For my 2 cents. I self host N8N and host Ollama, LocalAi. I have an integrated GPU on the server which has 32 GB ram, and running AMD 16 core. Like @James Walls and @Nate Herk I use cloud as well. However, the self hosting (if possible) is good and bad to a point. Using vps with no GPU, can be slow, but for me most processes take about the same time as if going external. Depends on your needs thou. Mine is mainly text, and manipulative tasks, that don’t require gpu support. Images I used Dall-e or others. I don’t create locally, as yes, gpu would be need greatly. Again, as my instance is local I have no bandwidth issues either or delays, so I might be one sided here. :). But as Nate and James said, your needs, need to drive your thinking n how to manage. My cost with OpenAi, runs about $3- $4 a month, as I use LocalAi and Ollama locally, and heavy lifting stuff I send to Groq, or OpenAi. Hope this helps a bit..

Been struggling with using two types of triggers as well, for simplicity in deploying. I was for a short time able to get it to work with session ID. Grab session ID from both and place in expression. One trigger would be red in the key field as the input was not found and error, I used continue on error for both nodes. Not sure if there is a way to ignore the key value when not present, but sees another? wondering if a IF node will work in front and if key value is >= "true", goes one direction false goes other. That way you key value will always have the correct value.. Still experimenting with this thou.. Memory for this is difficult to me. As it stores both chats, even with session ID I see it still goes through the entire history looking for the matching ID..

@Dale Thomas I have tried adding memory, and use session ID , however, I have found it reverts back to the session ID and its memory, and get confused. If you find a way to make this work, that would be great.

@Nate Herk , thanks for that :). But what you experienced and show is what I am seeing as well. In looking. at the node this, I do not see where we call for the ID? I see the ID when pulled just like you showed. I like the challenge thou and learning the process Lol.. Love the help and guidance from you and everyone here. I will see if adding a more strict prompt in calendar will work. Pull in the id, if asking for delete or update, etc.. confirm event and show id? If yes, delete. Think we might need memory for that? Lol. But, yes, I have noticed as well that the responses are quick and on point, but there does seem to be a delay 2-3 minutes even in creating event before it shows up in Google, and on phone. Using default google calendar for account. Thanks again

@James Walls Thanks. Yes I was going to implement header auth in n8n for the form. in testing i get the information back in n8n when form is triggered, however, nothing shows on form. Been reading and searching, and yes, the php file created by gpt worked great, and the response where the response should go works with what was created by gpt, but n8n response data from n8n does not populate. Will keep trying things to see if the paths, etc.. are correct. Would not think I would need to send a second request, but might try that as to what you said earlier, it might be timing out or something.

Great I’ll look at the log in the browser. Was only looking at n8n side. Thanks again

I do have a question. I'm having the same issue with the suggested temp change at .2. Using gpt40 and have tried gpt4 mini as well. When and where does the email address get passed? Meaning, I noticed that the contact tool pulls all contacts, so I am suspecting the agent is looking specifically for the name (s) being requested. with that said, I noticed that the actual email tool (node) is getting the incorrect address and it is fake. In my case madelin@example.com, instead of the address in the contact list. I cannot find the actual address being passed to the tool in any of the logs. Just the message being sent and the response. @James Walls where would I put the additional instructions for the agent? In the model itself or in the assistant agent, or the mail agent?