✨ Announcing Chatbots: Create, train and share

Announcements
22

@Michael_ES,

I believe these chatbots have been sunsetted. Instead, they have incorporated assistants directly into the ChatGPT piece. I think if you check the piece, you will see it has the addition of ‘Assistants’. So you are free to make them on OpenAI directly and link to them in your flows.

Here is a short gif showing what I mean.

gpt-assistant

It should also be noted you can also use something like Re:Tune that is supported as a piece for creating custom chatbots. (There may very well be others like Typebot, etc. I haven’t checked in detail).

Kind regards,
GunnerJnr

23

Hi Gunner, thanks for the response. I had gotten that they were sunsetting the Chat feature in AP, but I was hoping that it was going to be replaced with access to Custom GPT’s from within GPT4 plus subscription. I have many bots that would be useful to use for automations… Not sure if that will work from OpenAI’s Assistant feature. Doesnt seem to work quite as well.

I will have to look into the possibility of transferring a custom GPT to OAI Assistants I suppose.

24

@Michael_ES, actually, if you use webhooks, you can set ActivePieces up to work in the schema. I have been playing about with it myself, and so far, I managed to set up an action that connects to a flow which sends an email (to myself), but the plan is to expand it to have multiple actions like a tool, so I can have an action for sending an email, and action for writing a blog, and action for whatever I feel like. Here is a gif giving a demo of it working to send me an email:

gpt-assistant-ap-email-demo

Note: If the gif is too small, simply right click it and click open image in a new tab, you can view it full size.

If you get stuck with the schema when you go to create an action in the assistant, there are a few examples, like the one for fetching weather data. This is a good starting point for you to see how they achieve certain things and can be easily modified to work with ActivePieces.

I hope this helps,

Kind regards,
GunnerJnr

25

That is actually phenomenally awesome!

I was looking to set it up in reverse though. Where I could build a GPT like yours, and send it a request and get the response back.

For example, I have an automation that creates articles in a very specific way to my use case, and the first part of that process is manual because I go to my custom GPT to get longtail keyword ideas from an initial keyword phrase.

Ideally I would just add that keyword to the Google Sheet, and that triggers the automations that starts the research process for longer keywords… If that makes sense.

Just an example, I do not think I would end up actually using it that way, but it illustrates the direction I was hoping we could work with GPT’s.

26

@Michael_ES, I think I see what you are saying, and I am sure it is highly possible. Although I have not tested it personally (yet), I have been more focused on just triggering the flow for now, but because I am always so busy, my progress is a bit slow at times, but receiving data back to the assistant is definitely on my list of todos.

Regarding what you said here:

Ideally I would just add that keyword to the Google Sheet, and that triggers the automation that starts the research process for longer keywords

I have to suggest, similar to the email flow above, wouldn’t it make more sense for you to put the keyword into the custom GPT and then have the custom GPT send the longer keyword research/results to the sheet, which could then trigger the flow to write the blogs?

Or even cut the middle man out altogether and have a sheet that you put your initial keyword into and then have an OpenAI piece perform the keyword research for you? Then, from those results, you could either have a second OpenAI piece directly after it to write a blog from the results or even first output the results into the sheet and then have another OpenAI piece to write the blogs from those.

I also gather if the custom GPT is one you made in the assistant section of ChatGPT, you could simply copy the instructions/prompt directly from it and put them into an OpenAI piece in your flow on AP.

It is also worth noting that in my first reply (the one with the OpenAI playground assistants), you should be aware that these are the exact same assistants that you create in the ChatGPT interface as a custom GPT, so you could technically recreate your custom GPT assistant in the OpenAI playground assistants and then link it up through an OpenAI piece as I demonstrated above in the initial reply, so this would definitely aid in cutting out the manual step altogether, and in my opinion make a much more efficient work flow as all you would literally have to do is add an initial keyword to a sheet and then the flow would not only do the research for the longer tail keywords, but also create the blogs too.

I hope this is helpful to you in some way, but if not, perhaps it is some food for thought.

Kind regards,
GunnerJnr

27

@GunnerJnr I am right there with you. I have a pretty in depth set of pieces setup once a longtail keyword phrase is added to a google sheet. It kicks off a slew of actions to get precisely what I want out of it spanned across 2 pieces (the first one triggers the second one to kick into action once it is completed).

The GPT I created is pretty in depth, and I am unsure if I can copy and paste over to OpenAI Assistant, but you have intrigued me into seeing if it is possible… Maybe I could build a GPT that could make any adjustments from the information used for a GPT and recreate it specifically for the format needed for OpenAI Assistant to use…

I like to approve the keywords before a blog is created typically, as I have to make sure we are not cannabalizing other keyword phrases and also that the content makes sense for the site it is being posted on, but I would also like to work on that as an automation as well. Most of the keyword phrases remain in the sheet, even through testing, because of the limitation of how Google Sheets “new line” works in ActivePieces.

There doesnt seem to be much in the details for the GPT’s in the ChatGPT interface, and it makes me wonder if there is more to it in the backend that makes it work as good as it seems to, or if it really is that simple and rudimentary…

I appreciate the help, you have given me a lot to consider. The biggest obstacle now though for me is trying to figure out how to get the memory part to work in ActivePieces, it throws errors most of the time. But my peices were built before that was a possibility, so they work without it, but would be able to eliminate some steps if it worked more consistently.