I’ve watched a couple of your videos now and went to your channel to see what else you had. I was shocked you only have 10 or so videos! Your content is really good and polished. I think you’re gonna be very successful. Best of luck.
Great video. The way you broke everything down was brilliant. I just sent you an email not too related to business but more as a student. Would love to learn from you!
I'm not disagreeing with you but the process as designed would inject too much latency to be practical. Much cheaper to have a flow that grabs an uploaded voice note being processed though whisper locally or Groq.
Thanks! Kinda confused on how the "Send Email" node works. I assume there's some logic in the main OAI prompt, and you hardcoded the email address into it? Also would have liked to see how you're using the vector database.
This is really great. I didn't understand the value of summarise the database. I mean it is only referring to same database everytime because we have hardcoded in the setting. What is use of this summarisation?
One of the functions available to us was to update the database. So as new projects are added or current projects are updated, the summary would reflect these changes. The idea of this build was more to show everyone how to hook up different tools to and agent and open your eyes to the capabilities.
Hi Nate, great breakdown of the nodes and process. Regarding the database where you're using the Google Sheet would referencing the Pinecone DB be an appropriate use of Pinecone for the data you are keeping track of?
Hey there! Pinecone could absolutely work yes, as vector databases are more efficient at searching large amounts of data and retrieving what we need. It really just depends on the use case and obviously in this example the database was super simple, so google sheets sufficed.
During all of my testing with this build, I racked up about $0.10 of usage, all using GPT-4o. You could also significantly decrease your costs with 4o mini. I hope that helps!
For this video I kept the prompt very simple. All I did was tell the agent the current date using the JavaScript function {{$now}}, and I outlined each tool the agent had access to. A cool tip, when you need a quick prompt you can take a screenshot of your workflow and have chat write one. It won’t be perfect but you want to make sure Chat knows to include the following sections of your prompt at a bare minimum: Context, Role, Instructions on the Tools, and Examples.
If I want to create a SaaS-based module for this workflow, I have few questions. Can we create SaaS-based solutions utilizing n8n? For example, the user should sign up, grant access to Gmail, etc. I realize that we need to integrate n8n into our backend to automate processes and integrate with other services.
But it still can’t solve complex things right? It’s like Siri …. What about things like can you please cancel all my appointments for Tommorow via Mail etc.?
This workflow is currently not accessible because I’m working on expanding it and altering some of the tools to be more powerful. I am hoping to create a community in the next few weeks where I can share resources with you all
Chaining this Agent to then call upon a plethora of smaller, focuse lifestyle type agents would be and will be the way until we get all knowing. An example, a calendar agent strictly focused on all things caledaring, then so on with more. Cooking, Cleaning, Shopping, Researching, all based around you're databases and vectors that you've started here with as a foundation. Suggest moving to ollama or anything you like but local. No need in having your stuff going out unless you want to cotinue paying.. Great work on a PARENT/CHILD type model. That's what life is right ;)
Hey there! Feel free to fill out a form on my website or shoot me an email nateherk@uppitai.com. I am currently at capacity with projects but I will get back to you as soon as I can!
![Building AI Agents: Chat Trigger, Memory, and System/User Messages Explained [Part 1]](http://i.ytimg.com/vi/yzvLfHb0nqE/mqdefault.jpg)
![I Built a Virtual Assistant AI Agent that can do ANYTHING for Me [Step-by-step Walkthrough]](/img/1.gif)
I’ve watched a couple of your videos now and went to your channel to see what else you had. I was shocked you only have 10 or so videos! Your content is really good and polished. I think you’re gonna be very successful. Best of luck.
I really appreciate this! Thank you man!
This is awesome, Nate. Just to let you know, gaining tremendous valuable learnings from your work.
Thank you! I’m glad to hear that!
Can you share this workflow please? @@nateherk
yess!!
hello nate amazing videos
One thing i would suggest is working on your prompts
They have very basic functionality and sometimes dont work
appreciate your tutorials!
Awesome videos Nate. I am learning a lot and using it in my own projects. Thanks!
Thank you! This is awesome to hear
Great video. The way you broke everything down was brilliant. I just sent you an email not too related to business but more as a student. Would love to learn from you!
Incredible, Nate! Super excited to work with you.
My man!
Thanks, I have been looking for a tool to build agents with and this seems to meet my needs.
Great way to leverage workflows as tools man nice work!
Thank you sir!
With real time OPenAI would be awesome. Or even whisper sending voice message and return them in both message and audio from AI. Thnks Nate
I'm not disagreeing with you but the process as designed would inject too much latency to be practical. Much cheaper to have a flow that grabs an uploaded voice note being processed though whisper locally or Groq.
@@dad2979 would appreciate a tutorial for this workflow if you have a channel
@@dad2979 no worries, always happy to learn more.
thank you
Thanks! Kinda confused on how the "Send Email" node works. I assume there's some logic in the main OAI prompt, and you hardcoded the email address into it? Also would have liked to see how you're using the vector database.
Check out a previous video I made about building an Email AI Agent! I go into more details about pushing contact data into pinecone and the prompting!
@@nateherk awesome thanks!
Thanks!
No problem!
Hey im curious as to what is the difference between using the already existing gmail node and creating the tool yourself?
At the time of recording this video, the existing Gmail tool node didn’t exist
Excellent class my friend. How do I download this Flow?
Please check in my free Skool community!
This is really great. I didn't understand the value of summarise the database. I mean it is only referring to same database everytime because we have hardcoded in the setting. What is use of this summarisation?
One of the functions available to us was to update the database. So as new projects are added or current projects are updated, the summary would reflect these changes. The idea of this build was more to show everyone how to hook up different tools to and agent and open your eyes to the capabilities.
Could this be used with GPT's voice API ? It can output text and voice. You could speak what you want and it could output text for agents to use.
Of course!
Hi Nate, great breakdown of the nodes and process. Regarding the database where you're using the Google Sheet would referencing the Pinecone DB be an appropriate use of Pinecone for the data you are keeping track of?
Hey there! Pinecone could absolutely work yes, as vector databases are more efficient at searching large amounts of data and retrieving what we need. It really just depends on the use case and obviously in this example the database was super simple, so google sheets sufficed.
Any idea how much these API calls are costing you? What of using Ollama for some of the lighter lifting?
During all of my testing with this build, I racked up about $0.10 of usage, all using GPT-4o. You could also significantly decrease your costs with 4o mini. I hope that helps!
Thank you
Whats the prompt you use in the main agent?
For this video I kept the prompt very simple. All I did was tell the agent the current date using the JavaScript function {{$now}}, and I outlined each tool the agent had access to. A cool tip, when you need a quick prompt you can take a screenshot of your workflow and have chat write one. It won’t be perfect but you want to make sure Chat knows to include the following sections of your prompt at a bare minimum: Context, Role, Instructions on the Tools, and Examples.
@@nateherk Thanks nate, will definitely try it.
Yeah, thanks, this is very helpful/
If I want to create a SaaS-based module for this workflow, I have few questions. Can we create SaaS-based solutions utilizing n8n? For example, the user should sign up, grant access to Gmail, etc. I realize that we need to integrate n8n into our backend to automate processes and integrate with other services.
Hey! Feel free to join my free Skool community and shoot me a DM, or email me nateherk@uppitai.com
But it still can’t solve complex things right? It’s like Siri …. What about things like can you please cancel all my appointments for Tommorow via Mail etc.?
Hi, do you have the workflow json i can paste. or is it in the community search in n8n?
This workflow is currently not accessible because I’m working on expanding it and altering some of the tools to be more powerful. I am hoping to create a community in the next few weeks where I can share resources with you all
If you shoot me an email, I'd be happy to share the workflows with you! nateherk@uppitai.com
Greate video, can you please share your n8n flows
Working on a community where I can upload all of my resources associated with each video. Stay tuned for this!
@@nateherk would be great if you could share this initial one as starting point
If you shoot me an email, I'd be happy to share the workflows with you! nateherk@uppitai.com
If you shoot me an email, I'd be happy to share the workflows with you! nateherk@uppitai.com
You should show what actually happens and the use of it. Demo
I’m confused. I do demo the agent on each step
Does it transfer all your data somewhere?
Not sure I understand what you’re asking here
Ist depends if you use a managed instance or if you self host it (which is possible). I run it on a local machine in my room.
Same for the LLM.
does this work with ollama?
Of course!
@@nateherk do you need a bigger model like gpt40 in order to complete tasks like this or can llama3.2 handle it?
@@tokenization I generally use 4o as that’s what I’ve found to work best. I can’t speak too much on the ollama models, haven’t used them extensively.
Chaining this Agent to then call upon a plethora of smaller, focuse lifestyle type agents would be and will be the way until we get all knowing. An example, a calendar agent strictly focused on all things caledaring, then so on with more. Cooking, Cleaning, Shopping, Researching, all based around you're databases and vectors that you've started here with as a foundation. Suggest moving to ollama or anything you like but local. No need in having your stuff going out unless you want to cotinue paying.. Great work on a PARENT/CHILD type model. That's what life is right ;)
Great thoughts, thank you for the comment!
Can somebody help with a workflow automation ?
Hey there! Feel free to fill out a form on my website or shoot me an email nateherk@uppitai.com. I am currently at capacity with projects but I will get back to you as soon as I can!
Where can I steal this from
Working on setting up a community to be able to upload my resources for you guys