this for you. ;)

The template connects Manychat directly with OpenAI without any additional software, and this is my go-to solution for most of my projects at this moment.

Connectors: you can also set up these different options with connectors, often using Make or Zapier as middleman and sometimes additional tools to handle the back and forth of messaging.

I personally try to avoid additional software and complexity whenever I can, and since my Assistants API template works very well, I’m not using these solutions. But yeah, they’re also viable options if you want to explore them.

Hope this helps ;)

PS: I’ve replied to your DM with more details on the template.