I saw many people struggling to make function calling work using the Message Assistant module, so I wrote a guide. Let me know if anyone has any question
Thank you so muchâŚthis was a HUGE help!!
Nice work. Iâve also added some of the technical details of how this works available in this thread here.
@hideyukishibata your example is great! the response of the first scenario with the openai module âGenerate summaryâ works without issues while executing on make platform,
But the result response is not arriving to the assistant in openai assitant chatbox window on the openai playground (as i share with images to yo following this explanation). once the user sent links, says running in proccess⌠and just keep running as you can see on images belowâŚ
Please Can you suggest what could be the issue or what is missing? Thanks in advance!!!
PD: I share images below for better comprehension of the issue.
@Carlos_Maria_Maidana Thanks! On Playground, I donât think the assistant automatically runs the workflow (It doesnât know your webhook URL so it cannot work). What I think you should be doing here on Playground is manually enter the response the assistant is supposed to get so that the assistant is able to generate an actual message using the response / context / information (whatever you want to call it). So you can only see and try to adjust the final message there. Does this make sense?
Hi @hideyukishibata thanks for the time you take to response my question! i really appreciate it!
Now i understand that if i call an openai assistant function on the openai playground, it is required to enter manually some input like âsuccess or okeyâ in order the complete the process and receive the response on the user prompt,
I am using make (formerly integromat) to execute the functions via webhook, do you think it is possible to instruct the assistant on one of the modules of make, to automatically run all the function and then send the final response to the prompt (without requiring to enter manually some input like âsuccess or okeyâ) ?
Hope my explanation be understood (my english isnt enough good)
PD: I mention Make is only as an example, may i use other like Pabbly or Latenode or whatever that can do what i wanna do.
Thanks in advance! king regards!! Your Great!
@Carlos_Maria_Maidana Hey, you seem to be talking about doing function calling using two scenarios just like I wrote in my Medium post. If thatâs the case, yes it works without writing the success message. Itâs just something you have to manually do on Playground (and nowhere else do you have to do it). Iâm sorry I might not be understanding your question perfectly, but hope this is the answer youâre after! And in regards to other automation tools, I donât know if function calling can be done using them. Make takes care of the calling part on their end which you would normally have to code. Not sure other platforms support that. Probably not, I think.
Hi my friend! (Hey, you seem to be talking about doing function calling using two scenarios just like I wrote in my Medium post. If thatâs the case, yes it works without writing the success message.) Yes, that is exactly what i was looking to clarify!
Really thanks for your time and knowledge!!!
Best Regars!!!
Hey, my friend. Is there any alternative to use function tool with only 1 scenario? I mean, is it possible to get URL from anywhere else?
@Gabriel_Mercier Hey hey. Hmm, may I ask what you are trying to achieve?
Iâm building an SDR AI agent. I already have two active scenarios for it.
The first scenario is simply to retrieve lead information from a CRM and send a WhatsApp message.
The second scenario is designed to continue the conversation with the lead and qualify them. If the SDR assistant qualifies the lead as approved, the automation then creates an event on Google Calendar. Therefore, I need to use a function to store the qualification result and the date chosen by the lead.
I had to create these two scenarios because I need to use an HTTP request in the first scenario and a custom webhook in the second one. Since Iâm using the free plan, which allows only two scenarios, this agent is just for demonstrating the process to a potential client. Thatâs why I asked if itâs possible to obtain the URL without creating an additional scenario.
@Gabriel_Mercier Ah I see. Iâve never tried this but this might work. In your second scenario where you have a webhook as the trigger, try adding a router right after the trigger followed by a filter on each branch. Let the workflow proceed only to one of the branches by adding a query string to your webhook URL, something like yourWebhookUrl?branch=addToCalendar
and yourWebhookUrl?branch=functionCalling
. This way you can use the same webhook URL for two different purposes. Let one branch work as it is in your current set up and let the other branch handle the function calling. So basically, what weâre basically doing is self-invoke the exact same scenario during the run. I have just tried posting to a webhook via HTTP right from the same scenario that the webhook is set as the trigger (an infinite loop) and it seems to work. So itâs worth trying!