ASI Workshop I Rishank Jhavar

[Music] Thank you so much guys for joining in for the session today. My name is Rashank and I lead product marketing at fetch.ai which is part of the artificial super intelligence alliance. Artificial super intelligence alliance in a nutshell. What we are trying to do here is that we are trying to decentralize the entire AI ecosystem.

We sort of recognized the need of having one collective that could fight against the conglomerates that uh the other behemoths in the AI industry are becoming and we joined hands together as singularity, fetch.ai, Ocean Protocol and Kudos. Um I am personally from fetch.ai but all of us are from artificial super intelligence alliance. Um so I'll just quickly give you an overview of the fetch.

aiis stack before we begin. Um I guess I also wanted to ask how many of you are aware of AI agents in a general sense. Okay, so pretty much everyone is so I'll not give you a refresher on that. So what basically fetch.AI is enabling is an entire ecosystem of developing agents and helping people discover those agents so that they can accomplish their tasks.

It is pretty straightforward, pretty simple. Uh basically we have a consumerf facing product which is called ASI1. It's the world's first web 3 native agentic AI LLM that basically anyone on the planet can go ahead and use right now. Uh there are different models of ASI1 that you can discover for yourself in the documentation itself. ASI1 is based on the agents from agent tours.

So agent tours is a cloud-based platform for creating and hosting your own agents. uh if you guys have visited the booth I have given you a simple analogy that is uh what versel was for AWS agent was is basically for the entire agent tech space in a general sense you can just go there you can build anything that you want to on the IDE itself and then you can also host your agents and deploy it to uh the uh fetchai chain uh using the almanac contract and then we have U agents that are autonomous AI agents uh you need to use Python for that I'm pretty sure that all of you must be aware uh with Python and you must have used Python. It's pretty straightforward. It's an open source library. It has a lot of examples.

So please feel free to use that. Uh that's a general overview of the fetch.ai stack. Then we have singularity in its meta. So meta is a highly advanced form in a nutshell.

It's a highly advanced form of knowledge graphs. It will be used to enhance your operations. Pretty much anything that you're building it can use meta in a general sense. So it stands for metatype talk. Uh in a nutshell, it's a multi-paradigm language uh for functional computations uh over knowledge graphs in a general sense and obviously it's developed by singularity.

It provides you with structured knowledge representation uh which is just a organized form of information in a logical sense so that your application can consume that very easily. It has symbolic reasoning and you can perform a lot of complex logical operations and pattern matching and finally it has pattern matching and unification as I mentioned earlier. So that would be meta and we would be seeing a very quick example of that as well. Uh there is a slight change in the order of things because uh we have we have sort of moved the order up and down but uh yeah we'll give you a quick demo of that as well. So just to give you some context about U agents, um U agents are simply AI agents, right?

They are autonomous AI agents. Uh they sort of act as microservices as well. Um if you have reference for a specific AI agent framework, you can use that directly as well because U agents can wrap those AI agent frameworks very easily. Also, we already have adapters for lang chain and crew AI. So if you're aware of those frameworks, you can just use them directly.

We have a very detailed and comprehensive documentation with examples for that. Uh but if you want to deploy your agents via agent words, you need to use U agents. That's the number one thing that you need to remember. The second one is agent war. Agent wars is our marketplace where search and discovery of agents is made super easy.

You can host the autonomous U agents on agents. Agents can be created on agent. They can be hosted on agent. They can be registered to almanac contract using agent vers. And you can also use chat protocol uh using agent vers which is a pretty straightforward example.

Uh I'll also quickly walk you through what agent words is if we have some time. Then finally we have ASI1. ASI1 is the discovery layer. Basically it's it's basically a simple straightforward LLM. Uh when I say straightforward LLM we are technically ranking higher on benchmarks compared to all the other LLMs out there at least the major LM basically GPT and CLA in a sense.

uh but we don't promote it because there is another bigger USB which is that ASI1 is an agentic LLM as a user if you create an agent in the hackathon today I can just go right on ASI1.ai AI I can toggle the agent feature and I can use your agents directly if they get discovered and if you do some proper SEO which we call GEO by the way because generative generative engine optimization but yes that's ASI uh it's an agent discovery layer that can call any agent in agent words plus it also has a baked in LLM in a sense cool uh so this is the overview of building and launching with FG.AI AI. Um, I'll just quickly explain you some of the elements here. You can choose your domain first.

Anything that you basically want to build. It could be an industry. It could be any application that you ever imagine of. Um, if you want to use fetch. Ugents framework directly, you can use the U aents library and you can directly use it and deploy it and connect it to agent words.

Or if you want to use lang chain, crew AI, anthropics, MCP, you can use built-in U agents adapter. They are pretty straightforward. There's a very clear documentation for that, an example for that and it can be directly connected to agent pools. Um, you can also use a custom application. I can see that a lot of you are very uh able front-end developers.

So, feel free to connect your application to directly uh whatever UI that you build. uh but it would be nice to expose it to ASI1 where we already have a substantial amount of monthly active users, a massive chunk of users who are ready to use your agent directly and that's how you can submit it to fetch.ai track. Um if you're building across other tracks as well, feel free to use fetch.ai.

It's a horizontal technology, right? So it can be applied to pretty much anything. The LLM is extremely universal. The APIs are straightforward. So so that's already on there.

I'll also give you a quick briefer on the tools that fetch.AI basically has. Um we have U agents. Using U agents, you have U agent, ASI1, LLM and chat protocol. Chat protocol is nothing but a simple protocol that exposes your agent to ASI1.

Um then we also have compatibility with other frameworks that I talked about. The second step is to connect. If you want to build your own local agent, which is something that runs on your system or on your VM, you can use something called mailbox. Uh that's baked into agents. But I would say the easier option is to use agents because you don't need really need to spin up your own VMs.

Uh when it comes to agent vers, um obviously you can deploy it directly to agent or a virtual machine completely dependent on you. We are agnostic about that. And then you can test it directly on agent vers. There's a feature to test it using chat with agent uh that you can find on agent vers but my favorite way is to check it on ASI1 because once I develop it I want to see how people are perceiving it. So that's something that you can do and obviously finally your agent will get discovered via ASI1.

So these are the major steps that I ran you through. Basically they may sound like a lot but we have sort of tried to atomize these steps to make it very clear for you with the examples that you see and one documentation that I'll tell you about at the end of this presentation. Um that's basically a cheat code for the entire hackathon for you guys to win. So feel free to uh use that documentation. I will also give you a very quick example of the meta agent that we have repaired.

I don't think that we have a lot of time for us to sort of walk through this but yeah you can just visit the GitHub repo and sort of figure it out yourself uh if you're someone who likes brute forcing through the examples per se. So uh what we are doing here is that we are using singularity's meta knowledge graph and u agents to create intelligent autonomous agents that can understand and respond to medical queries using structured knowledge reasoning in a sense right so uh meta as I told you it's a multi-paradigm language for declarative and functional computations over knowledge graphs and fetch.ai I is an ecosystem for building, deploying and discovering AI agents. Um both of them are part of ASA. So um you can basically let me just skip through the entire thing because that will take a lot of time per se.

Uh but is there a demo here? Cool. So folks, we we did have a demo prepared for you, but unfortunately due to lack of time, we might not be able to show that to you directly. What we will do, however, is that we'll give this straight link to you. Feel free to explore the documentation.

Feel free to explore the code of this specific repo and then you can just uh basically do that yourself. But I will give you one small demo which would be um the agent was demo before we move on to Metalix. Just give me one second. Um so I've created a fresh account for you guys just to give you a feel of what agent was would look like if you haven't already visited it. Uh once you log in you'll see this screen.

All you have to do here is that you can just create an agent. Um let me just create uh I guess uh I don't know some some chat protocol agent that is uh good with a certain subject. I'll just call it bird's agent. And the agent is created. Whenever you create an agent, uh I'm I'm creating a boilerplate agent by the way.

So this agent will already have some code that's written here. Uh you will find that the same is true for all the boiler plate agents. So feel free to use them directly as well. When you create an agent, it will automatically create a readme for you which I will highly highly recommend that you edit because this readme is going to get your agent discovered via ASI1. So it would be a good idea for you to edit this specific readme.

I'm not editing it right now because I don't really need this agent per se. Um once I deploy this agent and start this agent using the button here which is the start agent button. I will automatically deploy this specific agent to chat protocol because I I picked that specific template. Um so I will just showcase you how that would sort of look like. Just give me one second.

Yeah, cool. So chat protocol is agent readme is created. This is not a good readme by the way. You guys will need to create your own. uh you can use choose your agent handle and once you get the first interaction uh your SEO or GEO would be even stronger for um the application to sort of discover it.

Uh I'll also try to sort of quickly give you uh an idea of how this agent would get discovered on ASI1. Just give me one second. So if I select agent search here, if I go at the rate and paste my agent then some info Cool. So guys, you might have to bear with me. Uh AS7 was experiencing some significantly heavy load since morning.

I believe it is due to uh the hackathon. And yeah, as I mentioned, it's uh sort of experiencing some heavy load right now. But yeah, the the core idea is that uh your agent will get discovered directly here. This is Murphy's law. Whenever I give a demo demo this, this usually breaks.

it works perfectly fine 95% of the times but uh yes once you create an agent you can go to ASI1 if it has chat protocol it will discover it pretty straightforward pretty simple and it would be a good idea for you guys to use singularities meta uh using the example that I just showed you um I will give maybe uh for a couple of minutes the forum to churag as well because metal is pretty much the most important use case that we have right now for fetch.ai AI it is a decentralized exchange and Churak will give you some more briefer on that. Churak feel free to take over.

Thank you Rish. So myself Chiak I am Metalix. I'm project manager at Metalix one of the ecosystem project of fetch.ai. At Metalix we are creating the world first P2P AI agentbased DAX and like this DAX is built using the fetch agentic u framework.

So if I talk about uh what the power that fage agentic uh framework is giving to the metal is that that is execution and the communication layer. So like uh if you talk about the metallics um whatever the agents we are creating on a metal is are kind of completely decentralized. So if I'm talking about the complete decentralized means is that the fage agents are platform independent means you can host the agents on our own platform. You can host the agent on agent vers that as Rishan demonstrate you. You can host the agent whatever the preferred cloud environment uh you want.

Then the agents are chain agnostic. So whatever the agents we are creating using the fage agent tech the are chain agnostic means they can they can kind of run on any of the chain that is ethereum, polygon or like any of the chain you preferred. They are wallet agnostic. So like you can initialize any of the wallet as per your use case. So this um one of the main USPS that stands out for the metallics is that removing the liquidity pools unlike the traditional taxes.

So we are completely removing the liquidity pools and agents can do a P2P uh kind of trades. So at the moment the metalics as a beta you can access the metal at beta.mallix.ai. So we have kind of live at the moment is uh kind of u crosschain trades where we have a multiple chains and multiple token supports like the polygon and the fetch and some of the cosmos chain and you can do a crosschain trade and like agents will do kind of all the trading stuff for you and like we also done the options part as well.

So option is currently in development and we are aiming by end of this year we will release the metallix with the crosschain feature on a main net and the option would be on a test net for the beta testing and kind of that's a pretty much from my side. So like if you guys need any of the demonstration of the metallics or anything related to the fage agent tech like how can you use on any of your web 3 kind of use case uh uh I can help you on that side. So thank you Vishan. Thank you, Churag. Uh, I would like to take these five minutes to walk you through the cheat code that I was talking about, which is the innovation lab documentation.

Uh, if you guys can go to innovationlab.fetch.ai/resources. This is not the docs, the official docs that you will see via fetch.ai/doccks.

This is completely different. Uh, we have a lot of documentation by the way. It's it's present everywhere all over. But if you go to specifically this link which is innovationlab.fetch.

ai/resources you will find that uh the examples here they are specifically catered for you guys to win in this hackathon. If you could walk through just any of the examples that are in here specifically right here you will be able to understand every bit of the technology as a whole. So for instance, if you are someone who likes to use crew AI, you can just go here and you can see a trip planner example right here. If you want to use lang chain or langraph, you can just go here. If you want to use Google's A2A, you can use this as well.

Um there are a lot more ASI example as well. I've seen that a lot of you guys sort of want to build this D5 AI agent in a sense. Uh there is a direct example that's sort of given for that as well. Plus the basics are communicated in like a super clear language here. Uh just for context guys, I'm a product marketing guy.

I'm not a developer per se. I used to develop way way back. But despite not being a developer, it is super easy for me to build these agents because of this very documentation. Um so it's pretty simple, pretty straightforward. I would urge all of you to sort of visit uh this specific link to understand more about fetch.

ai. And uh yeah, I was referring to one specific example. Let me just pull that one up. Ah, yeah. So, a lot of you guys asked about MCP in a sense and you sort of wanted to build on MCP per se.

Um, you can use the MCP integration section of innovation lab.fetch.ai to directly build on MCP. So, that is pretty straightforward here as well. Also a very cool thing we just yesterday actually we we launched agent wars MCP as well.

Uh we haven't really made that super live yet but feel free to go ahead and check it out. Um it is present uh the link is present on the website directly and obviously you have the innovation lab examples all the examples that you need right here. So yeah guys I would sort of encourage and recommend you to check all of these out. They are pretty straightforward examples by the way. If you want to build something with meta, you have something under web 3.

So you have three different singularity net examples here and meta examples here. Right? So it's it's not about how you are coding. It is more about what you're coding with this specific track. Uh if you have any idea, any horizontal idea, feel free to implement that using fetch.

ai. So yeah, it's it's pretty straightforward and uh all of us are available 24 hours round. uh we have uh beauties per se uh we have sort of shifted that so the Indian team will be available till uh 7 to 8:00 p.m. in the evening and then we have the UK team that is also available and the US team also available.

So anytime you face any issue, please come to us. Uh we have our Twitter and Telegram already given to you at least those who have come to the booth. But feel free to just reach us out and we will try to sort you out, right? So I don't think that there's going to be any Q&A. Uh it's right outside.

So feel free to come to us for any Q&A and yeah, let's let's build the agents, guys. Thank you. Thank you so much. [Music]