Hello Agents: Building AI Agents on Top of Hedera | Jake Hall - Hedera

[Music] hey everyone web 3 has a big problem user experience I think I think we can all agree with that all of us here have probably got past our first wallet but I think 99% of the population haven't quite and I want to ask you a quick question just before I proceed with my presentation on building AI agents on Hira can AI agents fix that we'll find out um so just want to quickly say uh this can be quite interactive uh I'm just put up on my screen that's my that's my Twitter I'm not I'm not telling you to follow me more than welcome to if you have a look at the pin tweet on my Twitter account or you scan that QR code you'll get taken to a Google doc which contains information as to how you can set up the uh relevant tools and and and agent sdks that we're going to be using so I'll leave that on here just for the just for the time being uh while people scan it or scan it for later or or have a look at my Twitter and get it up on your laptop uh on it you'll find instructions how you can interact and run the git pod for our Eliza plugin Eliza being a very common AI agent framework and also how you can set up our Hira agent kit and Lang chain Integrations cool before I just continue uh and and talk more about the agents I'll give a very brief background as to uh llms and what they are I'm going to assume a lot of people be using them regularly but on a very high level neural networks trained on loads and loads of data you give it a prompt it basically predicts what's going to come after that it generates whether it be text or images or other media forms based on the probability of all of the the the feed data it's taken in that big training phase and uh AI agents what are AI agents well they basically use large language models along side tools or actions which are deterministic pieces of code that an AI agent understands and knows okay I've got this function over here might be a JavaScript function or a python function I've got this function I know what that function does maybe it creates a token and it knows the arguments and the and the uh the required data to execute that function um if you haven't already noticed these images are all generated by AI so unfortunately our agents or agent they don't have ai quills or database quys but they do have API qus and database queries as well so uh thought that was quite funny and uh yeah may maybe not the best at generating text on on images yet all right so what do we have on hadir well we've got our Hira Agent K SDK this is essentially a JavaScript typescript class which Aggregates all sorts of different functionality on hadir it's relatively opinionated but the beauty of this is you can just use this and Inter integrate it with any sort of uh existing AI agent framework we've done that with our Eliza plug-in I'm going to be showing that today if you scan the QR codes earlier you'll also be able to follow along and also our hadir Lang chain langra agents all of those are essentially using the Hira agent kit but if you're wanting to build maybe a autogen agent uh you could use the hiir agent kit and leverage that also before I quickly go to actual like a live coding demo where we can see these agents in action we do have three bounties this year top one AI agents $225,000 that's uh split over the top four so the winner gets $15,000 and we've got evm stuff as well explorers for those of you uh who are sort of getting into Hira but that's a perfect time to sort of Leverage some of these tools you can maybe build a Dap which integrates them I mean but also it doesn't need to be with agents or you can actually contribute to our open source sdks all of that information is on this Google doc I'll quickly share that Google doc so if you go uh again on my Twitter that pin tweet AI agents on hadir uh as well as it's got setup guides uh I've already started the setup so I'm going to show you how to run the git pod environment for our Aliza plugin uh and then also we can test out our Lang chain uh langra agent let's start off with our uh hiir agent kit now we do have a that might need to be lower case uh we do have an open API KY or a router uh API KY which you can use so so be quick and and and sort of there's not there's not loads of credits but if you want to test it out you can do that uh I'm going to start plugging it into our git pod environment if you want to follow along click on that open git pod it will start setting up uh in the background by now mine should already be installed and it's been built all I need to do is just pass in my API key uh so I'll allow that and then I need to go to my Hira uh my Hira account idid so I can find that actually on the hiir portal which uh if you go over here let's uh click on that go to the hiir portal you can create your own account over here uh we'll use this account of mine with the hiir portal you get a th000 harar test net H bar every day which is more than enough for basically anyone uh unless you're really stress testing the network and I'm going to get the private key as well and this is a ed25519 key you can of course use ecdsa key so just for you evm Builders here if you do have an existing metamask wallet you can use those exact credentials over here uh Ed 2551 n cool so that's Now set it up it should automatically start starting the application and also running the front end cine I'm just going to be doing a few demos uh just showing the the user interface which comes out sort of out of the box and we can start playing around with it but of course if you're familiar with has anyone anyone here used the lier actually I guess I can gauge how how deep to go in you Lies over there anyone else heard of it no okay heard of it cool right we I I'll give a quick uh tldr hyzer pretty exciting uh framework open source framework for building llms very popular in the web 3 space not so much sort of in in in the web two audience um but it's it's pretty cool because they've got really nice Integrations into things like Twitter or X or uh Discord telegram so you can have your AI agent almost being like a social agent having its own personality its own character file as well as other plugins like the hadir plugin which would allow it to potentially create tokens send tokens trade whatever whatever sort of uh uh whatever your mind can think of um I I I'll before I go into that I I'll show broadly what the Aliza structure is so I've got these character files over here we're currently using this Hira character Json it's of specified for the hidea functions but there are also other uh character files that you can use this gives your agent its personality um and uh so yeah we we'll be using this one which is sort of optimized for Hira you can also see plugins and import different types of plugins you can have evm plug-in you can have uh non non- crypto plugins but we'll start chatting with it so this is the client which gets spun up you don't need to build agents using this let's just make sure it works hello Moment of Truth good stuff cool well uh let's show some of the functionality that it has and our our uh plugin is pretty extensive because it leverages that base uh hadir agent kit class so if we go over to our plug-in you can find that inside of uh this directory hia plugin elizer and you can see inside of the actions all of the all of the function ities that we have so these are the actions which is importing from that base hdea agent class making it really easy to integrate with Frameworks like lier and these are all of the tools or the actions which the AI agent understands and sort of if if we have a look at any one of these actions we'll see it comes with a qualitative description it comes with examples as to uh potential prompts that might be used to trigger it uh and let's have a look at the there should be a description here it Associates the private provided token with a given account this is what the AI agent uses to determine which function do I need to call so we can say uh Hey create a token for me called eth Denver and we could say with a symbol there are a few minimum requirements that you need so we need a symbol e FD and uh also with an initial supply of however many that is and two decimals so this now this qualitative prompt the AI agent will be understanding this and going okay huh this guy wants to create a token hang on I've got a function which allows creation of tokens and as you can see we've created this token uh back to my point on the EAS ux potentially making it easier for people to interact with uh W with with with stuff that typically might have a bit of a learning curve maybe needs to be a developer and we can see on hash scan which is the block Explorer for hadir that token which has been created nice user interface which one of our friends Ali at the back has really been leading on so shout out to ali uh but it's a really nice new user interface and I'll stick with dark mode uh cool so that that's sort of creating a token we could also transfer that token um just by doing a similar sort of prompt but I also want to share uh I also want to share the the Lang chain uh agent which doesn't have some of the features which the Eliza agent has it's sort of less aimed for a web 3 uh social environment it doesn't have immediate Integrations to platforms like X but it does have other benefits like plugging in really well to the Lang chain Library uh Lang chain is one of the definitely the first llm framework uh or sort of tool set which which I was familiar with that it's been around for a good couple of years now and definitely on the Enterprise uh and in the web 2 world Lang chain is is really leading the way so we can also uh let's also play around with with the Lang chain agent uh i' I've cloned the repo these instructions are on this part over here test Lang chain langra agent I've cloned it mpm installed I've done my whole environment set up let's go for the test and what this does is it pulls up a uh just just a CLI chat bot where we can also hang around and start interacting with it now I'm showing you a very very superficial uh overview as to what you can do but as I sort of mentioned earlier these are the total depth of of functions that you can have uh We've also had a really really happy about this one of the uh one of you guys yesterday came along we were talking about it and they did a a pool request last night for uh support with one of our leading decentralized exchanges so big thanks to Cal who did that and I want to Echo what I said earlier if you're contributing to our sdks and want to apply for one of these AI agent uh bounties this is valid because it's uh it's it's AI uh so if that's something which you're interested in if you've got your own little projects you need some extra plugins you can sort of double up and and uh I think that would be a really great look okay back to our agent kit so we've now got the option to go autonomous mode or chat I'm just going to do chat because it's a bit bit safer uh and again we can sort of be really really quite flexible here we can say hey uh let's create another token uh create a token called uh test with the same symbol and that many initial Supply yeah we do incorrect spelling mistakes AI is pretty clever now there we go you can actually also do this in in other languages as well because again these these sort of GPT 40 models are pretty decent so you can even do this and again back to my earlier point on on user experience uh having that interface where you can actually interact with it in multiple different languages could also be a a value ad for any use cases which you're looking to build so it's now created it uh let's do some examples of airdrops I've got my uh wallet over here let's uh let's go this is a different account so I'll air drop it to this account air drop that token to to this account oh uh 29 yeah 29 send [Music] them 500 of them or send them 500 5,000 so that's done it it's obviously remembered within that sort of context uh memory that okay that token is the one which we previously created and then we can also do things I guess let's try out the uh query get H by balance query so get the balance of that account that I just dropped I just say I dropped so again it's sort of it's contextualizing it within that whole uh with that whole sort of shter memory and the account of that account is a balance of 5,000 test tokens Okay cool so it's actually done that I mean we can also specify get the H bar balance of it and that should be uh also supported balance 426 balance 426 there we go so this is really just a very high level overview um as to how you can sort of set it up obviously we do have that AI agent uh um the AI agent Bounty where if you're using one of these tools that's great don't have to use these tools you can look at stuff AI Providence but this is really exciting because we've been working pretty hard on getting these open sourced and and and out on the market ready for you people to use so I wanted to give a really quick overview just as to how you can set it up uh this Doc is almost that golden source of truth and you can set up the alliza plugin try it out on git pod then start implementing it with your relevant uh specific use case looking at where it plugs in with other functions with other actions uh and and maybe having it on X maybe you want to have it where it's it sort of creating tokens transferring tokens trading them uh I've got a few minutes left I mean I I can keep on prompting it and pushing the boundaries but I mean if anyone has any questions now is also a good time happy to answer them up here and or may maybe we can push it to the Limit you test me with some prompts to try out and we'll see how uh how robust it is but uh o over to anyone who's got any questions um this might be like pushing its functionality but I was curious if you could like extend it to do like maybe like a uni swap or something that's a really so so that sort of ties in to that pool request which one of these absolute Legends did yesterday uh he's definitely in uh in the Hira good books he's integrated it with with sorcer swap which is essentially the uh which is essentially the the the Unis swap equivalent uh but yeah you can do that I mean and again holding back to that rhetorical question which I said at the beginning can AI agents make the user experience for consumers people who who don't really have the best understanding of uh defi all of the Concepts or maybe even just sort of w wallet wallet management you could even do stuff to the extent of okay I have X amount of hbar eth whatever that token is I've got a high risk tolerance find out for me what the what a good strategy is and then you could have an AI agent which goes right okay there's a plugin to scan all of the yields of a particular decentralized exchange or potentially the borrowing and lending rates of a particular exchange and then go through that endtoend process of right you've got your H bar swap it for uh another token and then provide liquidity to a particular token pair and tell that all to the user you could do a user sort of human in the loop approach that's what Lang chains really good at Lang Chain's got some nice uh functionalities where you can sort of get the llm to get to a certain point and then it goes right stop user do you want to continue yes no that's pretty nice for some of these more riskier use cases or you can sort of go YOLO mode just do it for me um these are all things which I mean potential ideas uh for you guys building user interfaces for this is is definitely a great great use case where AI agents can can really shine so yeah hopefully that answered it I'll leave the floor open because that was a good question hopefully we can get a uh followup as good or better oh I love it love it bro ask away man ask away I just a question I love question um I was you said that it has Lang graph implementation yeah does do you have like the base flow that we would be extending off of or yes is is there is that a like you have that graph yeah yeah yeah so so let me so so Lang graph is um Lang chain started like a couple years ago it was mainly good for things like Rag and just creating these pipelines but obviously AI has developed quite a lot since then uh so they then had L grath which I believe is specifically for agents don't hold me to that but agents is defin one of the uh core offerings of it and as you can see so what we're basically doing is similar to Eliza and similar to any framework you have your deterministic function which here is using the Hira agent kit we've got this create ft method that can take these uh these these uh object sort of uh parameters some of them are required some of them are not required but all that we do is you extend the Lang chain base tool you can then create your own plug-in I mean it's so easy it's just like mindboggling you give it a qualitative description this is what llms understand and then you can use this uh we've got an example implementation in that test script so over here we are uh we're we're importing currently all of the tools that we have uh support for that's just in this helper method that returns an array of the tools you can be more selective th this may maybe maybe all of these tools if we go uh so let's go to not that one create H tools So currently these These are the tools that we have supported with Lang chain uh We've also got consensus service support it's not yet integrated here but by tomorrow it will be and you don't need to use all of these for your use case maybe you just want to have a create fungible token maybe you just want to have a create fungible token and transfer or airdrop uh and that would of course make the model a bit safer I've only got 20 seconds that will make the model a bit more likely to to execute the correct thing because there's less uh almost entropy less room for error and also slightly cheaper because that model that token price will be reduced uh okay eight seconds I'm going to show you how you do this so you've got the got the stuff over there you import the list of tools that you want and then you just use this create react agent that's coming out of langra and then you go let's go 's out hopefully they let me on for a bit longer uh over here you pass in the llm you want that can be any llm and then you pass in that tools array and that's how it has access to it plus uh uh message modifier cool that's me uh hope everyone enjoyed that and hope you're also building AI agents