Hi Colin, I really appreciated your video, which explains much better than others that specific retrieval needs can be encapsulated within a dedicated agent acting as a domain expert. By the way, it comes to mind that classical LLM-RAG could be implemented by delegating the retrieval part to a Retrieval Agent that could search not only in relational databases but also in various types of documents. It would be interesting to see how a complex user query (say, involving multiple source searches) could be resolved by a multi-agent system that also shares data among agents using context variables. I’d stay with SWARM because it’s so nice and simple. Additionally, I wrote an article titled "SWARMing Conversational AI: Integrating No-Code and Code in Agent-Based Workflows," which you can find online. I would love to hear your feedback on my perspective. Thanks again! Giorgio
Thank you for easy understanding of swarm framework. So it's basically langchain but it's radically simple and easy to use. I don't like it makes many OpenAI API calls on every step but oh well, that's their business I guess. I think it'll be matter of time until someone makes it compatible with other LLM services and local models.
I'm glad you found it easy to understand! Yes, that's a great way to think about it. Since Swarm is open source, you can make your own version to use local LLMs so you aren't paying for every step of the way. That is something I am going to do myself and make a video on!
You're welcome! Thanks for the suggestion! I am actually planning on creating a video in the near future where I fork Swarm and make it possible to use local (open source) LLMs!
Thank you for such wonderful content! I got a similar video today. They don't explain it as well as you do. I have a high opinion of you. Look forward to sharing the next AI platform
Great content! Could we mix Swarm with Langchain and create Agents that execute many tools? I mean by having one main agent to not overwhelm the LLM and others with specific tools? (Actually, isn’t it similar to Crew ai?)
Thank you and I love your thoughts here, Arthur! Swarm does a lot of what LangChain would typically do to help you with a multi-agent setup. But you could still certainly integrate the two together to create something really robust using more complex chains made possible by LangChain! This is pretty similar to Crew AI, though I like LangChain and Swarm more because I find it easier to use and customize than Crew AI.
I really like the way Swarm lets you compose groups of agents like this. It feels a lot easier than LangGraph/Chain. Is there an easy way to use this with a local LLM like ollama?
Yeah I agree! I think LangChain/LangGraph is more customizable but Swarm definitely gets you up and running faster. Swarm only lets you use GPT right now, but since the library is open source, you can easily make your own version to use Ollama under the hood for the LLM instead! I am actually probably going to do this myself and make a video on it.
Thank you for the nice video. Is the any solution for asking complex questions which should trigger several agents and combine the answers back. For example asking about count of users and popular categories in one prompt?
My pleasure! And fantastic question! You could certainly extend this application to handle that sort of thing. You could change up the prompt to the router by telling it that some requests will need to be handled by more than one agent. Then you instruct each agent to pass control back to the router after it provides its part of the answer so the router can route to the next agent to continue/finish the answer. I hope that makes sense! Definitely a more complex implementation but it is certainly possible.
i was wondering if you would ever make a tutorial on how to create a custom node for n8n? since i want to create a playwright node and improved node for anthropic claud that allows me to input images - like chatgpt vision. i havent found anything good on the internet to showcase the how to make a custom node everything seems outdated or overly too complicated.
I appreciate you asking! I haven't actually made a custom node in n8n yet, but I have spent a good amount of time researching how to do it since that level of customizability in a platform is important to me. Once I have a need to make a custom node, I will 10000% make a video on how I did it!!
Great question! What VRSEN has built with Agency Swarm is awesome. I would say that Swarm is comparable in performance but I found Swarm easier to use. Granted I haven't set up the exact same set of agents on both, but that's just my general experience!
Yeah you're right it's not truly a complex query! But it's more than a simple SELECT so it worked for my demonstration. I will certainly be testing it out with more complex database structures in the future, just had to start with something reasonably simple!
Consider these actionable insights from the video: 1. Explore OpenAI's Swarm framework as a new way to manage and interact with your SQL databases. 2. Understand the concept of specialized AI agents and how they collaborate within the Swarm framework. 3. Identify areas within your data management workflow where specialized AI agents could improve efficiency and accuracy. 4. Visit the provided resource links to learn more about Swarm, explore examples, and access documentation. 5. Experiment with building your own network of specialized AI agents using Swarm's Python library.
You just forked Bolt can you add this to it? But using obviously OpenAI, but also Claude, Mistral through API. Then Ollama to run with this also Swarm is opensource. Then you will be the most Best Architect Coder ever. I am already using Swarm and it is damn good. But I want to see it work with other models...
So I actually am looking into developing an agent platform behind the scenes with my Bolt.new fork to get better results, especially for those smaller local models that don't always open up the webcontainer as many people have noticed! And I could certainly use Swarm for that to use something already out there that is open source. Thanks for the suggestion - I seriously love it!
@@ColeMedin Please do it. I will donate 500 USD to your channel. I want to use your bolt so that online models talk to offline models and the both communicate in the SWARM framework, through an API.
yeah PayPal or patreon which ever you prefer. I will send you the interface if you want. I am doing all this for the I am AuRa Foundation. Online model API + Offline model + Swarm framework to work with both. That is basically 01 + Sonnet. Llama 405B and its variants truly are amazing.
Sounds fantastic!! My PayPal is @ColeMedin, and I certainly don't expect anything, especially before I put out content on this! But I am at this point confident I'll be implementing this within the next month and making content around it, especially with your support!
GREAT question, I appreciate you asking! It doesn't out of the box, BUT since Swarm is open source, you can easily change it to work with Ollama as well. I am considering making a video on this!
There are some similarities! But LangGraph does less out of the box. It's more customizable for creating agentic workloads but you have to do a lot more of the work yourself with agents you set up with LangChain (or a similar/custom framework).
Good question! CrewAI is a similar tool but from my experience (and it's limited because Swarm is new!) Swarm is easier for me to use. Performance is similar but the developer experience is really important to me too.
Interesting! kyegomezb's Swarms looks quite different than OpenAI's Swarm, and agentic orchestration tools are such an obvious and important idea that it doesn't seem clear to me it is a copy. Maybe there is more to it though!
@@ColeMedin yep i can't find which part is the copy... well I'm not coder... however if this doesn't end well... in my opinion big companies should and maybe must be able to nurture open source developers... without that kind of support open source will be felt like nasty bottom tier of stepping stones where every developers hitting each other's just to survive sad
for me, just trying to understand swarm alone is already a challenge but then to be overwhelmed with sql is too much for one tutorial. i think a simpler example to just illustrate swarm would have been better, especially for an introduction. but thanks though.
I totally understand, thank you for being honest with your feedback here! I was debating what kind of solution I wanted to make with Swarm and I did consider going with something much simpler, but I'm also thinking people want to see Swarm do something more than the basics because they already have for other agent frameworks. Considering what you're saying though, I am planning more content in the future that keeps things to a true introductory level just to lay the foundation really well for topics I cover on my channel!
Thanks for the great video with links and no ebook or course selling. You will grow fast.
My pleasure - thank you man!!
Hey Cole, discovered your channel a couple of days ago and I love i. Keep up the good work.
Thank you for such great content mate! There’s so much fluff out there, I’m so grateful for you sharing your knowledge.
You're welcome - it's my pleasure! 😀
Have a sub for doing sponsorship right and staying open with your code.
Thank you very much! :D
Hi Colin,
I really appreciated your video, which explains much better than others that specific retrieval needs can be encapsulated within a dedicated agent acting as a domain expert. By the way, it comes to mind that classical LLM-RAG could be implemented by delegating the retrieval part to a Retrieval Agent that could search not only in relational databases but also in various types of documents.
It would be interesting to see how a complex user query (say, involving multiple source searches) could be resolved by a multi-agent system that also shares data among agents using context variables. I’d stay with SWARM because it’s so nice and simple.
Additionally, I wrote an article titled "SWARMing Conversational AI: Integrating No-Code and Code in Agent-Based Workflows," which you can find online. I would love to hear your feedback on my perspective.
Thanks again!
Giorgio
Thanks, a very good intro into swarm. Directly useable 😊
That's certainly the goal - thank you!
Awesome video Cole!
Thank you Pascal!
Thanks for this. I've been creating my own sql db's for the past few months. I think I'm gonna try sqllite, and pstgresql as the next stepping stones.
You bet! Sounds great man! 😄
Thank you for easy understanding of swarm framework. So it's basically langchain but it's radically simple and easy to use.
I don't like it makes many OpenAI API calls on every step but oh well, that's their business I guess.
I think it'll be matter of time until someone makes it compatible with other LLM services and local models.
I'm glad you found it easy to understand! Yes, that's a great way to think about it.
Since Swarm is open source, you can make your own version to use local LLMs so you aren't paying for every step of the way. That is something I am going to do myself and make a video on!
Thanks for this overview. Can you make a similar product with Open Source model so we can compare version?
You're welcome! Thanks for the suggestion!
I am actually planning on creating a video in the near future where I fork Swarm and make it possible to use local (open source) LLMs!
Thanks!
My pleasure - thank you so so much for your support!! 😀
Thank you for such wonderful content! I got a similar video today. They don't explain it as well as you do. I have a high opinion of you. Look forward to sharing the next AI platform
Thank you so much for the kind words, that means a lot to me! 😃
Great content! Could we mix Swarm with Langchain and create Agents that execute many tools? I mean by having one main agent to not overwhelm the LLM and others with specific tools? (Actually, isn’t it similar to Crew ai?)
Thank you and I love your thoughts here, Arthur!
Swarm does a lot of what LangChain would typically do to help you with a multi-agent setup. But you could still certainly integrate the two together to create something really robust using more complex chains made possible by LangChain!
This is pretty similar to Crew AI, though I like LangChain and Swarm more because I find it easier to use and customize than Crew AI.
I really like the way Swarm lets you compose groups of agents like this. It feels a lot easier than LangGraph/Chain. Is there an easy way to use this with a local LLM like ollama?
Yeah I agree! I think LangChain/LangGraph is more customizable but Swarm definitely gets you up and running faster.
Swarm only lets you use GPT right now, but since the library is open source, you can easily make your own version to use Ollama under the hood for the LLM instead! I am actually probably going to do this myself and make a video on it.
Greate effort and content. Thank you very Much.
Thank you very much - my pleasure! :D
Thank you for the nice video.
Is the any solution for asking complex questions which should trigger several agents and combine the answers back. For example asking about count of users and popular categories in one prompt?
My pleasure! And fantastic question!
You could certainly extend this application to handle that sort of thing. You could change up the prompt to the router by telling it that some requests will need to be handled by more than one agent. Then you instruct each agent to pass control back to the router after it provides its part of the answer so the router can route to the next agent to continue/finish the answer.
I hope that makes sense! Definitely a more complex implementation but it is certainly possible.
great concise content!
Glad you enjoyed it - thank you! :)
Great video
i was wondering if you would ever make a tutorial on how to create a custom node for n8n? since i want to create a playwright node and improved node for anthropic claud that allows me to input images - like chatgpt vision. i havent found anything good on the internet to showcase the how to make a custom node everything seems outdated or overly too complicated.
I appreciate you asking! I haven't actually made a custom node in n8n yet, but I have spent a good amount of time researching how to do it since that level of customizability in a platform is important to me.
Once I have a need to make a custom node, I will 10000% make a video on how I did it!!
Do you do custom work for clients?
Indeed I do! If you're interested, feel free to reach out to me at cole@dynamous.ai
Thank you for not doing the bare minimum of just reviewing code, instead doing a proper demo.
You bet man!! Thanks for the kind words!
Better than agency swarm from vrsen?
Great question! What VRSEN has built with Agency Swarm is awesome. I would say that Swarm is comparable in performance but I found Swarm easier to use. Granted I haven't set up the exact same set of agents on both, but that's just my general experience!
Single join isn't complex query, feed it proper monolith faked database and see if it works then.
Yeah you're right it's not truly a complex query! But it's more than a simple SELECT so it worked for my demonstration. I will certainly be testing it out with more complex database structures in the future, just had to start with something reasonably simple!
Consider these actionable insights from the video:
1. Explore OpenAI's Swarm framework as a new way to manage and interact with your SQL databases.
2. Understand the concept of specialized AI agents and how they collaborate within the Swarm framework.
3. Identify areas within your data management workflow where specialized AI agents could improve efficiency and accuracy.
4. Visit the provided resource links to learn more about Swarm, explore examples, and access documentation.
5. Experiment with building your own network of specialized AI agents using Swarm's Python library.
I love it, thanks for calling all of these out!
We were getting some errors when transferring back to the master agent. Did you experience this?
I didn't actually - I'm sorry you're running into that! What are the errors exactly?
You just forked Bolt can you add this to it? But using obviously OpenAI, but also Claude, Mistral through API. Then Ollama to run with this also Swarm is opensource. Then you will be the most Best Architect Coder ever. I am already using Swarm and it is damn good. But I want to see it work with other models...
So I actually am looking into developing an agent platform behind the scenes with my Bolt.new fork to get better results, especially for those smaller local models that don't always open up the webcontainer as many people have noticed! And I could certainly use Swarm for that to use something already out there that is open source. Thanks for the suggestion - I seriously love it!
@@ColeMedin Please do it. I will donate 500 USD to your channel. I want to use your bolt so that online models talk to offline models and the both communicate in the SWARM framework, through an API.
Wow that's very generous of you, thank you so much! It'll take some time to develop it but I do have it in the pipeline because yeah it'll be huge!
yeah PayPal or patreon which ever you prefer. I will send you the interface if you want. I am doing all this for the I am AuRa Foundation. Online model API + Offline model + Swarm framework to work with both. That is basically 01 + Sonnet. Llama 405B and its variants truly are amazing.
Sounds fantastic!! My PayPal is @ColeMedin, and I certainly don't expect anything, especially before I put out content on this! But I am at this point confident I'll be implementing this within the next month and making content around it, especially with your support!
Good content
Thank you man!
This is so good!
Thank you!! :D
does swarm work with ollama too?
GREAT question, I appreciate you asking!
It doesn't out of the box, BUT since Swarm is open source, you can easily change it to work with Ollama as well. I am considering making a video on this!
Looks a lot like LangGraph
There are some similarities! But LangGraph does less out of the box. It's more customizable for creating agentic workloads but you have to do a lot more of the work yourself with agents you set up with LangChain (or a similar/custom framework).
What about crew AI?
Good question! CrewAI is a similar tool but from my experience (and it's limited because Swarm is new!) Swarm is easier for me to use. Performance is similar but the developer experience is really important to me too.
OMG RSS is still alive
Haha maybe not for everyone but it is for me!
any one heard kyegomezB said openai swarm use his idea? x post kyegomezb swarms
Interesting! kyegomezb's Swarms looks quite different than OpenAI's Swarm, and agentic orchestration tools are such an obvious and important idea that it doesn't seem clear to me it is a copy. Maybe there is more to it though!
@@ColeMedin yep i can't find which part is the copy... well I'm not coder... however if this doesn't end well... in my opinion big companies should and maybe must be able to nurture open source developers... without that kind of support open source will be felt like nasty bottom tier of stepping stones where every developers hitting each other's just to survive sad
Yeah very true!
for me, just trying to understand swarm alone is already a challenge but then to be overwhelmed with sql is too much for one tutorial. i think a simpler example to just illustrate swarm would have been better, especially for an introduction. but thanks though.
I totally understand, thank you for being honest with your feedback here!
I was debating what kind of solution I wanted to make with Swarm and I did consider going with something much simpler, but I'm also thinking people want to see Swarm do something more than the basics because they already have for other agent frameworks.
Considering what you're saying though, I am planning more content in the future that keeps things to a true introductory level just to lay the foundation really well for topics I cover on my channel!