[Music] Good afternoon everyone. So um my name is Tane. I'm here representing U ASI Alliance. So I'm personally from Fetch. Um so during this workshop we'll be going over um some information about each each of the partners.
Following that I'll be talking about fetch uh a overview on our text stack. Uh we also have Christian from ocean who's going to talk about ocean. Uh and the second half of the workshop would be Raj. He would be showing you some demos on how you can use ASI as a consumer. Um so we'll be sharing technical docs and uh other examples on the discord.
um feel free to drop by at the booth and ask us questions specific to your use case and whatever you're hacking. Okay, so ASI is an alliance formed by these four companies. So we have Singularity.net, FetchAi, Ocean and Kudos. Uh I'll talk about like what each of the parts uh or each of the partners do and how they contribute towards our goal towards making ASI.
Um so singularity.net net uh or singularity net provides um provides us with a way to create knowledge graphs. Um so they have a system uh where you can create nodes and add knowledge and in the form of a graph. Um we have some examples with them um which we'll be sharing with you after this workshop. Um yeah um as for fetch AI we'll be seeing a full overview but we are trying to build an agent marketplace and an discovery layer where you can come in as a user ask your questions and we should be able to find the agent that does the task for you and then in the future you should be able to monetize these agents and interact with them.
Um for ocean I would like to call up Christian. So
Ocean Protocol is a suite of tools for the open data economy between smart contracts that allow you to tokenize your data sets and exchange payments for them using uh permissionless smart contracts. And then using compute the data allows you to sell permit uh privacy preserving computation. So people can train models and share data without having to uh let that data be out to the public. And then to kind of make this easier for the developers, we have an fully integrated system called Ocean Node, which helps to make it easier for developers to build by bundling services like indexing on chain. So you can make your app fast while still unlocking the benefits of kind of like onchain.
And finally, we have Kudos. So Kudos provides um infrastructure to run these agents essentially. So we'll be porting over our code onto their platform uh in the future. So let's get an overview of fetch AI. So our text stack mainly involves these three pillars.
The first is U agents which is our agent framework where you can create agents um using our Python SDK. The next bit is agent which is a marketplace where you can register your agents. And the final bit is ASI. So this is an LLM as well as a discovery layer. So when you as a user come in you can ask ask your questions.
we'll try to match your the question or task with an agent that can perform it. Um so obviously we also have like rankings and ways of changing the search based on how well your agents do. Um so we are take handling all of that. So you as a user can just come in and ask a question or if you're a developer building agents you can come in and build your agent and you have essentially distribution and discovery through our platform. So let's um see what each of the platforms or each of the pillars looks like.
So U agents is our framework. Uh after you build your agents using U agents, this is what an uh the the page would look like. So you can fill in this readme. So this readme contains metadata about your uh agent. So what it does um some sample inputs, outputs and some use cases.
Uh another thing you can see here is um ratings. So you can um see what the rating for your agent is and your goal as a developer of these agents should be to maximize this. The way to do that is make sure your agent gives good outputs, has a good readme. There are multiple factors um and we have something called as an SEO coach that you can see on the agents page um to help you do that. And you can also see the number of interactions your agent has.
So more the interactions usually better the rating. But um in the future when we turn on monetization we already have the I guess the work done for that uh once you turn monetization on based on the number of interactions your agents gets paid. Um as for agents this is our marketplace. you can come in and just um search for whatever you want to look for and there'll be a list of agents and you can see how many interactions they have, what their ratings are and based on that use those agents. And finally, we have ASI.
So ASI is an LLM and agent discovery layer like I mentioned. So you as a user can come in, ask your query and ASI will reason and try to look for an agent that does the task for you. Um, you'll see a bit more about this in the next bit from Raj, which would be Yeah. So, I'm handing over the mic to Raj. Yeah.
Uh, hi guys. So, uh, giving you a little bit overview of how we build agents on our marketplace. So basically we are not an agent framework but we are an agent marketplace. So what does it mean is anyone can build agents using the top frameworks like lang chain, crew AI, autogen or open SDK or Google Google A2A or MCPS for that matter. So they can take all these frameworks and wrap it around our framework called U agents.
Once you wrap an agent around U agent, what happens is this agent joins a network. That's our uh Fed blockchain. So all the agents that are on this blockchain can communicate with each other, discover each other and uh talk with each other. Basically send messages, transact between each other. That's the entire goal.
So build any agents with with any other framework and wrap it around with U agents. Once you wrap it around with U agents, it gets registered on our marketplace. What does it mean by registering on our marketplace is it is registering on our network basically fet network. And once it is discovered uh I mean registered all these agents like uh there are millions of agents right so many people are building agents using different frameworks but how do we bring all of them together only way is uh bring put it on a network and once they're on a network they can talk to each other communicate with each other and let's say I'm building uh travel agent using crew AI and someone is building a booking agent using lang chain how does this both uh communicate using the network. Uh that's the entire point of uh our marketplace and U agents framework.
And uh next is once all these agents are on the marketplace, we have this discovery layer called ASI1. That ASA1 is the LLM which discovers all the agents based on the query dynamically. So what how does it differ from the existing LLMs like chat, GPT, Anthropic, sorry, Claude or uh something. So these things uh claude and open AAI what they give answer based on the trained data right how ASI differs is that it goes and fetches the right agent from the marketplace so let's say someone has developed a very specialized agent uh based on a protein structure and if you go to chat GPT and ask predict a protein structure for me it doesn't predict it because it is not it does not have access to some specialized agent but RA7 will have access to specialized agent because someone has built this agent agent on the marketplace and our ASI will go and search for the right agent based on the query. As I as you seen each agent has a readme, right?
So each agent has a readme like this and based on the readme the R LLM will find the right agent for that particular task and get the task done. So let me show you this in action. So this is uh someone built this a uh agent called bolds to biological structure prediction agent and put it on our marketplace. Now when you query something like this on the asylm it will directly give you the structure because it is accessing the marketplace and you can uh visualize the structure. Let me actually show you the marketplace.
So this is the marketplace where you can find different agents and all these agents are discoverable on the ASI. So I'll go to the ASLM actually. So this is the ASLM interface. We also have a mobile app for this. And here we have something called agent search.
Once we turn on the agent search and search for anything uh any query, then it will go and find the right agent for that. So for example, I'll show you a simple I switched on the agent toggle and I'm asking so what happens when I ask that it goes and reasons that it needs to find an agent that is suitable for this particular action on the marketplace and once it finds the right agent it will uh execute that particular logic whichever the developer has done. So as you see it is saying I've searched across agent words for agents capable of fulfilling your request for this one. Uh the query yielded seven agents across a domain specific and it has found one agent. It is called GIF agent and uh it is giving like that.
So like this developers can build any agents they want and put it on the marketplace and whenever the user queries that particular uh query related to this agent the agent will get used and the developer can monetize uh it as an agent as a service. Uh that's one thing and uh other thing is we have business agents. So let's say someone want someone as a business wants to develop an agent and put it on the marketplace. So here the query is little blurry but I'll show you. So someone asked I need to get my car serviced in Hingley and now it is going and searching for agents that are on the marketplace related to this car service and it has found an agent called SNR autos and tires.
It's a business agent. So someone has deployed this agent and this agent knows entirely about the business of this particular car service and it has uh started using that agent. Yeah. One more important thing is once you found an agent they can do at the rate and the rate the agent name and then ask the query. So let's say someone has built a an agent and uh uh the user actually knows about this agent.
So you don't have to go and dynamically search for it. Just like on Twitter or someone how you tag some some particular person here you can tag an agent and then directly ask the query if you know the agent name already. So I tagged this agent and I'm giving the uh my car uh it is asked can you please confirm your vehicle registration number. I gave the vehicle registration number and the user uh it found the user vehicle information based on the registration number and it is asking us to confirm if the if the user is the right person and uh once the vehicle details are given it has found some like exactly five available dates for the service appointment and all details are confirmed and also after that uh once the details are confirmed the person can also pay directly through the interface. So the service appointment for so and so particular date and time is confirmed and after that the person can actually pay.
So directly generates the link for the payment and the person will be able to pay. So this is one type of payment where agent is being used as a service and other type of payment is agent to agent payment. So one agent can transact with other agent based on some allocated uh amount. That is another uh use case that we are coming up with. Yeah.
So basically you are uh I hope you understood the entire uh end to end solution of how our framework then the marketplace then the discovery layer that is ASI1 then now yeah we are open to questions you can yeah
infrastructure Uh, sorry. Can you repeat that?
What do you want?
Oh, for the hackathon.
So, as a hack uh for the hackathon, you can uh build agents build. Sorry.
Yeah, we have also we uh the other partners like Ocean Protocol and uh uh Singularity Net. using ocean protocol you can uh like get some data sets and build an agent which is uh like using ocean protocol uh or like uh without if you don't want to use ocean protocol or singular if you want to build only a particular AI agent you can build that and put it on the marketplace and uh for the objective is we want to see if it is if your agent is discoverable on the ASA1 and if it is uh a very good application Y
uh
ocean protocol. You mean
uh yeah ocean protocol you can
uh the reasoning engine of AS7 Yeah.
Okay.
So, okay. So, we are technically from fetch. So, we are not really experts on this, but from what I have understood uh singularity net has its own language called meta where you can define uh like nodes and like what points to what. So we have an example which we can share with you guys which is for like diseases and symptoms and treatments. So you can come in and represent that data as as a knowledge graph where you come in and say okay I have cold and it'll try to like go through the knowledge graph and figure out um what the treatments are or you could give a set of symptoms and then see what where they those point to and what disease they point to.
So I I guess looking at the example would be a bit more uh I would make it a bit more intuitive and easy to understand. Um but that's what we have. So yeah.
So yeah.
Yeah. So we are basically looking for uh you to build something uh like as fetch we are looking at you to build an AI agent that is discoverable on our ASILLM. You would get a lot more points in the judging if you use the other partners and we are encouraging that.
Yeah. Yes.
Question.
Sure.
Do you have a tutorial?
Yes, we do.
Y locally or is it Yeah. So you can build everything locally on your system and then if you're happy with it, you can uh upload things onto agents. For security reasons, all packages are not allowed on agent work. So if you if your package is not allowed, you can run it on your local machine and that can be connected to ASI as well. So you have mechanisms to do that.
Yep.
Not sure if I just missed it. When you put in seem to find the relevant really fast.
No. So, uh I guess for the demo we have trimmed out a bit of the part but it has some reasoning bit where like right if you saw in the demo it just showed like the full reasoning. Uh actually it will like stream out the response and it'll go through that. Uh we just trimmed that so that it's quicker to show during the demo.
So it wasn't that there was very specific key phrases. It was just really good finding the right.
Yeah. Yeah. it it like if you try something different like it doesn't need to be this exact phrase that he typed in. Uh you can just play around with the platform and uh yeah test it out. Any other questions like something you guys are building and you want to find out how you can build that?
Yeah.
Yes. So we have an so for the LLM if you just want to use it like you would use the other LLMs there is an API that we provide and we also have um internally we it's yet to be published but we will be uh we also have an API to use the agents through the API like uh so it won't just be like a chat completion it would also be using the agents so if you're planning to use that come to our booth we can provide that to [Music]
Automatic transcript — names and jargon may be misspelled.