The GLOW-UP Every AI Agent Needs

[@ColeMedin]

Think you have what it takes to build an amazing AI agent? I'm currently hosting an AI Agent Hackathon competition for you to prove it and win some cash prizes! Register now for the oTTomator AI Agent Hackathon with a $6,000 prize pool! It's absolutely free to participate and it's your chance to showcase your AI mastery to the world: studio.ottomator.ai/hackathon/register OpenRouter also became the third sponsor of the Hackathon recently! You can claim $10 in OpenRouter API credits for free as a part of participating in the Hackathon! Claim your credits to get access to ~300 LLMs now: studio.ottomator.ai/hackathon/open-router-credits

[@ThatNerdChris]

Dude YES! 30:05 is there ANY chance we could get support for the gemini voice api? Or even the chatgpt one? There are NO open source projects like this that can do this.

[@ColeMedin]

I do actually want to do this soon!

[@ThatNerdChris]

@@ColeMedin 👀🔥

[@jonathanadokwei2060]

Thank you so much Cole, always ❤. Can I suggest that you put the series into a playlist? It can be quite confusing to pickout the particular videos and in their order. I understand the 1st ever video is not a post-prod video but a stream, took me a while to figure that out.

[@ColeMedin]

You are so welcome! I do actually have these videos in a playlist on my channel! kzbin.info/aero/PLyrg3m7Ei-MrSXWv90oXXbuSsbdOP9j2n

[@jonathanadokwei2060]

@@ColeMedin You don't disappoint bruh... Thanks 👍🏾

[@ColeMedin]

Haha I appreciate it. You bet!

[@AntonioRonde]

Can supabase be replaced with a self-hosted Postgres?

[@ColeMedin]

Great question! So my interface relies on the realtime database updates that Supabase offers. So it wouldn't quite work with self-hosted Postgres. You can self-host Supabase for Postgres though which I would recommend doing anyway! Then it would work 100%

[@realsammyt]

Love your work, passion, and perspective. Curious about your video production process and workflow.

[@ColeMedin]

Thanks man! I record my videos with OBS and edit them with CapCut. Hoping to hire a video editor soon though. My wife helps with thumbnail creation and uses Procreate for the initial design and then I add final touches and text with Photoshop. I always create a script outline for my intros and outros and then a general outline for everything in between. Then I'll record everything in the video first where I'm not sharing my screen (generally just the intro clips and outro), and then finish with the rest.

[@miselgpt]

Pure gold! 🤘

[@miguelclare7999]

"Shoehorning"? sounds funny🤣

[@Qwerty-vm2zd]

Does live agent studio accept file uploads? Like my agent accepts a pdf file. How do I do that with live agent studio

[@ColeMedin]

It does now! Added this a few days ago. I updated the Developer Guide on the Live Agent Studio covering how to do this. I even have an example agent linked there!

[@oakgnarl5021]

Cole, Cole... you da man!

[@ColeMedin]

I appreciate it! :D

[@koyamikami2644]

Thank you Cole for amazing video! So you are saying n8n is not capable of serving as production ready backend for RAG?

[@ColeMedin]

You are so welcome! n8n is constantly improving so maybe at some point, but right now I haven't been impressed with the performance of my n8n agents when it comes to RAG. Don't get me wrong it certainly does well, but still not as good as my custom implementations in Python with frameworks like Pydantic AI/Langchain.

[@tuanmedeiros]

Could you create a video on how to containerize an AI Agent via Docker in an API with n8n?

[@ColeMedin]

I will be putting out more content using Docker soon! Though not necessarily with n8n, since typically the n8n instance itself is already in a Docker container when you self host it unless you are installing n8n directly with npm. Maybe you could clarify more?

[@Qwerty-vm2zd]

Do you want the n8n running in a docker instance and host an API?

[@gavienperez9329]

can i still get in on this hackathon event without knowing how to code? Like maybe make an agent with n8n or soemthing?

[@ColeMedin]

You sure can! n8n is actually one of the ways I'm recommending creating an agent for the Hackathon! I've got a ton of resources on the Live Agent Studio for building agents with n8n!

[@vladimirrumyantsev7445]

Thank you so much for such nice content, but i didn't get it, why we need our own agents if chatgpt or perplexity already exist? What's the real use case? Thank you very much and waiting further videos

[@ColeMedin]

You are so welcome! And totally fair question. AI Agents are really just LLMs that are given the ability to interact with the outside world to do things on your behalf like interact with your calendar, work with files in Google Drive, scrape websites for you, etc. Some of these "tools" like web search are widely available for things like Chat GPT and Perplexity. But other tools that you want your agent to interact with, like Slack, Google Drive, Asana, etc. are not. That's why you would want to build these custom agents instead of just using existing tools like ChatGPT - you can make these agents interact with literally ANY tool/service/site you can possibly think of!

[@vladimirrumyantsev7445]

@ I think it's like agent you're designed from scratch with structured and more robust output 👍

[@teebo6680]

why do i not see any agents when i visit the ottomator studio? "No Agents Found"

[@ColeMedin]

Hmmm.... I've had this happen like once before and I fixed it by logging out and logging back in. Have you been logged into the platform for a long time by chance?

[@pressplayg]

Hey bro, I recall you previously mentioned that Flowise isn't recommended for production due to limitations in logging, monitoring, and error handling, making it better suited for rapid prototyping. However, Flowise offers integration with several analytics tools - like LangSmith and others. Which solves the issue you raised right ? . I'm torn between your expert opinion and Flowise's capabilities. Can you help me out, bro?

[@ColeMedin]

I didn't actually know that Flowise integrates with LangSmith - that certainly helps with that big limitation which is one of the things that formed my initial opinion of Flowise not being production ready. LangSmith helps with logging and monitoring but not error handling though, and it's still unfortunate to be tied to a specific tool (LangSmith) and not having the ability to implement custom solutions easily. However these downsides don't mean Flowise isn't worth using. I definitely think it is and it's a phenomenal platform!

[@0xh8h]

Recently, I learned about MCP (model context protocol) from Claude and it makes me a bit confused about what is the most up-to-date way to implement tool call in AI agents. Because even before MCP was introduced, AI agents already be able to call 3rd party API or do many interesting things outside of the LLM. So from now on, when developing a new AI agent, should we use MCP or not? Thanks a lot

[@ColeMedin]

From what I can tell, MCP is meant to be more of a general standard for AI agents in the future. But right now it's just for using tools with Claude in Claude Desktop. What I'm developing is more general - you can create an AI agent with any platform/framework/language and then have it hosted to share it with the world.

[@0xh8h]

​@@ColeMedin from what I learned on their website (related to MCP clients), we can create custom clients so that we don't have to rely on the Claude Desktop app. In the future, I imagine that all AI agents will implement as MCP clients (at least a part of the code), so that we can "reuse" the functionalities of available open MCP servers instead of wiring all the logic ourselves.

[@0xh8h]

​ @ColeMedin from what I learned on their website (related to MCP clients), we can create custom clients so that we don't have to rely on the Claude Desktop app. In the future, I imagine that all AI agents will implement as MCP clients (at least a part of the code), so that we can "re-use" the functionalities of available open MCP servers instead of wiring all the logic ourselves.

[@ColeMedin]

Okay that makes sense! Yeah MCP is really a first of its kind standardization for agent communication. You're definitely right that in the future a lot of agents will probably have an MCP compatible entrypoint.

[@radowanahmedbaized3597]

Hi Cole. Is there a way to play audio responses or show AI generated images on the Live Agent Studio? Also, could you give some general ideas on what type of problems the Live Agent Studio interface can solve for its users? Asking to help come up with ideas for the hackathon 😅

[@ColeMedin]

Great questions! Playing audio responses isn't a native feature of the Live Agent Studio, but if your agent generates a .mp3 URL and gives that to the frontend you could create a custom component for the user to be able to listen to that mp3! Same thing for AI generated images. You could have your agent upload an AI generated image, give that URL, and then in the frontend create a custom component to display the image. Check out the "Handling Extra Data" section in the developer guide: studio.ottomator.ai/guide The Live Agent Studio can really connect you into any agent you can build! Obviously some more complex than others!

[@radowanahmedbaized3597]

@@ColeMedin Thanks a lot Cole!

[@ColeMedin]

You are so welcome!

[@amrpyt]

How make this agents as a SaaS platform?

[@ColeMedin]

Good question! First step is getting it deployed to production which I'll be doing through the series. You'll also want to build a frontend where users can pay with Stripe to get credits to use your agent and probably connect their own GitHub account too!

[@edism]

Working on that right now.

[@amrpyt]

@@edism some details plz?

[@prabhic]

Thank you once agian👏

[@ColeMedin]

You bet!

[@mesiiu407]

Master🎉🎉 Try with notebook and project IDX with GKE?

[@SuperNinjaman1234]

Hey Cole, I recently asked you the question about local setups and getting a MacBook with how much RAM, I wanna know your thoughts on the NVDIA DIGITS, you think it’s worth it?

[@ColeMedin]

Great question! It's hard to tell right now since we don't know the price of Nvidia Digits yet (at least I can't find it), but I'm guessing it's going to be VERY worth it. According to their site and based on the specs, you'll be able to run up to 200B parameters locally, which is insane.

[@SuperNinjaman1234]

@ they’re saying $3,000 which is way better than the same amount of RAM on the M4 at $5,000

[@ColeMedin]

Yeah it's an insanely good value! Did some more research myself recently.

[@The_AI_Buddy]

I saw your comment saying flowise is not production ready but are you sure the new verison of flowise is super updated . And also i got an idea for you that will be gane changing, why dont you develope a no code open source platform like flowise but for pydantic AI😍 i beleive only you can do that ❤u have that passion bro. Isnt this crazy good idea ?

[@ColeMedin]

You know I haven't checked out the latest updates to Flowise yet so it sure could be now! The biggest issues before were that logging/monitoring/error handling wasn't the best in Flowise - do you know if those things have been addressed in recent updates? Making "Flowise for Pydantic AI" is a FANTASTIC idea! Something I have actually put some thought into ;) I appreciate the suggestion and the kind words man!

[@EduardsRuzga]

Ou wow using Prezi! How do you like it? :D

[@ColeMedin]

Indeed! Always think of you when I'm using it haha :) It's fantastic!

[@build.aiagents]

Phenomenal

[@ColeMedin]

Thank you! :D

[@The-ism-of-isms]

Yo boss can you make n example a customer support rag chatbot with pydantic AI framework in this series or like tell us how to customize github pydantic ai agent code you created to a customer support rag agent what all codes to change in it atleast.

[@ColeMedin]

I'll be continuing to create more use cases with Pydantic AI on my channel, including agents with RAG! What would you want your customer support chatbot to do exactly?

[@The-ism-of-isms]

@@ColeMedin Thank you , Actualy a prodcution ready superior customer support chatbot that can recommend product , make neegning converstation with cleint , convert intrets to sales , collect lead etc. more like virtual company represneative that knows eveyrthing and working 24/7. how is this idea ?

[@ColeMedin]

Yeah fantastic idea! Most businesses need something like this!

[@DemocracyDecoded]

Let me ask you something, you've been building agents this entire times, what projects have those agents actually finished for you? What have you automated in your life? Thanks.

[@ColeMedin]

Great question! Generally I find AI agents are best when used to save my time, not entirely replace a task for me. So things like drafting email replies for me to review, helping me write scripts for videos, and transcribing my meetings and building up a knowledgebase so I can ask questions to remember things from my meetings.

[@GoodBaleadaMusic]

Hi free idea. Make an agent that goes into social media and cleans it up. Organizes it. Goes back through 20 years of facebook and brings old albums to life. Delete things. Omg delete things. Professionalize. Turn my my shit into something coherent. I'd pay 20 bucks a month for that.

[@ColeMedin]

Love it! I think a ton of people would pay for that haha

[@GoodBaleadaMusic]

@@ColeMedin change the tone of my social media to reflect my new position at XXX. And then it just goes through everything and then turns you into that person

[@Soulscribez]

I don’t understand why it is called a agent it is only a chatbot!!!

[@ColeMedin]

It's much more than a chatbot because under the hood is an agent that is able to ingest and understand GitHub repositories on your behalf! A "chatbot" is typically just an LLM - it doesn't have access to other things besides maybe just being able to search the web.

[@pressplayg]

I hope in this series you will include how to turn this agents as a saas platform with subscription and payment features ❤. That will make this series officially a zero to hero super course ❤thank u boss

[@ColeMedin]

I wasn't planning on it initially but I've gotten a couple requests including yours, so now I'm looking into how to fit that in! :)

[@The-ism-of-isms]

​@@ColeMedin perfect

[@pressplayg]

@@ColeMedin Thank you bro , plz do it

[@ByronWatts]

A normal URL never includes a 'backslash'. Ever.

[@ColeMedin]

You are right - that is my bad. Always get my back slashes and forward slashes mixed up for some reason. I appreciate you correcting me a lot!

[@SteveDigital-qx6qs]

kzbin.info/www/bejne/mXmndHaJp5aMiKs

[@ColeMedin]

Yeah I had a couple other people share Agent Zero with me, unfortunately I do remember seeing it now but totally forgot about it when I created Agent 0. Not trying to rip on anything! I actually named it Agent 0 just because of the URL for it on the Live Agent Studio: /agent/0 It's kind of like the "0th" agent on the platform!

[@SteveDigital-qx6qs]

@@ColeMedin yeah I just wanted to give you a heads up, keep up the good work

[@ColeMedin]

I appreciate it!

[@amrpyt]

First ❤️

[@trevorpowell4851]

Your naming is terrible! I thought you were going to hook it onto the actual agent zero 😕 you need to really start coming up with your own unique names instead of ripping off other names

[@ColeMedin]

Shootttt.... I remember hearing about Agent Zero a LONG time ago now that I think about it, but I totally forgot it existed haha I really just named it Agent 0 because of the URL for the agent: /agent/0 Since it's the "0th" agent on the platform. Not trying to rip off of anything else!

[@trevorpowell4851]

@ColeMedin I am glad to hear you didn't do it intentionally but I thought oh he is going to hook it into agent zero framework. Wich may be a interesting video anyway haha

[@ColeMedin]

Yeah that would be pretty sweet!

[@Timers124]

1st dislike ( sorry i had to kinda)

[@ColeMedin]

How come? I'd love to improve and I'm curious why :)

[@Timers124]

@@ColeMedin idk i like the vid but i saw no dislikes so i just wanna

[@PhuPhillipTrinh]

@@Timers124I dislike your troll he’s doing great things for the community

[@sum1likeno1]

Agent 0? That name is already taken by @agentzero 😅

[@ColeMedin]

Shootttt.... I remember hearing about Agent Zero a LONG time ago now that I think about it, but I totally forgot it existed haha I really just named it Agent 0 because of the URL for the agent: /agent/0 Since it's the "0th" agent on the platform. Not trying to take any names!

[@TheDJHollywood]

@@ColeMedin Yes I am involved with the Agent Zero. I can not have 2 different Agent Zero's. I am sure that will get really confusing.

[@ColeMedin]

Yeah fair! I'll have to think about how to change the naming of it. For now hopefully it helps I call mine Agent 0 and not Agent Zero, though I totally get that isn't enough in the long run