Agency Swarm: Why It’s Better Than CrewAI & AutoGen

In this deep dive, I’ll explore how Agency Swarm is challenging CrewAI and AutoGen. Join me as I demonstrate how to use Agency Swarm’s framework to develop a simple web search agent. Stick around till the end for my final verdict on this innovative tool.
Need to develop some AI? Let's chat: www.brainqub3.com/book-online
Register your interest in the AI Engineering Take-off course: www.data-centric-solutions.co...
Hands-on project (build a basic RAG app): www.educative.io/projects/bui...
Stay updated on AI, Data Science, and Large Language Models by following me on Medium: / johnadeojo
GitHub repo: github.com/john-adeojo/agency...
Chapters
Introduction: 00:00
Key features of Agency Swarm: 01:00
Python Code - Prompts & Schema: 06:10
Python Code - Custom Tools: 11:00
Python Code - Agents: 12:50
Demo: 18:15
The verdict on Agency Swarm: 41:10

[@vrsen]

Wow! Amazing tutorial. I am so glad to see someone finally taking a deeper look into the code and building actually useful ai agents, instead of a trip planner.

[@Data-Centric]

Thanks for taking the time to watch! Keep up the good work

[@jfrposada]

Hey @vrsen and @Data-Centric! Im a big fan of agencySwarm but I ma lacking of knowledge and eould like to learn more. Is there any documentation or tutorial on how to build your agency from zero using the agencySwarm framework?

[@ChefDomein]

One of the most useful tutorials on agents I have ever seen! Keep up the work and please more videos on agency swarm. This one is by far the most useful agent framework!

[@MYPROTIPS]

Dude!!!! Amazing work!! Everyone else in this field assumes your a coding genius and noone explains thoroughly how to do anything. Im pretty sure that a tutorial should explain whats happening lol. Your Video secured me as a viewer within 3 mins and I subscribe rarely. Cant wait to see what else you put out.

[@augustus6660]

As always, amazing content, seriously. Good to se an impartial discussion about these frameworks. Impartial doesn't mean not giving your opinion, in my opinion. It means explaining the reasoning behind it, and inviting others to inform themselves so they can either agree or disagree.

[@OliNorwell]

Good video, for anyone out there wondering the “next premier league game” question did correctly determine that Arsenal are playing at home in the next round of premier league games. *however* all 10 games are kicking off at the same time, meaning the answer is impossible to give. It didn’t seem to realise that, so is a fail really. I wish the same question had been repeated with 4o, then we could have seen any difference in approach.

[@CharlesZerner]

Data Centric, I am so glad to discover your channel. Your explanations are so wonderfully clear and well spoken, informed by lived experience (on cutting-edge topics), unpretentious, measured, and thoughtful. Video after video--the quality is amazing. Thank you so much for your contribution.

[@benh8199]

Thank you for making this video based on your viewer's request! Keep it up!!

[@jarad4621]

Did you request as well? Are you using it AS? Havent started yet so busy but wondering if its good, better, such lack of good agents

[@scotter]

I love your way of giving tutorials! I'll dive fully into Agency Swarm if/when I can use underlying LLMs other than OpenAI. Philosophical reasons.

[@jarad4621]

Finally agency swarm thanks for doing my request! Legend, ir you can do a review for open source model version that would be epic

[@paulmiller591]

Thanks, John. Great video. I am new to agents and love your explaining how to use this and why Agency Swarm might be worth where I should start on the agent side.

[@elcaribbeannomad2079]

Since John recommended it, I'd better check out this framework now!

[@donconkey1]

I appreciate the excellent and informative video, which taught me a lot and saved me a significant amount of time.

[@keithmifsud]

Awesome video. Very informative and you sounded very charismatic. Thank you for adding your opinion at the end of the video. I agree, we should wait for the Assitants to be out of Beta and also have a clearer fee structure.

[@oryxchannel]

you're in my algorithm. Subbed.

[@sd5853]

Thank you i'm grateful for your videos, liked and subscribed

[@bernieapodaca2912]

VRSEN’s Agency-Swarm is by far the best framework because it is Open Source and because you can reconfigure and make it your very own. He also has the Genesis project where you can incorporate tools contributed by others into your project. He has also reconfigured it to use Open Source LLM’s. Great project and team!!

[@yamnel]

I enjoyed this video. Thanks for the engaging and easy-to-follow information. I especially liked the explanation of the Premier League I saw the effort it took you to say "Soccer League!" 🤣

[@lovestocofish]

Thanks for reviewing all these tools so we don't have to. 😊 I think I saw a video from the creator where he did add support for OpenSource LLMs

[@SeattleShelby]

Manager not doing anything… so you’re saying it’s just like real life?

[@Bana888]

Great analysis and verdict, as always! Feedback from some on Agency-Swarm was its hard dependency on just using OpenAI models, and no support for open source models. It supposedly was addressed recently so if you could evaluate that, it would be awesome.

[@SethCohn23]

Yeah, seems rudimentary at best, no ollama/etc yet. Nonstarter

[@jarad4621]

Nope it can do all open source models now it's been upgraded

[@tomw4688]

agreed. would love to see a video evaluating its open-source support.

[@zatoichi1]

You weren't mistaken in seeing John Quincy Adams. He was John Adams' son and the 4th President of the US.

[@realCleanK]

Thank you!

[@lLvupKitchen]

Great video

[@maxh96-yanz77]

Appreciate to all videos that give us enlightening knowledge. Could you please also make video about LLM OS , how and what the pros cons compare to other agent 🙏🙏🙏

[@Dominik-vk4ie]

Thanks for your videos. The materials are very good and well explained. It would be nice if you could add something about open-source models.

[@JulianHarris]

Super helpful thank you! You’re keen on the Assistants API but the LLM lock-in is pretty severe as a result. What, over having server-side memory vs your local chat history are the compelling benefits? The cost is a major issue that would be alleviated at least a little with open weights support.

[@vidfan1967]

Thanks for the good content! Seems like good system for one-off activities. However, I am wondering how consistency across multiple runs is ensured as the tools and the websites could be different with every run.

[@aimademerich]

Phenomenal

[@csmac3144a]

Great explanation. Just a note regarding cost of inference -- what may seem expensive to some is trivially cheap to others. I work on tools that supplement people who cost $700 US per hour. The cost of inference isn't even a rounding error for us. The quality of the result is all that matters. Other projects may be far more constrained in terms of budget. Cost is a relative thing.

[@Data-Centric]

An excellent point! Would you mind telling us a bit about these tools you're building? At $700 per hour it seems like something in professional services...

[@vastvitamins1966]

Really great video i like this framework you should start a community.

[@DanielBandi]

Great video, I just hope you do more videos with open source alternatives like ollama, instead of openai

[@jarad4621]

It now works with ollama and open source it's been updated making it the best framework for open source agents

[@am0x01]

Hi John, Thanks for this content. Seems like you came to the conclusion with your first test, that what Andrew NG mentioned recently about GPT3.5 outperform 4.0 Turbo in agentic setup was not totally true. I'll look into the repo to find out if it works with OpenLLM's, can't afford paying OpenAI😅

[@scotter]

Apologies if you gave the following information and I missed it. What was the price difference between using 3.5-turbo and 4o?

[@domjoe18]

The problem with the temperature seems to be with the OpenAI API. When I try to retrieve an agent with a temperature of 1.0, the API returns a value of 0.0.

[@travisporco]

How flexible is it? I found no way to use crewai to look at images

[@yt-caio]

Great content! OpenAI models are great but if you don't have revenue it gets expensive pretty fast to prototype and repeatedly test. I spent R$5 in two dev sessions using agents. Then I went all in open-source models running locally. Currently I stay away from content that pushes to paid API's but I like your analysis and opinions.

[@yt-caio]

Actually it was $10 dollars I spent since you get $5 for free and I put more $5 on it

[@yt-caio]

Hidden prompts also increase costs because of additional tokens

[@jarad4621]

Please note bit outdated swarm can now use all models not just openai, local and open source i think it's a new repo called Lab

[@Data-Centric]

Going to see if I can do somthing with Open LLMs, maybe using Ollama.

[@jarad4621]

@@Data-Centric That be awesome thanks man, enjoy your content well delivered, andm more will keep an eye on here

[@alvinjamur1]

Definitely look at Langroid

[@asi_karel]

thanks ou for the hint

[@lavamonkeymc]

Does it work with azure open ai?

[@MrAnt1V1rus]

No. I've wasted a ton of time trying to get this working and the Discord is totally unhelpful.

[@zippytechnologies]

we need to be able to use locally hosted llm or private server llm - no more giving all our prompts and data to OpenAI - there def is a time and place for OpenAI - such as need for the new 4o model - it is fantastic - but no need to waste money on simple thoughts and research tasks that can be done with ollama and local hosted llm's

[@ayushman_sr]

Assistance api is expensive

[@SeattleShelby]

Yeah… but… Langchain RAG implementations for research agents appears to be way more robust than just using CrewAI or AutoGen by itself.

[@fbrand]

Thank you for this review. But please do a complete setup with Agency-Swarm-Lab using "GROK\LLama3 70B" and compare with what you've shown in this video. Thanx.

[@SixTimesNine]

Thanks. One thing though: Zoom in a bit next time please. Well, a lot actually.

[@carktok]

Yeah or increase your font size CMD-+

[@JoeSmith-kn5wo]

I like most of the videos but decided to click off as soon as you said only works with OpenAI. I am not a fan of being locked into a closed source models provider. Unless Ageny Swarm opens their framework to other models they are going to have a tough time building traction.

[@veryVANYA]

its open-source compatible as of last week

[@JoeSmith-kn5wo]

Well I stand corrected. 🤣 I'll check out the rest of the video. I was confused why Agency Swarm wouldn't offer open-source model capabilities now a days. I mainly use Ollama for model hosting.

[@jarad4621]

Yep updated will be perfect for your ollama models in super excited by that

[@MrGluepower]

Agency Swarm might be good, but I doubt it and have no intention of checking it out. The founder's attitude, acting like he is the best thing since sliced bread, belittling other solutions instead of pointing out his strengths, just doesn't work for me.

[@zatoichi1]

So do emotional responses and arbitrarily shutting out technologies create the most effective solutions for you?

[@nullvoid12]

BS