THIS New AI Agent Can Do EVERYTHING! (Magentic One Test & Install Guide)

Рет қаралды 3,759

Күн бұрын

Пікірлер: 36

@OminousIndustries Ай бұрын

I took a go at getting the autogen-magentic-one repo to work with llama 3.2 vision (11B) and had some decent results: kzbin.info/www/bejne/Y4jUeYxpqqqSgZI

@locommotionmusic Ай бұрын

enjoy the way you leaned into the YT ai meta with the over-the-top video title :D very interesting magentic demo too!

@OminousIndustries Ай бұрын

LOL yeah, I had to do it for the algorithm.... Thanks!

@MrKasper0001 26 күн бұрын

I love the content! , I hate the keyboard! Do a video on having magentic download, setup, and interact with Moshi AI? That way you can interact with your agents by simply speaking. I bet it would go viral!

@OminousIndustries 26 күн бұрын

Thank you! I do not have any experience with moshi and will look into it, your idea about vocally controlling the agents is very interesting and I agree it would be a cool thing to set up. I could def get that working using microsoft azure tts/stt since I am pretty familiar with it.

@MrKasper0001 26 күн бұрын

@OminousIndustries it's been a nightmare trying to set it up lol. I don't actually know how to code or know much of anything, that I jumped into the AI. I learning curve at 90 mph. Moshi - Even carefully using conda to isolate dependencies. I'm still having trouble with pip finding compatible s***. I'm half tempted just to go with a telegram bot for text-to-speech and speech to text then connect that to the open interpreter so I can continue trying to dissect magentic one.

@OminousIndustries 24 күн бұрын

@@MrKasper0001 Though frustrating, jumping right in can be a great way to learn even if it is frustrating hahaha. My best suggestion for that is to have an llm help you dissect and understand the codebase and how to make the specific changes you want to make. It takes time but is a valuable experience.

@NickYoung16 Ай бұрын

A search in KZbin for magentic one auto corrects to magnetic one, only one on topic video loads. Dunno if adding a magnetic tag would help, might do until KZbin catches up to it as a topic at least

@OminousIndustries Ай бұрын

Good point, thanks for that. I will add one into the description!

@userou-ig1ze Ай бұрын

It's pretty annoying, I'm unsure if it's not renamed to magnetic one at this point

@wareagleish Ай бұрын

Great video. Please show tell us how to connect to Ollama!!

@OminousIndustries Ай бұрын

Thank You! I plan to do a follow-up video on Ollama integration soon!

@OminousIndustries Ай бұрын

Here is the ollama update: kzbin.info/www/bejne/Y4jUeYxpqqqSgZI

@augmentos Ай бұрын

Nice video keep these up. My only input is at the start maybe show or explain what you're gonna do and if you can how what you're about to do compares to other well known alternatives (in this example other agentic frameworks) to give us the value proposition of why to watch all the way through. But I did and it was fantastic! Tnx sub'd

@OminousIndustries Ай бұрын

Thanks for the feedback and the sub! I am still working on refining this video style to make it more consistent and I will take your suggestions into account while doing that :)

@Rejekts Ай бұрын

Cool! I just tried it with gpt-4o-mini since it's cheaper though it doesn't do the OCR, sadly

@OminousIndustries Ай бұрын

Good thought to try it with that. The prices will go down as time progresses (hopefully)

@QuantumXdeveloper Ай бұрын

GPT-4o-mini is not multimodal in nature. It doesn't support image input

@Rejekts Ай бұрын

@@QuantumXdeveloper The openai api does support sending it images though. You can use it in both the playground and api

@ewasteredux Ай бұрын

Thanks for this. I didn't know this existed so...

@OminousIndustries Ай бұрын

Of course! I had learned about it recently myself.

@ElshadFeyzili 29 күн бұрын

Thanks for the video! Is there a way to automate the Web Surfer Agent of Magnetic One? For example, when we provide username and password credentials and instruct the agent to log in and navigate to a specific page to perform an action, it often fails due to authentication protocols. How can such authentication issues be resolved? Any insights or suggestions would be greatly appreciated!

@OminousIndustries 28 күн бұрын

Thank You! For this case, I can't say for sure, but I would likely begin with playwrights "storageState" to save and reuse browser data, I believe you will have to map the credentials you save to a location in the docker instance so that they can persist across different sessions. You might be able to login to the service you need from your main browser and then transfer the cookie over to the browser used in magentic one. It may also be helpful to enable the human in the loop flag so you can supervise some of the actions.

@currywhite-p3n Ай бұрын

Thank you very much for your video. In the orchestrator module, an error occurred while executing the code "ledger dict: Dict [str, Any]=json. loads (ledger str)". The error message is as follows: "json. decoder. JSONDecodeError: Expecting value: line 1 column 1 (char 0)". May I ask what is the reason for this?

@OminousIndustries Ай бұрын

It sounds like empty or not a valid json was returned by the model. The orchestrator shows the ledger results in json formatting iirc so if the model does not give a response with the expected syntax this can happen. While not with GPT4o, when using this repo with a different model I noticed that sometimes I would get json formatting errors but it would keep going as the model would sometimes return correct json and other times not. I did not have this issue with gpt4o, but I suppose it is possible. If you are having this issue every time you run, I would maybe take a look in the orchestrator script and see if you can print the returned result to see if the error lies in the formatting of the response being incorrect or something else.

@ItsLightMind Ай бұрын

Hey, sorry for bothering was wondering if you'd ever share your thoughts on the MSI 40-inch monitor. Would love to hear long term thoughts on it, would really value your opinion.

@ItsLightMind Ай бұрын

By the way to give more context, i'm wanting to hear your opinions on it on productivity and design. I dont care if you dont game.

@OminousIndustries Ай бұрын

Of course. Truth be told I am lukewarm on it. I had seen some other folks speak of the text not appearing as clear on some other monitors and I have noticed that myself as well. It is perhaps a setting or something but I keep it in "eco" mode as it is easiest on the eyes and haven't bothered to change any other settings as it works fine even with the weird text. It seems to not like powering back on after being off for a while, the one time I have had it off since purchasing it it had a weird effect on the corners until it "warmed up" I am not sure if this is normal or not but since it looks fine while on I haven't bothered to investigate further. In terms of productivity, it has been a big improvement over the old 27" 2440 monitors I had been using. The extra 500px on each side really makes things easier in terms of multi tabbing, etc. I was never someone who liked having multiple apps open at the same time and proffered having any single app full screen, which meant I had to switch back and forth a lot. This has totally changed that, and I in fact never have any windows full screened on this thing anymore at all. For productivity and not gaming this is definitely a large improvement to my workflow at least, and I would imagine it would be the same for someone else's as well.

@ItsLightMind Ай бұрын

@@OminousIndustries Thank you so much for sharing your thoughts. A lot of what you've said aligned with what I was hoping i'd gain from it. I've always been a one thing max on a screen guy (ended up with a quad setup recently). So, I hope that this ultrawide can change that in me. I do game with friends sometimes, but I hope the experience is good enough in game to tied me over until more mature 38-40in oled options enter the market.

@strongme80 Ай бұрын

Buy this. Trust me. Aerospace Engineering, CAD, Embedded Systems, Agentic AI programming, that's me. Samsung Odyssey CRG9 120hz

@jayt4849 Ай бұрын

is there a way to use the groq api by chance also when am running the last example am getting [Errno 2] No such file or directory

@OminousIndustries Ай бұрын

Yes, but you would have to modify the code to point to groq instead of openai. As for the issue, I would check the issues tab in the repo to see if anyone else has dealt with that.

@jayt4849 Ай бұрын

@@OminousIndustries thanks for getting back to me i ended up figuring it out, i was watching your ollama video on how to run this with the local llms i got it all set up, edit: i got that working too grate videos bro and thanks for uploading your ollama to github

@OminousIndustries Ай бұрын

@@jayt4849 I am very glad to hear, getting to use it with ollama makes it feel more special IMO hahaha. Thanks for the kind words as well!