Congrats @Nate Herk You have helped me and many others just by pointing and sharing what you had learned. You share in a way that is easy to understand and put into practice.

@Nate Herk Great video love the details, in it. Just wondering, how do you create (if needed) a second table within the same project? If you use the same "quick start" it throws an error about vector extension already present. Me being me, I trying to setup locally as well, but having issues with the postgres connection, as the ports are not exposed for external use.

@Nate Herk I did some playing around, and noticed I can duplicate the table in Supra. When you duplicate it asks for a new name. Like you I am not to familiar with Supra, but it looks like it keeps the structure, so the only thing it changes is the name. Once you make the duplicate, you can then modify the column headings as you need. I suspect that can be changed in how you upsert and what data goes where? Still playing thou.. :)

@Wayne Ergle I use the serp and it works well. Have noticed at times it does use, but it is based on what is typed in chat or webhook. In my prompt message I use #web search and under that I then call attention to the serpapi so it knows to directly use that tool if it needs to or if user requesting specific info. Sounds like you are on the right track. Nate or someone else might have a good system prompt that might help.

@Wayne Ergle yeh, I’m in the same boat. I have used activepieces, make as well. FlowiseAI is good as well, but I have noticed the need to use third party apps for actions that are already some what built in with other apps. Like N8N, Activepieces, etc.., but is great for internal processes and logic flows.

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..

@Bertrand B . Ram is not really used to much except for loading the model. VRAM in chip (threads) help with cpu only. I have LocalAi running on a amd 4 core and it runs ok and responses are jerky (response type is slow). 1st setup I use Ollama llama3.2, and gpt-4 when using LocalAi. But it runs! If you don’t mind waiting 30-45 secs, it should be ok, for smaller models. 2nd setup above My server is home lab, using proxmox hosting vm’s. I have dedicated VM for Ollama, using 8 cores with 10 Gb ram, and same for LocalAi. I task between both systems to balance load. To be honest, not sure if it is the best setup but it works for what I have it doing.