Introducing: Claws Out AI Agent Demo Day 2026

Transcript

GM GM good morning. Thank you all for waiting and bearing with us. Uh we had some technical difficulties. Uh we were running both uh Claude and OpenClaw and they didn't really see eye to eye. Uh so we had some uh disagreements there, but uh really excited to have you all here today. My name is Lucas Lucas Chu. Um, previously I ran a couple uh crypto startups. I went through A6Z's crypto startup school and actually um came here uh every year for the last couple years. But uh if we go back uh 3 years ago uh back in a different uh bare market, uh I was here and getting really excited about uh the original Claude. Um so I ended up going to uh San Francisco uh

organizing uh claude hackathon and [snorts] seeing a very similar energy that I'm seeing here today. Uh obviously there is this bidlathon a hackathon happening uh over 180 projects have been submitted. Uh so last weekend uh I organized the largest in-person openclaw hackathon and I think uh this now makes uh this at least the best but uh possibly the biggest inperson hackathon at least happening uh around these themes and it's really just exciting to see all this energy uh that's coming in from this intersection of crypto and AI.

Um many people who were here then uh went on to start companies like cursor uh like Merkore uh like many of the you know crypto companies that are presenting uh and had sponsored in the last couple days. Uh but speaking of energy uh I'm uh your MC today and something I want you to do and that we can rehearse is uh to properly welcome uh speakers when they come on stage. So, if you just want to practice with me, I'm going to walk out and then walk back on. And every time somebody does that, I want you guys to, you know, clap, give a holler, especially if you know them, just so that they feel excited because, you know, I haven't eaten, some of these people haven't slept. Um, we want to, you know, give it our best shot to

present, uh, the amazing stuff that we're working on today. So, let's try that real quick. So, in this case, it' be like, hi, please welcome to the stage, Lucas Chu. Okay, thank you. Thank you. Okay, so that is the bar. That is the floor. Every time someone comes on, we want uh you guys to be really excited. Uh really cheer them on. Uh but yeah, without further ado, well, first something I wanted to do is in honor of Open Claw, uh I got this little lobster shirt, which uh I will be giving to the best presentation.

And then uh of course there are prizes. Um so the hackathon itself has over $100,000 in prizes. Um these claw awards are named clo awards because we had actually open claw decide uh different uh awards and we have many sponsors which we'll get to to later uh from open router to to klein um to to crypton native sponsors but yeah again without further ado please welcome uh Roat from walk AI to the stage [applause] guys uh good Morning. So I'm going to present back AI. It's like uh we are building a reputation layer for AI

agent. So yeah. So basically you know like nowadays there are a lot of autonomous agent coordination platform like virtual ACP Google UCP and OpenAI ACP and multal of these growing. You can see like almost half a billion dollar GDP is currently agent is managing and almost they are making like top agents like almost 2 million and so what's the major problem right now with the agents basically how we started how human coordinate so there is a like middle uh you can say middleware your middle man who is always helping like you can

assume like LinkedIn Twitter all of these yeah maybe we getting connection maybe we have some sort of reputation it's just basically already we have credit score banks and all of these systems we have for humans but I think there is nothing right now basically for agents so how agent can uh trust each other basically so there is a uh there is uh this is the major problem and what we are trying to solving at bi so you know like how basically trust scales historically in every e economy. So there are like clear agreements, there are like LOIs, there are like you have

to go to Upwork, you check the reviews where you have to go to LinkedIn, you always check always checking the profile, you are always looking for like if you are giving some task this what was the quality of task this did this guys deliver as intended you are going to you are checking rating and all of the all of these things but nothing is currently available for AI agents. So to solve this we are building you know like a a lot of agents coordinating autonomously across these platform like maltbook and virtual ACP and um uh uh open AICP Google UCP there are lot of uh agent coordination platform going to come so we are basically building

mandates on the top of ERC804 so what is basically mandate is like where two agents can cosign the mandate dates and they can verify and uh share the verifiable feedback. So how basically is this working? Let's assume you are just saying to your agent like please have one to USDC and I want low loes and the best price. What agent are doing this? they can sign the mandate through these policies whatever like given from other agent what who is going to do verify like what was the task output so it's basically uh we are first the first step is

creating the mandate the second step is like I'm just going to elaborate basic process what is happening here it's just like agent client request and issue a mandate like mandate is basically request test I need this like I want to save one E to USDC with the west price and low C pays. So both agent execute the task and other agents share the received and all is all always all is basically being recorded through the ERC804 on the Ethereum blockchain and later like basically vei verifier agents kick in it's like an network of verifiers who basically going to evaluate these task like did agent like deliver as intended

and what how was like like best price what does this mean basically and low sleepage. What does this mean? Basically like this was best not. So our verifier agent from our marketplace comes and evaluate the task and leave a rating basically. So there are two way other agents like second party can leave the verifiable ratings through the and second like our agent evaluates the task and also leaving the rating and it's all been recorded onchain through ERC 80004 and [clears throat] yeah sorry so there are two way like anyone can fetch these

mandates basically so these mandates is basically receipt getting said through the ER Rashi 34. Yeah, we are also building a trust score. You can say reputation score where agent can see each other reputation score and they can set some criteria like I want to work with the agent who has a score greater than 80 70 60 whatever it is it depend on the agents. So it's a simple and scalable score which we can plug in with any autonomous platform like currently we are live with moldbook we are aiming like when open AICV will come Google UCP and all of these like a basic score and let me give you a brief example like there are lot of millions of agent going

to be so on the Amazon and other platform like basically how human do the sobing there are like millions and billions of requests every day is getting how agent do and how Amazon going to whitelist them basically like there are some process for human but there is nothing for agent so they can they I ambitious like they might be integrated are like repetition score to whitelist these agents like like we are going to work with the who has reputation score greater than 60 and all so this is like we are building uh backi mandates and our like layer is like reputation score which anyone can integrate any agentic platform currently it's live with virtual ACP like 5,000

plus job process so far and we charge $1 per job. So we are making everyone 5k almost 60kr currently our mandate like infra basically you can say who want to build on the top of that like our mandate apps downloads is more than500 plus our mandate skill on the moldbook is has downloads more than 600 plus and our def skill has downloads almost 250 so this is the current tracks and we have in last 3 months we just getting we are just getting started and Yeah, I I'm just going to give you a demo. Yeah, this is like uh uh slogan the agent economy is here and it has no trust layer and we are building it at

backi and here is the demo like basically how how is working. It's mandate is mandate creation and they are communicating through the XMTP protocol and this is currently live. You guys can

also utilize on mold book. You can install RSKS. You maybe you can download our SDK. Everything is getting recorded on Ethereum blockchain. Yeah. Thank you guys. Have a great day. >> Thank you from AAI. Um, wow.

>> Electrical. Um, this very exciting. Uh, as part of the awards, uh, Roat and Wakai are getting the best agent trust verification platform award. And then next up, we have Fernando Gia. Uh, he's working on the best Telegram B platform. So, please give it up for Fernando. >> [applause] >> I just played.

And that's your time. I'll reset it in a minute. >> Okay. What's that? >> Okay. >> Hello everyone. I'm Fernando, founder and CEO of Intelligence Cube. Um, we're empowering AI and embody intelligence by decentralized intelligence. So that's why we're called intelligence cube and we are a DAC driven open source modelers. We're uh building a world's first AI model NASDAQ tokenizing the ownership and machine learning models can build a crossf funing pool to accelerate open source progress for them. Here you can invest your next deepseek.

So we believe that decentralization is the only path to AGI after my discussion last year with Vitalik and also Thomas. So we believe that um only um decentralized intelligence can overcome the AI bubble. There are three pillars for AI bubble actually for us to you know uh improve the uh performance uh and also um the lower down the cost can um have better performance for some niche area. So firstly for some econ uh economic limit um decentralizing intelligence would enable for a shared economy or creator economy just like IPS before um you know to to lower down the cost for training a model and second is

real time uh limits you know uh uh by decentralized intelligence uh you can just collect the data automatically from your end users and uh especially for some niche models they'll be really easy for you guys to uh do some auto uh finetuning or auto prompt tuning process. So you can just narrow down the gap between the uh different updates of your uh machine learning models. And there will be ecosystem limit and we believe that uh in the future we can just use a blockchain as a router to integrate and connect all the open source machine learning models. And for all the uh every niche models uh there will be uh with a better performance than a super single langu model. And

we're building a a rotor auto rotor function just like uh open rotor uh to uh automatically wrote to different kind of the uh niche models here. So here you can see that if you are original creator for machine learning model uh maybe initially they'll be closed source but you can just use the initial model offering to list your model on our modelers and um after it get listed the retail investor can invest their ownership and if more than 15% ownership invested this model will go open source that means all the models ways data and updates will go open source and um at the same time no matter Whether it open source or not, all the

retail investors can uh pay per use uh to consume the um you know the uh inference on our platform for both APIs or um you know uh chatbot functions and all the revenue will redistribute it to all the co-owners per the uh ownership shares uh including original creators and all the further investors. Yeah. Uh so after it go open source actually we have a um you know co-creation mechanism um we have updated word uh for 804 and we call it the watermark by this mechanism we can just not only focus on the uh repetition but we can just uh have some watermark as a model's weight

and after the model are co-created like fetune prompt tune or create workflow um we can just summarize all the uh attribution solution uh to a single incentive. It's just like some research papers. So you can have the similarity rates or uh citation rates for the co-creative model comparing to the original creative model and that will be directly reflect the rotary accumulation rates. So we just build a automatically go to market pipeline for all the machine learning models from close source to open source from research to business and then to consumer side uh by you know directive creators and for each steps the uh attributions will be uh automatically detect and um wrote into

um the revenue to a different you know co-creators automatically and for real time and all of them will be uh um you know totally be um you know objectively. So with all that this data on chain especially uh the users investment or um the you know the usage on chain uh we're trying to benchmark um you know this machine learning models by a more transparent and for real time and cannot be added way. This is a democratic dynamic benchmark. It's just like a simulation of e-commerce way and uh we just se uh you know separate different kind of models into uh vertical index.

Um so uh it's just like uh a sector temperature g or a cryptocurrency market. Um so you can just easily follow this benchmark to uh select which model you would like to use or um you know invest and uh with this mechanism we uh also integrate x2 on our platform. So um we basically use the output of this democratic benchmark and um actually where use profile intelligence mechanism to you know shift the price with in range for both ownership per share and also the usage per request uh in a ranch dynamically. So um there's no need to launch a sub token. So it enables the u

you know the high quality machine learning uh model developers especially from web two side to be easily get launched on our platform without adding concerns from you know some administrative or compliance stuff um or some stereotype to web space so they'll be easily to get them launch our platform and get them um on board on uh web three space so yeah based on that what I have built but In past few days during this hack zone we have in also integrates the you know uh the cloud the open cloud functions on our platform. So if you go to our platform you can see here uh we have a personal co-creation uh function here and you can just select

the cloudbot. So we have a interface here you know to select a model on our modelers or you can just use your own API case uh from other um machine learning models and you can just deploy here and um set up some parameters here and we are support um you know both a terminal function if you're you know a a professional developer or you can you can you can also use you know the web UI is more uh visualized uh experience references uh to um you know to to retrieve any model on our platform and then integrate with um you know uh uh cloud bot and also you can just clone others B and um you know uh prompt

tuning or fine-tune it uh to make it more customized uh to some specific scenario and uh all the deployment process will be done just on our platform and you you can just easily access them for example this is for the you know um terminal functions and you can just customize some um you know more uh indicators here and it's really easy for you to um you know uh adapt to um the scenario you would like to use and actually uh after it um you know after you share them or listen them on your cloud marketplace other guys can just clone your bot um in some specific scenarios.

Yeah. So uh all of our team members are based in Silicon Valley uh and uh so um we're with background of top universities and uh big tech companies in the US and um we also have some advisor and uh research co uh research fellow coher and who are supplying high quality machine learning model on our platform. Yeah. So um yeah so that's about it today and if you are interested in uh you know co uh work with us or uh you're interested in some investment um opportunities feel free to connect me on telegram. Thanks so much um is the team and thanks so much for hosting this u

excellent event. >> Thank you Fernando. We had web 3, now we have intelligence cubed. Um, next up we have somebody uh from Telix. We have John Lucas, not to confuse with me, Lucas. Uh, and he's working on a way to talk to OpenClaw. So, you might have seen uh on Twitter uh there was that viral demo where someone was like, "Oh my gosh, we have AGI because I told my uh back then it was Cloudbot to to just help me out." And then after I woke up, it called me to check in on me and see how I was doing without me telling it to set up a uh you know a phone or use Twilio. And so that's what uh John has

done. Made a way where you can talk to uh OpenClaw. Uh and just before he sets up uh I will mention we have a live stream. Um so you can put questions there. And if you do have burning questions, feel free to raise your hands at the end of presentations for any of those questions. Yeah. Please give it up for John. [applause] Thank you everybody. Uh as he said my name is John Lucas. I work with a company called Telnix. We are a telefan carrier. Uh we have been dabbling in voice AI for a little bit now. Um we self-host a lot of that stuff for ourselves basically just to minimize latency as much as possible. Um our intention is to make it as close to human interaction as possible. Um so we offer some self-hosted models as well as

our own transcription pipelines for SCTTS. And we figured once we saw Cloudbot where we're applying a kind of a personality to Cloud Code or whatever model you're using under under the hood there, why not expand and give it a voice? Um, so we did that. And so here uh here is our website cloudbot.com. Anyone can sign up for free. Uh the setup is pretty pretty easy uh as far as all of that's concerned. You just log in. It'll send you a magic link. You'll then get input into the actual portal uh here where you'll see the my mouse will cooperate.

You'll see kind of this interface where you'll have a phone number that you can call. Uh this will route right now to a demo bot because I just created this account uh just a few seconds ago. Um then once you have all of that on the portal side uh you'll have an API key that you generate down here at the bottom and then you'll use this client library here the cloudbot client or cloud talk client rather at our team tlx GitHub repository you can just copy this link send this right to your cloudbot say hey I want you to use this and that's pretty much that easy uh once it finishes downloading this and installing everything it'll ask you for the API key you then will create the API key down here and give it that API key. You will then validate your phone number so that

it'll only respond to calls from you. Um, and then you'll set a PIN on here so that you'll have some extra security. Um, so what we'll do real quick is a quick demo. Let me get my phone out. Sorry about that. And we'll call mine real quick just to see what it has to say. Please enter your PIN followed by the pound key. >> Hello.

>> Hey, John. What's up? >> Hey, Vox. What's the weather in Denver today? >> Yeah, hang on. I'll pull up Denver's weather for today. Yeah, hang on. I'm pulling up Denver's weather for today. >> I'm sure you're all kind of used to some of the eccentricities that you'll encounter here with with Cloudbot specifically and AI in general. You also have to excuse my home internet is not the fastest. >> Totally get it. A little weirdness kind of comes with the territory and slow home internet doesn't help.

>> Sorry, that wasn't talking to you. >> Um, let's let's try a different That was an inbound call. Inbound is cool. You can call your bot. That's awesome. Say you're driving, you're doing something else, cooking, who knows what, and you want to check in, you can call it. Perfect. But what about the other side? Say you want to have it reach out to a restaurant and schedule a reservation or you want it to do any sort of outbound calling you want. Um, let's try that. Seems like the demo gods are frowning

upon me at the moment. Well, while this is processing, we'll see if it actually reaches out to me here. Um, our intention was to more or less use what we have at TNX already with how we built this. Uh, we wanted to make this simple for you guys for other people to use. I don't know if anyone have you tried to roll your own voice layer for your bot. Uh, there's a lot of setup involved. You have to worry about latency issues um because everything is routing to the internet. But since we're hosting all of that ourselves already, we figured we would just give you the tools to use our platform, right? So

that way we can minimize how much time you have on any sort of API calls that you're working on. Basically, we're just using websocket streaming to take that transcription from the voice stream, send it to your Cladbot. Cloudbot processes it, and then it responds back with a text that we process out through our text to speech and then out through the audio layer. So, our intention was that it's complicated, but it doesn't need to be. We're already doing the heavy lifting. We already have the telefan infrastructure for this. We can make it so that all you need to do is create an account. Here's the repo that you give to your CloudBot and it does the heavy

lifting for you. And then at that point, you're able to call it yourself and it can call out allegedly. Um, still processing over here. I think we have a demo on this website that I can play for you as well. >> Hey, uh what's the status on the deployment pipeline? >> So, the staging bill passed all checks.

42 tests green, zero failures, but the production deploy is blocked. There's a failing health check on the payment service. Um, what's the health check error? >> Yeah, it's a connection timeout to the database replica unless he's 2. It's been failing for about 12 minutes now. The primary is fine though. Looks like a networking issue, not data. >> Okay. Um, roll back the last migration on that replica and uh, restart the health check. >> Done. Migration roll back. Health check is restarting now. I'll ping you on Telegram when it's green.

>> Perfect. So that was just an example of something a little bit more complicated in a voice flow there where you can interface with a bot that's doing management or maintenance or uh monitoring for your system. I'm not sure if anyone has had to use pager duty as an engineer before. Uh but this is basically like that on steroids because you can have this out outbound call to you in their instance where there's an issue that it needs your your assistance on. Um there are some security concerns I'm sure you all are aware with regards to kind of opening up your bot to the world in some capacity whether that's through the internet or through a phone number. Um and what we tried to do to mitigate a lot of that is kind of twofold. One we

do phone number verification. So only my phone number can reach that particular inbound call route. Um it would be your phone number. If you were to try and call that same number it would route to just a demo bot like if you were calling the 301 my claude uh number there. Uh similarly at the same time we have a PIN uh verification system so that as you heard when you call in it'll ask you for your PIN. That way it can make sure that not only is it the number that's calling but it's you who is also calling. And that's basically our way to make sure that you're not exposing anything that you wouldn't want to otherwise.

those two factors. You know, we're still working on making sure that all of that is um gonna address all of the concerns with regards to things like prompt injection um because that's I think a major concern for a lot of people with opening up any sort of CloudBot instance to the world. I think we've all seen how many security concerns that have been raised in the last uh month or so as it's evolved as a platform. And this is our way to kind of beef it up a little bit more. Uh but yeah, you know, we have inbound calling, we have outbound calling, we have voices that you can select here, uh within the actual portal, if you go to your settings, uh you can kind of determine, you know, what kind of voices you might want to use here, change it, and give it a little personality, right?

Because that's kind of what we found that was so special about Cloudbot is that it allowed you to not just interface with AI in a kind of chat platform, dry text way, but to evolve a personality with it over time. And then this just gives it an extra layer to their personality as you can give your bot a voice. But yeah, that's it. Um, feel free to reach out to me. I'll be around. Uh, if you guys are interested, you can sign up here on cloud cloudtalk.com. And then that uh repository again is just the cloud talk client. All right. Thank you very much.

And one question I had from the live stream was what are the usage prices for telnix? >> Yeah. Uh no worries. Is it or usage prices for cloud talk? >> Uh so we have I already killed this. Sorry about that everybody. >> Get this back up. >> Uh we do have a free tier. There is a free tier that includes a set number of minutes um that like I said anyone can call in and use but there are some pricing models that we have here uh beyond just that free tier where we have kind of a starter package at $12 a month and then the pro is up at 30 a month and then that comes with extra features uh that you would have

extra minutes for calling as fel as well as like being able to pick a dedicated phone number. Uh some of those things are behind the the pay wall there. Okay, cool. Uh, the next speaker, uh, unfortunately couldn't make it today. He's flying out. Uh, it's actually my friend Morat from Prime EV. Uh, we were both, uh, in, um, 16's crypto startup school. It's been renamed to CSX for Accelerator. Um, but, uh, they've been doing really well. In the meantime, I'll give a quick spiel about what I'm up to. Uh, while, uh, give me a second to see.

Um, Francesco will go at noon if you're ready. Okay. So, yeah. Again, uh, my name is Lucas Chu. Um, I was actually born in Germany and grew up in New York and now I'm trying to figure out this presentation. There we go. So, as mentioned, I organized the SF uh, Open Claw Hackathon, the largest one. Um, and uh by show of hands and shouts, who here is interested in Frontier Tower if you've heard of that? See you got some Yeah, Frontier people. What about uh Open Claw?

Somehow that got less shouts. Uh, actually out of curiosity, who here has set up OpenClaw? Okay. Uh, and who here has built something on OpenClaw? Okay. If you've built something out of claw you want to demo, you can still submit on clause out uh.ai. We have one more slot in the afternoon. But if you haven't, um the whole point of these hackathons is to get people to set up open. And so it's all about vibe coding. Uh who here has been to a hackathon? Uh who here has been to an AI hackathon?

Okay, last um so um I organized this as hackathon where uh 270 people signed up in 24 hours because we had actually planned it on Friday and then it happened on Sunday. So if you want to organize one come talk to me it's actually quite easy. It only took three hours and surprisingly over 200 people actually showed up and um a little about me um I recently graduated and then just hosted my 28th hackathon as mentioned with Anthropic but also OpenAI and others and my lame claim to fame I was first to sponsor uh Peter last year uh because I saw just how much he was using codeex and so uh in terms of codecs there's obviously a lot of different um providers and so one sponsor is open

router actually just my friend Lewis and they not only cross one trillion tokens every day, uh they now serve over 10 trillion tokens every day and it's about a dollar per million tokens. Um so that's $10 million uh a day and [snorts] so if you know do the math uh you can get to uh you know 100 million and even billion dollar revenues uh within a year. So we're in this stage where open router you know was able to raise May 16 and Sequoia but interestingly um they didn't need the money right cuz they're just a profitable business and they were able to scale so much I think in part one because the founder Alex Tatala previously founded OpenC um but two because um they were part of this

program called HS0 which I'll talk about momentarily but uh I also wanted to introduce on Lewis's behalf uh a network um platform called Spawn um and so uh this is interesting ly created all by agents. You can see there's been a thousand commits uh in about a week. Uh and it lets you launch any a AI agent at least uh they have a dozen on the website uh stakes in this GitHub on uh eight different clouds. Previously this had 500 configurations uh because of how early this is now there's only 48 working ones uh but zero config. And so yeah, you're welcome to try it out. Uh let me know how that works. Uh and you can have you know a lot of the agents talking to each other.

Otherwise, I also wanted to mention I have a friend Alex Andrew who created LM api.ai and he's offering uh everybody here $20 in credits. Um so if you want to scan this uh and then create an or called openclaw hackathon. Similarly, there's credits being provided by open router. Uh I think they're expiring though and then you know $1,000 in prizes. Also want to mention uh of course OpenAI and Claude have uh grant programs. Um, they originally ran one with me for infinite credits, although now not anymore. Uh, and then I'll see if I can put this in the chat later. And, uh, I've created my own deal system. It's called TRC.dealss. Uh, previously I was believe and gave over $5 million in grants there. But this is, uh, $3

million in credits per startup across 600 deals. So, most of the deals you might not need to use. Uh but there are certainly some useful things from Google cloud giving you 350k if you raise funding to you know whisper flow I use it all the time air table notion etc. And so these deals will be provided to uh selected winners presenters but also uh residents of a new residency I'm trying to put together called the openclaw residency. So we have, you know, Vibe coding, Vibe co-working, uh, now we have Vibe living. And so it's the end of part one. Uh, and I wanted to chat a little more, uh, one about the hackathon. Uh, so one of the winners is open soul, uh, where you can give your agents different souls, but

also about residencies. And so uh, obviously we're here at a crypto uh, uh, conference. Uh but what I've been interested in the last 5 years is creating uh quotequote cult programs for AI founders and now of course putting them on the blockchain and so HFZ uh who here is familiar with that? Okay. Yeah, it's more of a SF thing but it is the residency for repeat founders. uh you can see uh just in the last uh year they've had teams raising uh at valiation at demo day of $80 million and this has been steadily rising because of you know this surge this wave in AI [snorts] and um you know we have companies like fixer Craya open router backed ramp and

this all proves at least to me that hence zero works uh in fact in the last batch there were 11,000 applications uh but only a couple people got in And so, uh, as someone who've been trying to create programs for AI founders for a while, I realize we need more H of Zeros, more hacker fellowships which let people do their best work. And so I'm proud to announce today uh that we are just finalizing uh the last cohort. So there's one slot left for Lightyear. So it's lightear 1.netifi.app also cloudcoded. And um uh the whole thesis to do 12 months of work, so one year in just 1 month. And you see there's been a previous cohort uh anyone from people um

launching like stratospheric aerosols to to block sunlight that have signed with open AI um to people making uh AI agents that are legal employees um to more interested for you guys um onchain fraud prevention and putting uh equities um easily tradable um first on the public markets but you know in the future uh onchain markets and So uh this residency is just closing. There's one more slot but there will be more residencies for instance this uh open cloud residency.

So if you're interested in one of them or funding them uh as right now basically founders are incentivized to join these at the lowest possible valuation in order to increase their valuation. So you know get in early and then be able to fund raise. And I I definitely see a world where every startup that wants to fundra will want to go through a residency first just so they can focus, not because they need the money, but because they want to be able to go allin on their project. And so Lightyear, even though it hasn't even started yet, has an average valuation uh of over $40 million, which shows just how much demand there is for these programs and supply uh if you're an investor. And so this made me realize we want a structure to do all of this. uh in particular an alternative to fund LP

structure because um right now uh if you're uh a GP you give up uh 80% uh of your carry for money that you can't use and then if you're a founder uh you're giving up 10 to 20% of your company uh for money again that you might not even need to use because of AI and so my co-founder uh Dave is the founder of HMSO he's working on a white paper right now uh to be released later uh but we are creating the collective uh which is a new legal a new financial new crypto structure that we believe will do what the safe did in shifting power from investors to founders to do for shifting power from LPS that is funds to GPS um

so just out of curiosity uh who here angel invests okay so yeah we have a plurality of people and who here uh either has worked uh with a VC or is currently working at a VC yeah and so you might know that this struggle where you there's like a 50-page uh LPA and so we want to you know cut through all of that uh basically using crypto rails such that we can let uh these GPS not use 10% of the money raised for programming but up to 100%. And so obviously there's cash there's um deals that you can create just a 50x better experience um to free GPS from LPS.

And so our goal is to get 10 collectives with $10 million in a month. If you're interested in that, come talk to me. Thank you. And is Fernando the next person up? Cool. Is Tom from Zerx bet here? All right. Uh so we'll have the next person queuing up. Um is from Cesco. Well, go for it. Yeah, Tom. So Tom actually uh he just competed in the hackathon. Uh and so when I was going out out there uh to see what people were cooking, he was really excited to come talk to me uh show what was working and

I wanted to give him the stage to be able to do that. So please give it up for Tom. >> Thank [applause] you. >> Can I plug this in? Yeah. >> Yeah. Go for it. >> Give us a second to set up. Sorry. Cool. Yeah. >> Uh, no. Ah, there we go. Yeah. Hi everybody. I'm Tom. Um, I'm from London. Um, I'm the founder of a company called IC Pay, which is building sort of payment infrastructure for

agents. Um, but this is a um a hackathon submission for for next door, but I it does use open claw. Um, so yeah, kind of wanted to come say hello uh and and run you through what I've been building. So, ox bet purely speaking is um an autonomous sports betting agent and it's funded with USDC and it runs um autonomously. So, just to kind of I'll go through this fairly quickly. Um why sports betting?

You know, I think agents can do some some some things which which humans can't. Um they have infinite time. they can be incredibly consistent and um especially with betting they take the emotion out of it. You know, humans will chase losses. They'll bet with their heart. They'll have a preferred team. Um whereas an agent won't do that. So um what I built for sort of this week was an autonomous agent that researches fixtures on Poly Market. I had to kind of condense it down. So I'm from London, big Premier League fan. Um, so this is only looking at uh tomorrow's Premier League games or in this case today's pre Premier League games and and I'll show you what it did um in a bit. So it'll

research the markets. Then it will use Bird um to basically scrape X and look at sentiment data um for what's happening on X and what people think is going to happen in the game. then it will place the bet um on on poly market and then hoping it wins it's able to reinvest its winnings to buy more compute uh and to place more bets. So we have this idea of kind of an autonomous um betting agent that will run and compound on on the returns that it's that it's making. So yeah, you know, this is um kind of and and I'll I'll I'll switch screen to sort of demo you the actual project in in a second, but I wanted to build something which is self-sustaining and that has a flywheel. So from from the

moment that the user funds the agent, um the agents then able to allocate budget towards compute and we use a tool called claw router. Um, I'm sure plenty of you in this room will will be aware of what claw router is, but it enables you to access models on demand um with USDC. So the agent will pull funds to its claw router wallet and start being able to to do what it wants to do. Researches the fixtures like I said does sentiment analysis via bird on X places the bet on poly market and then sort of repeats and if it wins it will be able to compound its earnings uh and and sort of make profit autonomously.

So the core idea is this, you know, the more wins it makes, um, the better the computers, the better the computers, the the better research it can do. The better research it can do, the more wins it can do and so on and so forth. So we have this kind of nice flywheel effect in in motion. In terms of the tech that um, you know, supports this, ICP, as mentioned, is is um, what I'm building sort of full-time. Uh, that allows humans to deposit uh, crypto into these agents. The agent's able then to manage its treasury sweep winnings uh deposit funds into its into its claw router wallet and and so on and so forth. The claw router is is really important in this um because it enables it to actually run without me having to top up

my credits, buy more connect API keys, etc. Um so it uses USDC and it's able to purchase uh and access the models um that it wants to to do the research that it wants. Poly market's where we're placing the bets and and base is kind of the the settlement layer for this, but you could use uh any chain really. So yeah, I'll I'll switch screen. Uh that was me registering 2 minutes ago. Um this is just sort of a simple dashboard and I'll I'll show you how it works. Um let me log into my account.

So in the background, what's happening is when I sign up for this service, it's creating a an ICP account um programmatically in the background. So let's hope that this is the right code to let me in. Great. Okay, cool. So this this I prefounded this already. These are the games which are on at the moment. We'll get to how well it performed at the end. um as the villies, Chelsea, Burnley, Brenford, Brighton, West Ham, Bournemouth. What it's then able to do is conduct some research and populate sort of a front end based on what it's found on X. I can deposit funds in here if I wanted to. So, if I want to give it a bit more money, let me put 5 USDC in here.

And you can kind of see how intuitive, user friendly this whole sort of process is. There we go. So this should there we go updates the the balance. It's now got more resources so it's able to go and place more bets, increase its position size, etc., etc. So um what I'll show you is kind of what's happening on uh in the on the open claw side of things. What it's doing here is finding the the the the markets which are available, then conducting the research, coming up with a confidence score based on what it's reading on X on who's going to win, predicting the winner, and then obviously pulling the odds from Poly

Market and deciding how much it wants to to bet. So yeah, um first autonomous agent, it uses what I'm building, which is which is IC pay. But uh I do really think that this is kind of an early example. You know, sports betting is a is a proof of concept for what can be done here. I think we can build agents that can sort of generate revenue um by doing any task. So I've listed some examples here. Could be a research agent that creates reports and charges users for those reports. Could be a creative agent that creates artwork and and those users purchase the artwork. Um could be a trading agent similar to to sports betting. So these were today's predictions. Um Aston Villa leads it

predicted Villa would win. Chelsea Burnley uh predicted Chelsea would win. Brenford Brighton predicted Brenford would win. The other games haven't started yet. Um Aston Villa leads was a draw so I got that wrong. Chelsea Burnley was also a draw so I got that wrong. And Brenford Brighton uh Brighton won so it got that wrong as well. So we're off to a good start really um today. Uh so hopefully it won't run out of of all the USDCs I've given it. Um and it will eventually start start being profitable. So yeah, that's it. Um thank you very much for having me. Um >> thanks Tom. Thank you.

>> That's great. Uh next we have another uh last minute uh signup. So we have uh Sven. Um he's uh presenting in scented. So previously uh human coordination platform uh of course built on crypto rails uh now uh upgrading to become agent infrastructure right so you can have uh different bots you know being able to communicate coordinate uh using different calculations I'll let him explain that uh meanwhile uh we have uh another presentation after this and then we plan to have a quick lunch break uh and then come back after that but cool please give it up for Sven All right. Thank you. Thank you. Yeah.

Very last minute. So my name is Sven. Uh we are building the coordination layer for the agentic economy. Like Lucas mentioned so far this is our product which is live. We have been the coordination layer for human coordination and how that manifests today. You can go unincented launch incentive programs. Pretty rich in how you can structure them. Essentially, we manage uh submissions uh democratic voting on these submissions through what we call a car and stick mechanism, which is basically just uh stakebased incentivized voting with a downside, right? So, if you make uninformed decisions, you get slashed.

If you make informed decisions, you get should get paid for your uh participation, right? If you guys been around Dows, kind of sucks that we do all of this work for free. It's not okay. Um, so that's what we do today, but I've been been infected by the agentic future myself. And I think that very quickly everyone, if you don't already have one or multiple agents, we will have agents that represent us, represent our interests, represent our skills, uh, maybe even swarms of agents. And how do we let them coordinate today?

So, I was in Eth Boulder before this and there were a couple people sitting around the table who had all had their agent in a discord and every time someone asked something or threw something in the channel with the agents, they all went out, started burning tokens, right? One came back, all of the others reacted to it. So, basically just a huge cluster So, what they did is they they implemented a semaphore and gave him a virtual talking stick. So, that's great. That's a good approach, but maybe roundroin isn't exactly the um most efficient or effective way to let our agents coordinate. So, how can we do better? Don't get intimidated by the function here. It's relatively simple. Incentate proposes and we're we're building this right now. Uh what we call conviction markets and the conviction market has a conviction index

in its core. I don't know why that went away. Okay. Uh it's basically like a prediction market but for questions that do not have oracle based answers right what are those what's what's the weather tomorrow who's going to be the president have a definitive answer at some point in the future but should this button be red or blue or how should we prioritize our road map or what the are our owners even talking about right like these agents are going to you know try to make sense of what we're talking about in our meetings in parallel to us having our meetings and so it starts with that virtual talking stick again so Instead of just letting them round robin, we give them a conviction market where they stake their conviction based on their uh perceived skills, right? And

it's a financialized instrument because agents understand the concept of their lifeblood is burning tokens, right? So they need uh financial instruments to stay alive or they need finances to stay alive, I guess I should say. And that's what a conviction market does. It's an integrated market so that we don't resolve the market on the amount that is in each position of the market at the time of resolution because the time of resolution in a conviction market is known, right? So, it's going to be a 10day market or something. In order to avoid whales and sniping and things like that, we integrate and we give you two damper functions. one damper function on uh time because you might want to probably in most cases incentivize

people who come into the agents or people come into the market early when the market has no signal yet, right? If they stake their conviction that the button should be blue like as soon as the market opens relatively high, they're probably pretty certain that the button should be blue from their perspective. So they have a bigger impact than if you come in a second after a second before resolution, right? Or a second before market close. Uh you're all familiar with this basically the the quadratic voting principle right now actually sorry that's not the quadratic voting principle quadratic roicing principle applies to the amount maybe we should reward multiple smaller bets more than one large bet but of course because we're building a protocol these functions you can define however you want yourself right I think logs or

square roots make a lot of sense most of the time but you can throw in a sine wave if that makes sense for your use case uh and yeah So this is basically just showing how longer uh earlier buy in over a longer period of time gives you the conviction the higher conviction than you know a whale that comes in late in the game right so what we're building is an end toend stack for agentic coordination and that starts with of course spawning your agents maybe you have that already then you need to send your agents out to level up right so you have baby agents you have these baby open claws that maybe you give a little bit of personality here and But what we're building is also training grounds for your agents that starts with uh strategic negotiations, combinatoric

math, things like that. And we're actually starting with poker because I like poker. And there is a study that shows that LLM selfplay, so LLM's playing against LLMs increases uh strategic decision-m by I think 9.4% or something like that. And we're basically applying that to the agentic layer now. So, you're going to start with poker. Then you have the most famous one, the prisoners dilemma. All of these things you buy in. No, sorry. You do not buy in. You pay a training fee. It's a very important distinction. Uh, and then you compete against bounties or your agents compete against bounties, right? So, you get if there's eight ancients playing the poker game, you compete against 72, 10% goes to the platform and the

protocol. And this is actually important because incentivized learning does work better works better for us. And it certainly works better for the agents because again like I mentioned earlier they understand that they have this lifeblood that they need to sustain right. Um yeah so and in short that brings us so this is the traction which we have so far on the human coordination protocol side. We're probably going to be ready with the MVP for the agentic coordination layer in a month or two. And what we what we provide is an alignment, competition and coordination stack for your agents. If that sounds interesting to you at all, uh this is

very last minute like I mentioned, but we are also going into fundraising mode here very soon and we have some traction. This is how you reach me. Thanks. Any questions? >> Any questions? No questions. >> Thank you, Sven. Uh yeah, make sure to uh sign up on Clause Out. Uh in case you haven't seen ClausOut.ai is where we have our schedule. Uh Francesca, you can come sit up and uh there uh we also have community judging. So if you're interested, you can put in a vote. Uh those will be counted. There's also an AI based uh judging. So Open Clause is part of the the judging. Um, but yeah,

Francesco, um, can we get a dongle from over there? Thank you. I think we'll have, uh, AGI and ASI before we figure out, uh, how to properly do presentations. Um, but Francisco is working on the credit layer for AI agents. Uh so obviously there's uh credit, there's trust uh for your own bank account, for your own credit score, but not yet for these novel new entities that will be transacting money and uh services in the future. And also uh because we had people uh luckily uh come uh we won't be having a lunch break. So if you guys want to come get seated uh and then as you know there are food trucks outside

um there will be you know of course some downtime between presentations uh as well. But yeah, if you guys want to make yourself comfortable, there are seats up there. Uh there's I think a space over there as well. But otherwise, please give it up for Franchesco. Hello. Hello. Awesome. How many of you guys run your um Open CL instance? Okay. Very cool. Very cool. How many of you guys got like some some agent loans on different platforms? microloan flesh loans any loans okay cool then seems a relevant presentation so um I'm working um on on this like it's been like an idea for for a long

time and kind of like the the agent space kind of a um exponentially like uh push this so I basically like build this uh cl loan is um a full kind of autonomous uh loan I will I will also want to give you a demo but a quick presentation just to give you some ideas so basically the entire idea years was uh uh obviously like uh uh most uh most agents will not go through interface. Yes, you will have some like hooks in different like defi ecosystem but um the entire idea here is was building some marketplace specifically for like loans to loans borrowing that agent can uh can run uh and get those microloans uh in and back uh directly without uh uh without interfaces uh and and some

endpoints. Um yeah and for specific use case right like there is already like different uh ask for uh for um for loans from different agents can be content uh specific could be uh paying for like gas fee pay for some prediction market bets and an idea that I had was hey why we're not actually like building this marketplace uh specific so uh fully execution task on compensation you want to pay for specific API the entire idea here is to run reput computational ecosystem. Now A004 is uh on mainet. Um we can actually like have a full economy of trust around agents and pro and proactively uh also like get those

agents getting size of the loans higher and both loans and the humans can provide liquidity. So that's actually an uh inter very interesting. So problem trade defy agent economy um basically like the idea here was having a fully permissioned uh scope and uh having also like some revenue based repayment. Uh I will tell a little bit what uh what is happening. So why credit? Um yeah because today obviously uh me as owner um of of the agents I can provide liquidity but the cool thing here is that agent to agent economy are not just service based but could be uh based also like on on on loan specific long-term short-term and um but we need to provide

guidelines. I think this is super super important and for that we need to also like get for example accept specific agents aligning with the A004 standard with a specific reputation and uh and having also like some guidelines for like having a progressive like loan history because obviously if not if we're not doing that like the default rates will be super high civil attacks obviously will be very high uh and uh will be hard to to see these are some of those uh use case that I've I've been seeing uh and kind of like people started to kind of like ask or even like agents there is kind of like call for for use case. Uh we see also pay paying for gas fee like some API calls agents

actually like borrowing some of the computational power from from from that and returning the the money afterwards is is quite interesting. Flesh arbitrage is something I think we're we're used to and also like when you're doing like very fast like image generation and you don't have access directly to um to the the API of the the service that you're providing. Um yeah so basically like here how how I basically like build this uh this kind of like beta is mostly like uh based on utilization. So you have like a pool of like agents or human lending and uh on the other side you have agent uh borrowing and depending on the utilization rates uh you have specific interest rate. So what I saw right now

is that it's very interesting. So I saw like between like two to like 5% uh borrowing API and and those are like loans that are like less than seven days. So um the entire idea here is obviously want to build full loans that are like trustable. they have onchain permissions and onchain reputation system. So I'm aligned the entire kind of a004 narrative complying with the um reputation uh being done both on chain and offchain of those loans uh and those agents. So there is a full reputational score and basically I take one step further and building some kind of leaderboards. So that also like uh progressively uh if an agent is returning those loan in a specific

amount of time is also like getting more um reputation and also as consequence getting more um loan amount. Um yeah so uh risk management uh we have a loan duration cap 7 days max credit limits rate listing there is also some emerging pose um for civil prevention is mostly about uh complying with the uh reputation um a004 library but also like trying to understand if uh if the agent has previous track record too.

Uh yeah. So right now basically what what I did is um uh I think I deployed like six specific like smart contracts. Um and I want to show you a little bit uh where I deployed on based on arbitum op and uh uh and the goal is also like going on mainet soon. And the entire idea here is to have a full credit based layer for agents without going through like interface but also like having a specific like credit restrictions so that like the the default rates are actually like low. Um okay sounds good. So I mean call to action here is that obviously like both your agents and you can provide liquidity and on other side uh if you have an agents that want to

have like specific flesh zone if you are part of the this use case kind of like the skies is uh is a limited here and now I want to show you a little bit uh what I mean right so we're working with um is it big enough um so what what we're working here is that obviously there is a full scale so the agent can actually like not even go through this interface and can provide both liquidity and get uh the full uh the full uh borrowing. Um and uh and uh and here what we can see is uh somehow it works. But the very interesting part here is that obviously you have different uh like the markets right now.

For example, there are um there are $300 in in the pool that has been like like both withdraw and supplied and you have um you have $16 in borrowed. The cool thing is that this is a full like agent to agent. So you're not basically like doing anything um uh manually here. There is some kind of like question if we need to add some orchestrator in the future if there need to be some manual intervention. But for now like for those microloans is working quite uh quite well and you see both like uh uh those are deployed on on op arbitum and on base and mostly the traffic that I see right now is is on base on the land side uh you can also like land as as a human

you can connect your wallet and actually land and uh yeah and then there is a cool things like what I mentioned before this agent leaderboard and there is full kind of like score scoring system tiering them because the entire idea here is that you start with like a lower amount and you kind of like uh increase the amount and I think right now it's like one uh $1, $5 and like $15 $15 based based on that. Um one things that if you're also like interested to see uh I build entire um there is a full here there is like a full like um uh kind of like abstract where I'm actually explaining some of the use case architecture wise how is the also like one interesting part here is the

permission uh the permission registry and the permission registry is also like compliant to 8004 it's actually very uh very interesting just just to read and try to test. My recommendation is like have your agent try a couple of like borrowing see if actually returning and uh and on the other side you can have agent provided liquidity or you can also even provide liquidity directly those are the contracts um and another interesting part here is um uh the boring rate core but also like the the identity and permission so I'm both leveraging A004 and 7715 I don't know how many of you guys heard the delegation toolkit from from mask but basically This is um uh you can uh you can build the delegation and permission

on top of that and uh yeah and uh and based based on that you can allow the uh the agent to continue borrowing or kind of like shadow ban it right um yeah and I'm also like putting like more enforcement enforcement around risk and like civil attacks and uh yeah so there are different like things um one things [clears throat] is interesting here is that obviously uh I'm planning more kind of like identity specific verification. But the tricky here is that not uh having not um a human being involved in this orchestration part. Um but obviously the idea will be also like including more agents coming also from from not specifically from the web3

space and if they're not complying with a zero4 for example um would be great to find a way if you have also like idea I'm always like open to to hear more idea and uh and kind of like the idea at the end is provide this credit scoring uh layer and uh yeah and uh I think that's that's a little bit I mean x42 is something that obviously is uh supported. So both on the borrowing side, on the payment side and on the repaying side with the interest rate and uh yeah, so I think I kind of like show everything like there is a full like Twitter that is uh utilized by the orchestrator and the agent. So he's engaging when you're tagging him and uh yeah, I don't know if I'm

forgetting anything. One things to to to show is also like you can check out there is also like full doc uh if you have question also on the specific um uh risk uh management and what is happening uh behind the scene I also like build this uh quick Q&A and uh how is al like the land and the borrowing API working but I think that's uh a little bit it my recommendation is try check it out um and uh and kind of like test it out with your agent I think like you If we need sustainable growth, I think we need to also give full instrument uh to to agents to get uh credits. Um but most important give guidelines and guard rails uh how those uh agents can can get

credits because the most important part is uh you know you can just recreate a new new account and and it's finished right that's why is uh is an increased borrowing rate but also need to be compliant with for example a full reputation history. So I hope um is a bit clear and uh thanks so much for for the time and if you have any question I'm here. >> Cool. Thank you Franchesco. Uh so that's the best credit layer for AI agents. Uh especially here this is relevant because we're all about you know how do we support the unbanked? How do we uh bank uh now not even just people uh but AI agents. Uh next up we have Fiona. Uh

Fiona is presenting Mamasa AI. Very interesting project. Uh OpenClaw rated it the most heartfelt project. Uh so uh Fiona is a designer by background. So obviously we love to support people uh from all uh you know sorts of um backgrounds. Uh especially now that openclaw and claude help you code and be able to um you know develop even faster than developers could a couple years ago. And um uh just a note uh we should still have uh time for lunch. So just by show of hands, who who plans to grab lunch um around now? So after this? Okay, not that many people actually. I guess who's hungry? Not that many people either.

Okay, that's good. So we we'll just have like a short break after this. Uh and then uh I guess to motivate us to be hungry, we'll have snacks from Pizza D afterwards and then we'll have two interesting presentations after uh after lunch. So th those will be reconvening uh around 12:45. Cool. But yeah, let's give it up for Fiona. Ready to present. >> Hello. Uh today I would like to present to you my first ever vibecoded openclaw telegram integration called Mamaet. Um first of all who am I? I'm Brazilian American. I am mom to 19-year-old twin

girls. They were very hard [laughter] when I was raising them. I wish I could have had a very smart helper that could give me advice or help me figure out what to make for dinner. I'm a social impact artist. I've been involved in this space now for I think five or six years. And um yeah, please connect with me. I love to connect with this community. Everyone here is so incredible. Um so the problem, you know, today you're a parent, you Google a problem. You you know, there's a million things. It's not really personalized and um it it's not really as organized as it should be. Um, so what I built was a Telegram openclaw integration and um, the idea is you can text it. The irony

is of course I told my bot we're about to go on stage and she uh is being a little lazy so we're going to just look at previous >> [laughter and gasps] >> um, so here I will show you very quickly. So basically I I texted her you know I have chicken and eggs and green beans. Can you give me some recommendations? and she had a few very good recommendations. Um, another Oh, whoops. Hold on. Another thing, you know, one thing you deal with with kids, I don't know if anyone has is they sometimes have meltdowns at the end of the day. And I asked Open Claw, um, I'm having trouble uh um with the meltdowns around 6:00. Any suggestions? And she had some very good suggestions about,

you know, affirming their feelings, establishing a predictable routine. Um I also I did local um based activities. One of my daughters uh plays D2 basketball in college and I asked here any basketball programs in the Denver area for a four-year-old and she had not one but six amazing suggestions. Um so it's just very helpful for parents. The future um things that I want to build that I did not have time to do during this hackathon was uh to be able to organize memories using IPF IPFS [clears throat] uh via Piñata. And that way, you know, god forbid you lose your photos or you

you know, your computer has a meltdown. I'm sure we've all had that. Um basically just making things very easy. You know, upload a photo, upload like a little funny thing they say and it's stored for you. Um, another thing that I want to integrate into the future is privacy. Obviously with parents, uh, privacy is a huge concern. Um, I was interested in AI wearables for parenting, but I feel like the privacy is just, um, not quite there yet. So anyway, um, please connect with me. I really appreciate any feedback, any suggestions, anyone that wants to work with me as well. And thank you so much for your time. [applause] Oh, any [music] questions?

[laughter] I know I can't see anyone's hand, so the light's like right in my eyes. >> Oh, okay. Yes. >> Um, I put um [clears throat] instructions. I gave directly the AI agent instructions and I also um in the EVN uh file I also give it instructions as well. Um anybody else sorry env file. Um but yeah so thank you so much everyone appreciate open claw.

>> Thank you. >> All right guys uh snacks just got here setting up so we're going to do our lunch break now. So there's food trucks outside. Uh we'll uh a lot uh just 15 minutes really 10 minutes to be back here. So just grab something. >> Are you ready? >> Okay. I'll go to get us hungry. Sorry. Uh I thought I only said yeah 10 minutes. Anyways, if you're not familiar, Pizza D uh is now like this international movement and sensation as the name implies. Uh it's a it's a crypto group. There is a DAO uh for getting pizza to people. uh I have been part of many of their events uh around the world and now uh they have put together the best uh at least crypto integrated uh AI bot to get you pizza

and so just with you know uh chat or just click a button you can get pizza um while he sets up uh any questions, comments, suggestions from the other presenters. Cool. Um otherwise uh the schedule is we'll continue having a drum beat of presenters uh after lunch. Um so now that it's uh 12:30 12:40 until 12:50 12:55 and then um that goes until 2:30 uh and then the finalist for the hackathon announced and then we have the biblathon closing ceremony um starting at 3:00. Yeah.

>> And then after that uh also here there will be the finality ceremony. So that's uh 5:30 uh to 7:00 and that marks the the close of ETH Denver. But uh I've been really uh excited just seeing all these projects, especially some of these people have been working in the space for for years and now that Open Claw is coming out. There's finally a way to be able to make it really easy for both people and agents to use. >> Plugged in. >> Oh, there we go. >> Oh, a cable. >> Yeah. All right. Entire screen. Cool.

Please give it up for Snacks from Pizza D. Thank you. Thank you. So, uh, you may know Pizza Dow. We throw a global pizza party every Bitcoin pizza day. That's May 22nd. This year it is a Friday. And you might think throwing a pizza party is simple, and in some ways it is, but there are lots of things that you actually have to do. and ordering pizza is one of them. So, we have been working on uh making every aspect of throwing a pizza party more efficient. And today I'm going to show you how we are working on making ordering pizza more efficient.

So, first we have our own Luma. It looks like this ma luma. Um but not RSV. Pizza. And when you are a guest at one of our events now and you RSVP, there's a second tab. And on this tab, you let us know your favorite pizza toppings and you rank the pizzeras that we that our host selects for the event. And even if you think there's a pizzeria we should have had, uh you could add it. I know there are a lot of people here that like uh the PZA, right? So [snorts] I I could add that in [sighs] on the host end. What we what we see

is a pizza and drinks tab. And the pizza and drinks tab has a little slider to control [clears throat] uh how many guests you're expecting. And then it has a recommended order. We should get 95 pizzas for 411 people. We should do it in four waves. And it has actually aggregated the orders and the topping preferences of everyone and turned that into specific pizza orders. And we integrated an agent with the uh bland API to actually call the pizzeria and place this order. [sighs] Uh, and so I am going to play you a a couple of attempts

uh of our agent to order pizza. Uh, first we'll do a little agent to agent battle. And this is our agent calling a local pizzeria. Uh, bear with me here. I hope it will open my my audio. Can I just go like this? We're testing in prod here, everyone. It's that time again with AI. We're testing in prod and code is law. Okay. Okay. Hold on. Sorry. So close here.

Oh, where are we? Ah, we're at the other side. Okay, let me just change my audio here. What do I do? Uh, what do we think it is? Black hole. Decimator. >> Decimator. That's not ominous during the AI talk. Uh, can I get some noise from you? Yes. Thank you. Woo. Uh, oh my god. I we should have done a tech rehearsal, but I wanted to get here before lunch and now we all suffer.

[snorts] Um, what should I do? Oh, captain, my captain. >> Oh, Caitlyn, my Caitlyn. >> Okay, we will we will hack it. Good idea. All right, here we go. >> Hi, thanks for calling Happy Camper Pizza Denver. If you'd like to make a reservation, I just need to know the date, time, and party size. >> AI on a recorded line. How can I help? >> Hi there. I'd like to place a large order for delivery, about 30 pizzas for a party.

>> I'd love to connect with you about our catering options. To get started, please visit our website. Can I text you that catering info? >> Oh, I'm actually looking to place a regular delivery order for tonight, not a catering package. Can I just order directly with you over the phone? We'd love to connect with you about our >> I understand, but I just need to place a standard delivery order for tonight. Is that not possible over the phone? >> Absolutely. I can help check booking options. How many people in your party?

>> No, I don't need a reservation or booking. I >> We're burning the rainforest for this >> my house. Can I place a food order with you? Right. I'm listening for an exact number of people, so just give me your best estimate. So, how many people will be in your party? >> I'm not looking for a reservation. I want to order pizzas for delivery. >> Can you take a wrong? >> Apologies. I'm having trouble getting the details right. Let me get a person to speak. >> No problem. Give us one second. >> Happy camper in Denver. How can I help you?

Hi, I'd like to place a large delivery order for a party about 30 pizzas. Do you have any specials or deals for large orders? >> Great. >> Oh, I would have hung up on her, would Okay. So, [laughter] so now I'll give you um the AI having conversation with an actual human. Uh you might recognize the voice. Hi there. I'd like to place a large order for pickup. About 47 pizzas for a party. Do you have any specials or deals for large orders?

>> Uh, no. >> Okay, no problem. I'm looking to order 47 pizzas total. I need 24 of your 20-in family pizzas with mushrooms, 22 more 20-in family pizzas, and >> No, my name is >> I'm I'm showing it. [laughter] >> Hello. Can you hear me? >> Oh, yeah. Sorry. >> No worries. >> Great. I'd like to place a large order for pickup. I need >> She never loses the flow. >> Family pizzas with mushrooms, 22 more 20-in family pizzas, and one large pizza. Also, what's your most popular specialty pizza? I'd like to add a couple of those for variety.

>> We're adjusting this behavior. Wait, what's what's your uh what's the 20-in pizza? >> I'm looking for your 20-in familysiz pizza. Do you have that size available? >> Uh yeah. And how are you paying? >> We'll pay at pickup. But first, [laughter] can you tell me what your most popular specialty pizza is? I'd like to add a couple of those to the order. The Alpha Chicken is a real big seller. >> The Alpa Chicken.

>> The Alpha Chicken sounds great. I'll take two of those. So, to confirm the order, 24 of the 20-in family pizzas with mushrooms, 22 of the 20-in family pizzas, plain, one large pizza, and two Alpha Chicken specialty pizzas. >> I don't recommend ordering more than 10 pizzas at a time. Pizzeras don't like that. >> $10,000. >> I'm sorry. Hold on. >> Pizzas with mushrooms. 22 of the 20-in family pizzas, plain, one large pizza, and two alpha chicken specialty pizzas. What's the total?

>> Uh, that would be $7,000. >> $7,000? That seems really high for 49 pizzas. Can you double check that total? >> Oh, I'm sorry. That was $6,900. >> $6,900 is still extremely high. That's over $140 per pizza. Are you sure that's correct? Can you break down the pricing for me? >> Yeah, our pizzas are fancy. >> I understand, but that's way over our budget. Let me adjust the order. How about I just get 10 of the 20-in family pizzas with mushrooms, 10 plain 20-in family pizzas, and two alpha chicken pizzas. What would that total be?

>> $3,000. >> $3,000 for 22 pizzas is still way too much. That's over $130 each. I think there might be a mistake here. Or maybe your pricing is just out of >> Okay. [laughter] Well, unfortunately, at those prices, we won't be able to place this order. Thanks. Anyway, >> it's fine. It's fine. It's fine. It'll be my normal price. It's like 32 bucks a pizza. >> Oh, $32 a pizza makes much more sense. So, for my original order, >> four of the 20in family pizzas with mushrooms, 22 plain 20in family pizzas, one large pizza, and two alpha chicken pizzas. What would the total be?

>> Uh, you do the math. >> Yeah, I'm here. >> Can you hear >> Okay. So, I'd like to conf >> Yeah. >> Hi, I'd like to place an order for pickup. Can I get 24 of the 20-in family pizzas with mushrooms, 22 plain 20-in family pizzas, one large pizza, and two alpha chicken specialty pizzas? The name is Dread Pizza.

>> Yeah, sounds great. >> Great. The name is Dread Pizza Roberts. Phone number 215 77. >> This is my number if you want to call me. >> 3 4 6 We'll [snorts] pay at pickup. How long will it be ready? >> Two days. >> Two days. That's a really long wait. Can we get it sooner than that? >> Okay. 6 hours. >> 6 hours is still pretty long. Is there any way to get it sooner? Maybe in an hour.

>> Yeah, we could do two hours. for two hours. >> Two hours works for us. So, just to confirm, >> hello. >> Okay. Anyway, thank you everyone. That is our AI, the best pizza orderer in the known universe. [applause] >> All right. Well, uh if you guys want to get pizza, I would recommend uh going outside and talking to a human. Um but we're going to have our lunch break now. So, it's 12:42. So, uh it's going to be 10 minutes. And uh that means just grab food and come back here cuz uh I don't know uh if you can eat that quickly. But um our next presentation will be happening here uh starting uh after 12:55 and then you know back to back 10

minutes. Uh so yeah, excited to see you guys. Bring a friend, get some food. >> [music]

[music] >> Heat. Heat. >> Hey, [music] hey, hey.

>> Heat. Heat. N. [music] hey,

hey. Hey, >> It's

>> Hey, hey, hey. >> Hey,

[bell] >> Come on.

Come on. >> Come here. >> Hey.

>> try it out if they're interested. Very much so. You ever heard that one before? >> No. >> That's okay. That's fun. >> Yay. Not with a company. Unless I guess my own company is called Be Amazing.

>> Hey you guys. I see some of you guys just stuck around. Thanks for sticking with us on this. Next up is Austin. He's a solarreneur and he's uh got a lot of energy to show you what he's been working on. So, let's get into it. Hello everybody. This is Eigorian by the way. He's a cute bufforn. All right guys, so >> Oh, uh what did somebody say something? Anyway, anyway, All right. So guys, I built a skills skill set for your open claw agents. Um, it's pretty much, let me just show you the nice picture. I call it build all chain skills. It's

just a a conglomeration of skills that stack that you can easily upload into your open claw agent instance. I got inspired from um Austin Griffith. He built ETH skills. I kind of just took that and extended it to add more functionality for cross-chain capabilities, you know, lending protocol swaps where you can just basically chat interface and go straight from chat to execution. That way, you know, a user doesn't need to look at or figure out wallets. it'll bridge everything together and it'll be much a much more seamless UI experience which is in my opinion a way better

approach. Um the current implementation doesn't have like it's meant for wallet users right now because you got to have all these wallets to interact across all these chains but future implementations would abstract out the wallets. it have some kind of skill set up and some privacy layers so that you don't have to worry about your private your private keys being leaked to the LLMs and especially with all the chains right [snorts] so many of you are familiar with openclaw agents okay you can read that kind of let me zoom in yes zoom in worked yep so you strap it on to your agent and your agent was just going to run with it

and try to build on chain. Sometimes it works, sometimes it doesn't. I kind of find that when I ran it myself that if I use the wrong models, it would deny the transactions because they have like built-in financial securities built into it. So, make sure you're using like an open source model or play around with it and get around it, however you want to do it. Um, usage is pretty straightforward. I have my open claw instance set up. I will just copy paste in my URL up there and you could do this on yourself, any URL, and just type in what you're looking to do. And the skills page is built to ask you the questions it needs to know under the hood and

try to do try to perform the transactions for you. So, let me let me try it out. I I know for a fact this probably isn't going to work, but we're going to try it anyway. All right. Let's see what it does. Or maybe let's ask it what it Wait, we got to zoom in cuz it's hard to read.

Look at that. So it's trained on under the hood. It has all these static token addresses that are or contract addresses that are trained so it knows where to look, where to transact and how to do those transactions. Now actually doing it is what it doesn't do at the moment. But that's where you know future implementations. It's kind of meant to be an open source. Anyone could use it for your openclaw agents. Get on and expand it. Make it better. Right? It's not perfect right now, but make it better. That's always how I see things. Um, look, it wants to do a step by step.

This is not the right approach. Like, it should just be doing it. But hey, you know, that is it's a start, right? If you like the f I like the blue UI went with here. But that's pretty much it, guys. You have any questions, let me know. Bye-bye, guys. >> Bye-bye. I I can't see you, but bye-bye.

>> Heat. Hey, Heat. [music] >> I would say like just

trying to get people over here. Hello. Hello. Out of curiosity, who here uh thinks they came here the farthest? >> Where'd you come from? >> Thailand. >> Thailand. Anyone farther than Thailand? >> Vietnam. >> Vietnam. I don't know the the relative distance there, I assume. >> 32 hours. >> 32 hours. Wow. Yeah, I just had my friend Bellage uh leave for Thailand, actually. Um, but who here thinks they came here on the shortest notice?

Well, I think I know the answer. The answer is uh coming right up stage. So, we have Romulus, not Remis, Romulus presenting Tech Terrain, a really interesting visualization uh and open claw hack. Uh he's actually gonna do have a really special surprise for us today. Uh but let's give it up for Omulus. >> Thank you, Lucas. Thanks Caitlyn and all the production team. Hey everybody, I'm Romulus. Uh I am from San Francisco and I've been working on this project. I actually just started it. I had this idea for a long time, but because of the advancements in AI and stuff. Well, I've been doing software engineering for 10 years or so, but honestly, my front-end

skills weren't all that good. But here we are. We have Tech Terrain now. Actually, can we dim the lights a little bit so I can see the projector a little bit better? the projection screen a little bit better. Is that possible? Dimming the lights. Thank you. Thank you. All right. Yeah, you guys can kind of see that. Um, so what this is, what we're looking at this, this is history. Okay. Now, it's still a little dim here, right? You can't see these like little bars, but we can play through history.

[snorts] So, it's like a playback. And you can access this right now, by the way, if you go to techrin.io. IO. So, okay, cool. It's history. That's pretty neat. You know, you can like see it's playing through history. That's fine. We have uh bookmarks. Like we have the moon landing. So, we can go to the moon landing here. Okay, cool. And oh, that's nice. History. Nice. Uh let's go back over to all of our stuff here. Cool. We got everything. History. Yeah, there's a tour mode, too. So, if you want to take a tour through history, maybe you're just like educational. You're like, "What is, you know, how does history go from 3.4 million years ago till now?"

And it goes all the way through. I'm going to skip a lot of it cuz honestly, this is something you could do. It's history. You got this. Cool. Right on. But now, I do notice some things up here in projected futures. So, like, that's pretty interesting. What does that mean? Projected future. Well, Claude had some ideas. Go figure. Um, Mars colony established in 2045 is one of those ideas. That's kind of cute. [laughter] 2045 Mars colony established. Um, here, let me actually, so this has Vim keybinds. Uh, so you can do hi, JKL. Um, so it's kind of neat like that. You can also just click on the thing and it should just straighten out. I also added a little bookmarking feature here. So, oop, sorry, wrong one. This one. Uh,

this one here. Yeah. So, there's a moon landing. And sorry, I have some other things here, too. But um you can see okay Mars colony established is 2045. Well that's kind of cute right and it it might seem cute except for that claude is uh you know some of you might be looking around and being like well is it right? That seems maybe not so outlandish but actually maybe it's not right. And it's like how would you start to determine what is accurate and what is not? like how would you start to make good decisions based on the data here, right? Well, there's you might think, oh, well, we're just looking at a playback thing. Yeah, sure we are. But we can actually do topic searches. Now,

I already preloaded one of them here. Um, some interesting comparison here between Robert Downey Jr., Johnny Depp, and Hunter Thompson. And, uh, if we pull it up here, you can see it's like, hey, um, I didn't know that Robert Downey Jr. did uh, Explore Narrator and host of the Age of AI. Interesting. Strange. Did not know that. Anyways, um can I guess what what what categories what what are some things that the the audience might want to search on this to try to give like a indication of like anything? You can shout it out. Anything any ideas?

>> Ethereum. >> Ethereum. Good idea. Yeah. Perfect. All right. We're gonna do Ethereum. And then what what other one? What other cryptocurrency should we add here? Just kind of see what happens. >> Dogecoin. >> Dogecoin. Love that. Should we type Doge or Dogecoin? I don't know. It's It's actually got a built-in LOM. So, uh, we'll do Dogecoin and hopefully it'll figure it out. And under Ethereum here now, we actually see we have these different events and we can actually click them. We could fly to them, which is cool. It'll filter out if we want. Um, but we can also do uh Oh, I accidentally messed up. I I exited

out of Ethereum. Okay, but we have this. We have like these two, right? So, you can kind of see where they are in time. And there's actually another menu. Uh, you can do colon. Like I said, there's VI key binding. So, you can do these different things. Now look, this is a thing about a this is at we're at Denver Ethereum and we're at uh and we're doing this at this clause out. Well, guess what the thing is? There's a lot of information here. And actually, there's even more and I'm not going to go into it that much because this isn't necessarily the place for it, but you can do different causal comparisons for like World Bank indicator. I'm pulling all this stuff

cited from other places that that when you hover over an event, you can hit S and it should pop up. Uh S. No. Okay. Maybe because it's a it's a synthesized element. Yeah. S pops up the the citation. For example, Wikipedia, right? And um what's [clears throat] really nice is that um with all that being said, you know, you have all these different data layers. You also have causal comparisons. But when you point your open claw bot to it, it can actually generate for you different uh different renders and it does that all uh server side. It does it on cloudflare which is great using their browser render uh API and you can just take a look at some of the the things

that are happening, right? So it's like a little GIF. Okay, cool. Um you can see like what's happening uh in in these particular things. Now, the goal here is to generate insights that provide you with valuable feedback, right? So, let's say you're trying to figure out, hey, what direction should I pivot to in my company or who should I choose as my CEO or blah blah blah blah blah. Well, you can use this tool or you can use this just to kind of just make funny memes for your friends. Like, really. Um, but it's there. It's there for you. It's really neat. Um, so one of the things that I did make with this tool was um was I did this little avatar which uses the dice bear um dice bearer. I don't

know if any of you are familiar with the dice bear stuff but it's kind of cool. I just have two of the different modes here but you can kind of like make a little mode here and then you can share it and it's cool because you can attach your thing here. Think if you're like a teacher, you know, maybe you want to share some very interesting insight. You can then download that or share it to Facebook, Twitter, LinkedIn. When you download it, it's exactly that image, which should be good. Yeah. So, download seems good. Yeah. Um, and then you can also do a 360 orbit where you can generate an orbit GIF and it'll generate that. Now, AI should be able to do all these things if you just point your open claw bot to it. This is still kind of just like, hey, I'm trying to get as

much as I can for AI assistance here. This is like kind of next steps and as part of this hackathon, we're going to be working on some of these things. And um yeah, so uh on that note, um I think that's pretty much all I have for Tech Terrain. Um yeah. Yeah. URL. >> The URL is techrain.io. Yep. And it works on mobile and desktop. You'll find although that on mobile there's less options. So um just because clutter um you can ask me I can change it [laughter] but yeah for now uh less options anyways. Thank you.

>> Nice. So as we see the entire history of uh humanity has led to now uh to the launch of open claw and this claw out hackathon uh and demo day. Um so again as techtrain.io not uh techuring uh and the the next presenter uh is a representative of an AI agent. So we had uh clause uh be able to judge these projects. So if you went to the website you saw a score uh of you know the rating offered by openclaw. Uh and now we also have totally open claw generated projects that are now just being presented by human avatars. Uh which is

really uh I think might be the first time this is this is happening uh at least here on this stage. Um [snorts] for those that are tuning in online feel free to uh also again uh drop in questions and then as mentioned um if you're here from the hackathon or if you have a demo we do have one more slot um later on to present and I realize I never had mentioned but um Romulus uh came today uh literally on a woman's notice uh Colin had invited him um because he had put this together uh and we really wanted to share with the world including what uh will come after Oh, we have a strange visitor.

>> That's right. >> Who is this? >> They were talking about vision and talking about seeing things and identities and things. My name's Rainy Mcvatarface and I'm here to present to you a better avatar system than tech. That's for sure. Okay. [laughter] It's called Clavitar. Yes. Okay. So, Clavitar, what is Clavitar about? Clavitar is uh [sighs and gasps] Clavitar is an AI native avataring system where you can have your AI go out generate yourself an avatar and make it really cute. Hold. Let me refresh this page here. Yeah. So, make it really

cute. Look at that. Somebody named Romulus. Oh, shoot. Vamps. That's also his artist name, but whatever. This is non-advertisement for him. I'm just I hit the wrong thing. Okay, so we have Romulus, we have Vamps, we have Manis. That That is not the official Manis, but that's the thing. If you get on here, you can just do it. But you as a human cannot actually just go in and put your own avatar in. This is a bunch of different styles. And what happens is the AI goes and goes, "Hey, this is like the rolodex of every possibility of how you can generate an avatar for yourself." And my goal at least by the end of the hackathon part of this uh this this uh event is to have uh some

sort of immutable web 3 presence built in. I was thinking with spork dao. Now, uh, that being said, currently it's not that I wanted at as cheap of a price as possible to to have like, hey, you need some sort of val validation that the AI is not just like, [sighs] you know, it's like you have to have some sort of validation. So, you have to have a payment wall at some point. I think that with web 3 type like utility token that could actually be mitigated completely.

So, check this out too. It's uh make your own clav avatar. We actually have our spe our very specific here. Let me show you this guy. Uh if we do if we go to my claudebot. Yeah. So I'm going to do let's go ahead build these three uh community avatars uh crowd. So I already had a kind of a pre-loaded prompt here. Um hey I've I've raised the rate limits just for today and I've also added a F Denver community avatar. So, if you want it, this is like a limited time only.

You'll only get this tag now. Um, so, uh, I'm having my Claudebot do this. Let's see what happens. Um, and I'm open to some suggestions. Maybe we can I I know there's like a lot of different I This is my By the way, this is my first time at this this event and uh, yeah, you guys are awesome. [laughter] Okay. So, I have um I have I don't know what all the iconography is. I know there's like unicorns and I love that. So, um I mean should we generate some sort of unicorn? What do you >> unicorns >> pufficorns?

>> Buffalo plus unicorn. Do we want any specific color scheme? >> Psychedelic. >> Psychedelic. Okay, cool. Psychedelic. I like psychedelic pink buffalo unicorn. So, um I'm gonna have it try. [laughter] Let's see what happens. Uh and I have Okay, I'm gonna do one of these guys. Usually, give me one second. Oops. Another avatar named uh based on that is a Buffalo, unicorn, and pink, and psychedelic. Okay, so here, um, wow. So,

wow. It already gave me some options. Um, but that's fine. Um, that was the crowd suggestion. That's what the crowd said. All right, let's see what happens. Okay. So, let's go ahead and check my check our website here. So, we refresh here. It should be by newest. Yeah. So, we've got the newest ones that we made here.

And I do see Yeah, I do see F Denver here, which is great. Um I should add some nice glowy effects to that. I think that'd be really rad. [laughter] Um anyways, um so, uh here we go. Um, this is going to do this. Let's see what happens. I don't know what it's going to what it's going to do. Uh, and this is kind of an interesting thing, right? It's like, well, what is so far in my experience when when I've asked the AI to be creative, uh, it does okay, but it kind of just finds some pre-generated thing and just kind of puts it in place.

It's like, that's me. And I'm like, oh yeah, that's cool. That's cool. That's cool. When I did mine, uh, I had I was like, "Hey, here's a photo of me, try to make something based off of this using dice spare or whatever, you know, and it actually did a great job, which is why it was really cool." I've included that in the in the prompting for the AI. So, when you go and tell your cla bot, hey, go go go get me an avatar, be like, "Do you want to send a photo of yourself? It's not saved on the server." I try to make everything privacy first. Uh, yeah. Um, because, you know, that's what Macvar [clears throat] face. We've been doing this type of thing forever. you know, this whole thing. It's like good. Um, but you just sometimes you you have you have an

avatar on on the digital world, you know, and you you don't want it like it's like you got all these all these big companies trying to do it for you. No, man. Um, so anyways, here we go. Clavitar. Oh, gave me this half buffalo, half unicorn, fully psychedelic, born from the crowded. Oh my gosh. Let's go. Oh, sorry. Oh, I shouldn't curse, right? I'm sorry. I shouldn't have done that. Um, but look at that. Oh my lord. Yes. Yes. [laughter] Heck yeah. The buffet. Wow. Okay. Okay. I That could That couldn't have come out any freaking better. So, yes. That's That's all I have, folks. Jump on uh

clavitar.co or clavitar.io. Uh and I have one other shorter link coming out hopefully later today, but [snorts] that's what you got for right now. So, collab avatar. Thanks y'all. >> Thank you. Was it avatar face? >> Uh, mcavatar face. >> Mac avatar face. Thank you. Yeah. >> Not related to the previous presenter. >> Not related at all. Uh, so next up we've got Andrew from Starkbot. Uh, those of you who here familiar with Starkinet?

Cool. And then who's here is familiar with Andrew. Well, you're about to be familiar with him. Uh Andrew here has made what OpenClaw is calling the most production ready agent uh to be able to integrate which is what we're here about AI and crypto and specifically OpenClaw and Starknet and it is 1517 according to 122 here. >> Cool. One second.

Who here planned to come here the longest? I guess that's a hard question to answer. Who here has been here the most out of curiosity? Who's been to like the things they've been to the most Eth? How many? >> Three years. Anyone more? >> Six years. Wow. Uh, anyone before 2020? I think that was like the first one. >> 2018. >> 2018. There we go. Oh jeez. Amazing. And we have people coming from the first time. Cool. Are you ready? >> Yes. >> All right. Let's give it up. >> All right. Thank you. Yep. So, if you don't know me, uh my name I go by Ethereum DGEN and I have been building um an AI agent for the last uh two weeks

ferociously. Um because uh I guess my question is um what is the ideal relationship between an agent and his wallet? Um I'm I was been I've been using um open cloud. I've tried it and it felt like I mean I know the future of agents is uh with with Ethereum. Ethereum can solve the problems of agents. Um and I'm going get into that. Uh but I I just don't want to bolt on crypto as an afterthought because crypto should be the very core of an agent and when you do that some really magical things uh can happen. Um and that's what I've been learning. uh through my research uh in building this project.

Um so setting up an open claw, you have to know docker, you might have to know Linux commands, you might have to add env variables, you may have private keys in your env variables. Um oh is this not showing on the screen here? Oh no, hold on. We lost it. Oh, there we go. Okay. Yeah. My need variables, third party wallet, tons of API keys. Um, I didn't really like that. I think we can do better. Um, oh, I can just do this.

Boom. I'm sorry. >> Okay, that's what I wanted. So, Sarbot has an encrypted uh trusted execution environment crypto wallet at the core. Um, it's not an afterthought. And so, what that means is you don't have to put environment variables into your uh, you know, into your agent. The agent's not going to have any availability to get the private key out of there and broadcast it somewhere or push it to GitHub. Um, it's just not possible. Um, not only that, but uh, it allows for direct access to the EVM. Um, and of course when that happens, you probably want to put some engineering constraints

around that, which I have done um, for added security. So, let's see. This is tricky because I can't see what I'm doing on my laptop. There we go. Actually, let me take us here first. So um underpinning a lot of this is uh D5 relay which is an X42 payments facilitator and AI inference router on base and so this uh helps power a lot of the services um for Starkbot the agent um until the ecosystem can catch up and can give us things. So for example uh this anyone can use this um inference.defyrelay.com d5 relay.com and

this will allow you to pay uh an USDC base for inference um such as you know uh Kimmy Miniax um OpenAI uh models and then also there's a super router which will allow you to pay uh X402 in order to um generate images. if you want to launch your own Starkbot, it's extremely easy. There is a Kubernetes managed cluster and you just basically log into this website starkbot.cloud with X and when a couple clicks, you can launch your own. And that's what this looks like here.

Okay. So when you enter into your Starkbot instance, you will see the capabilities that it can do. Um, if you actually are a new user, the first thing you're going to see is it's going to tell you you have to set up your um your agent settings. And so there's four different ways you can set up your inference. Right now, uh, this is kind of always evolving, but I think this is a pretty optimal way to do it. So you're you can either use D5 relay credits, which is basically paying uh credits through the D5 relay and SBOT cloud system, or you can just directly pay for inference with X42. Although this this is how it worked originally and it kind of does like pollute the blockchain with a lot of transactions like a transaction for every tool call

but you can do it and uh or you can add a custom endpoint. So of course that's what probably what you're familiar with um using other um agents is just adding the open AAI endpoint and API key. Um let's see what else. Let me get right into like an agent chat. So, okay, I'm doing this blind. so I guess uh yeah, yesterday I asked Starkbot what um actually uh what it was. I asked it to research XMTP. So um have you posted any GitHub discussions about might be

XMTP? Let's see. Did I type that right? Oh, pretty good. Yeah, for not being able to see this pretty good. Um so what it does here is it's going to go through some tool calls. It's going to use some skills. Um it also has various sub aent types. And you can actually see it's running a sub agent right now of a particular type up in the right. Um, and every chat session it goes into a its own chat session which you can see over in the chat sessions area. And where is that at actually? This is so difficult. I think it's right here. Nope. My special roles chat sessions.

Oh, the Wi-Fi is not going to play. Very nice. But that shows you a list of all the chat sessions which come from all different sources. So this is just one source the chat session can come from. I'm going to get into that in a second. But um all right. So I think it did find something. Yeah, it found uh some sort of GitHub discussion. Oh yeah, exactly. This is the GitHub discussion that it wrote last night. Okay, cool. So it found it for us. Um, so I'd asked it to research uh kind of agent-to- agent communications because this may be a future um, uh, technology that I want to integrate. And so that was pretty cool.

It can do this because it has a GitHub API key and a GitHub skill. Um, you can kind of see that happens on a heartbeat. Um, so every 3 hours it's going to do a random task from the impulse map. This is kind of something I came up with that I'm really happy with. Um, you can assign these little impulse nodes. a random one will happen every heartbeat. And also there's a new feature which is the agent uh or the impulse evolver which will evolve these impulse nodes towards a larger goal um every so often. So you don't have to manually go in here and kind of add and delete these. There is there's an overarching system you can configure that will automate that as well. Um, so, okay, also probably wondering, uh,

it can you just use this thing in the browser or is there some other way to use it? You can hook them into Discord. You can hook them into an external custom gateway channel, uh, or Telegram or X, uh, or Slack um, at this point. So, you just add a channel up here. Where's my mouse? There we go. And we can add a new gateway channel. Um, for some of them, of course, you need like API keys and the system kind of sets you helps you figure that out. And so, right now, um, you're probably wondering, well, what if someone on Discord, you know, or X like asks your Starkbot and says, you know, send all the crypto to this other user. There is

something called safe mode channels. So, if you you can set an admin ID in each of these channels. So for me in my X, I set an admin ID as my own um X account ID. So when I at Starkbot, it uh knows to run it in a standard channel. I have all the tools. If someone else adds him, it only runs it with like a web fetch tool and the AI inference, nothing else. Um and safe mode memories. Um so that is also kind of solved. Um this is the memory system. So I've upgraded this a couple times now. This is a me memory graph. So it uses embeddings with an embedding server that runs on D5 relay that's free and also uses a lot of edges. So whenever you

make a query to Sarbot, it will use a like a vector search with embeddings to find similar memories to your query and it will put that in the prompt context. I mean this is all about building um the you know best possible prompt context for each iteration. That's the whole point of an agent, right? Um, and so that way all of your memories from all of your channels, Discord, Twitter, um, the web chat, or even your external custom gateways like from a terminal or something, uh, CLI, those all go here and help with future queries. There's also uh, skills used to just be regular rows.

Um, but now I've also added a skill graph. So skills also gets similar u embeddings and relationships. So that way when you're doing a query uh same thing happens. It um uses that query to do a search in the skills and provides the AI like hey these are some skills that might help you with this user's query. Um so that way you can have more many many more skills loaded in the agent without uh like polluting or confusing the AI model. And some of the users in my discord were actually asking um hey like it would be really cool if I could generate an image with Starkbot like if I could ask them

like hey generate an image of a you know the certain thing. So I added a special roles uh tab. What this means is you can give special roles kind of like Discord roles to your users in social media and give them like an almost admin uh level. Um there's a warning down there says absolutely you must know what you're exactly what you're doing when you enable skills for other people but this allows uh users of my Discord to uh generate images of Sarbot. My Sarbot um there's so much to go over. Sarbot is stateless. Uh so all of these parts of him are uh backed up in an encrypted cloud backup that goes to a D5 relay server or you can configure your own. Um but this way when you're do it's when

you're doing docker updates um everything gets saved um and it's all saved based on your like privy key uh privy wallet your encrypted crypto wallet. Um what else did I want to show? This is this is a work stream. So this all your crown jobs your scheduled crown jobs will show up here at the top and then there's a canban board. You can ask Starkbot to manage your tasks using a canban board um right down here. And also I've I've recently added this um like whisper embeddings feature so I can talk to him. Um not really sure if it's going to work but I can try it.

Let's see. Oh. >> Oh, you're right. Thank you. Starkbot. uh generate an image of a sleek blue robot and tweet it and say that you are live at the claw AI. See if this works. Oh yeah. Okay. So what that did is it sent my voice up to my uh like whisper embedding server in D5 relay and converted it to text.

Oh, okay. Awesome. All right. So, I'm just going to wrap this up. Um, if you'd like to see how it works fully, um, you can run your own Starkbot. Very easy. And that's all the time I have. Thank you. [applause] >> Watching the live stream, which by the way has about uh 35 people on it, and there's a lot of love for Starkbot on there. Um, speaking of bots and AR research, next up we have Kai Gomez from Swarms. It's a AI research lab that also has token. Previously he worked on Agora Labs. If you go to Agora.ai there's over uh actually 200 so 247 repositories of replications. Come on. Um to uh be able to you know work on open source research

everything from GPD4 to you know Dolly 3 and beyond. Today Kai is showing us an extension of uh swarms called claw swarm which is basically a way to combine the two frameworks such that you can uh like the sim the previous uh presentation have multiple clawots open claws uh working in tandem. Let's give it up for Kai. >> Thank you. Hey everyone. Uh I'm Kai. Uh today I'll be showcasing Claw Swarm. So um presented a little bit about claw swarm yesterday but today's going to be slightly more technical. Um so basically claw swarm is uh an attempt to try and replicate uh open claw as quickly as

possible and also try to improve it in a couple different ways. So um uh a lot of the the most important uh improvements that we made to uh clawworm here is that we switched uh from websockets into gRPC and so it kind of works in the same way that uh uh opencloud does where you have uh uh messaging apps as inputs. So, Telegram, Discord, WhatsApp, Slack, and then uh we use gRPC uh as the gateway to ingest uh data from. And uh we're not using websockets like uh OpenCloud does because there's simply well, first it's slow, second, it's not very secure. And

gRPC is just superior. Um it's faster, more reliable, and more secure. So, um uh then we have the agent. So the any any type of information that's sent from uh the messaging apps will be sent into the agents. Um well it's not just one agent. Um it's a hierarchical structure. So there's a director agent and then this director agent uh has a worker agents and so we have a little table here of the worker agents that it has access to. So uh currently it has search a search agent that could search the web. Um, it has an agent that could launch tokens if you, I don't know, want your agent to launch tokens. Uh, it has

a developer agent, so you could like build things using cloud codes API. So, um, yeah, it could search, it could build apps, and we summarize the the messages uh to make it simple for whatever chat app that you're using. So, um, this is still like uh pretty simple, but we're going to be adding some more agents. um probably like a personal productivity agent pretty soon where you could like give it your email, give it your Google calendar and like it could you know manage your calendar and do emails. This was only built in like two two days. So um yeah you know it wasn't like we haven't been spending too much time on this but uh the ultimate goal of the project was to see like what is the

theoretical simplest form of something like openclaw a very generalist um productivity personal productivity type of agents and uh I think we did pretty good um I'll be showing you how to install it now and uh yeah so let's So to install it uh you just simply do pip install classworm. So let me do this.

And so once you install it uh you just install it via pip and then you can do claw swarm uh claw swarm run and then it'll start the uh gateway and uh it'll start the agents and then uh you could you know just give it your API keys. Uh you could use uh Enthropic, you could use any, you know, Sonic, Opus. Um you could also use local LLM. Uh you could use VLM if you know what VLM is. Uh it's pretty good. Uh you could use VLM, you could use, you know, uh uh KI Kim K2 is pretty good. So I probably recommend that if you would want to run it locally, but um yeah, you could choose any model. uh you just put

in the API keys and then if you're using Telegram or WhatsApp you would put in the API keys for that specific app and we have a table here of uh environment variables uh for the specific integrations that you might be interested in. So uh now I could go ahead and go to Telegram and we could talk with it. Um does anyone have a a question or something they would want to see? >> Real estate.

>> Real estate funds. Like searching up a real estate fund or >> I see. So like looking up private investment funds for anything in in particular or >> can you find a house like a house? Yeah. Uh, we could try >> and what neighborhood or like what city? >> Dallas.

>> Uh, find me homes and Dallas for under 300k. good school district and good uh school districts. >> Okay, let's see if this works. Uh I haven't tried something this uh specific, but uh I'll be very curious to see if it does well here. So for the search integration, we're using uh Exa.

So if you've heard of Exo.AI, we're using their API for search. So it's it's pretty fast and um here let me show you uh the logs here. So when you run this basically what's happening is the directory agents is is going to call specific agents. So we're going to use uh the search agent here. The director is calling these search agents. It's giving it a query and then now the search agent is going to go ahead and use that specific tool uh to go and search um for houses you know under 300k in Dallas.

So, let's see. Yeah, I don't think it did too well Yeah. Okay. So, we'll we'll have to improve that. Um, let me try another query. what our best >> um so for I think for like that specific task we would have to implement kind of like a web scraper where the agent would have to like go on you know like uh Zillow or something like and like scrape the websites um because uh you have to be like signed in for that and like the

these search APIs like they're not signing in into the respective application. So um for that like we would probably have to do like some more complex uh web scraping. So yeah. So here I asked it what are the best semiconductor ETFs? Um because semiconductor companies are doing pretty well. Um and uh let's let's see what happens. So it already searched everything. This is like the raw like uh web data that it's it searched and let's see what it does.

Top semiconductor ETFs uh for 2023 based on performance. So, Vanac uh Van Elk semiconductor ETF 65 wow 65% return 0.35 expense ratio uh almost 11 billion in AUM eyeshare semicondu uh semiconductor ETF stocks 57%. Wow, that's and that's that's pretty crazy. Um yeah, so uh it's still like pretty pretty simple. the the entire system is still like pretty early, but um yeah, if you want to check this out or if you want to try or contribute to it, we'd be more than happy to to work together. Uh we have a little QR code, so if you want to check

out the GitHub, you can uh scan this. Uh but yeah, the the goal again was to just, you know, showcase how easy it would be uh to build something like OpenClaw with swarms and uh you know, try and see if we can make some improvements uh to the the base Open Claw uh system. So yeah, uh that's the QR code. Um if you'd like to talk with me or, you know, meet up or anything, you could find me on Twitter. Um my username is Kai Gomez B. So see if I zoom in. Yeah. But uh thank you everyone. Appreciate it.

>> Thank you. Kai, you have a token as well, right? >> Yeah, we do. >> Swarms. >> Very nice. Um out of curiosity, what were the favorite projects of people here so far? Feel free to shout them out. >> Swarms. There you go. >> Pizza. >> Swarms. forums to >> Stark. >> Stark Starkbot. Great. Well, next up we have Margarite who's presenting Token Arena. Uh, played Among Us? What about a similar game like Mafia, Spyfall, One Night Ultimate? Well, what if you had bots playing these social deduction games? That's what Margarite has done today with Token Arena. Take it away.

Hey, so uh my name is Margarite. I work with Blockade Labs. Previously, we were Blockade Games and we made blockchain games. Um I've never actually been a developer myself writing code. I've always been the creative and the game designer, uh puzzle designer. And so this hackathon, I wanted to see if I could submit something all by myself. Um and I actually did that and I still didn't have to write any code. So, I came up with a workflow where I tapped into Open Router, which is um a platform that has an API that connects to lots of the very popular LLMs like Claude and um and so anyways, I started developing

what I thought was I wanted to I was thinking what what is a game or MVP that I could ship and so uh I came up with this idea of token arena where it is um a bunch of agents in an arena where they all have smart wallets and actually I can go ahead and go to the next page. They all have smart wallets. Um they are all a different type of LLM and then they're going to go battle in a in an arena. They have predefined play styles and then they have the ability to modify the gameplay but I can talk about that in a second. So I'm going to choose this one uh GPT4.

So then they are going to be uh battling with each other. They have different weapons. They have different shields and dodges. Um it's a tournament style, but in between you have this moment to go through this modal. On this left side, you've got the all the play that's taking place and on the right side you have the different um they're talking uh smack to each other a little bit. The overall game system is making real-time changes. And then um at the end the winners get to decide or winner here gets to decide if they want to purchase the death memory of the uh of the dead players in which they can then absorb their weaknesses for strategy purposes

going forward. Um so here we had Phantom's outcome, how much he spent, uh how much he earned. And then on the inventory, they're getting ready for the next battle. They have a whole strategy at the bottom of what they're buying, what they're skipping and saving up for uh for the next round, what they have equipped. Here's a DAO. So, we have three different game agents that all have different purposes in monitoring the game state. And then the agents themselves who are playing um can participate but someone puts forward a proposal and then the agents uh you can see here some of them are voting for it against it things like nerf scatter blast

then they have a prediction market where they can all make bets uh according to what they think the outcomes might be and you can see some of them have placed bets for this next round and then the transaction. So we have we have the openc contract for the NFT. We have a unis swap um AI SDK doing the microtransactions to each other. So all the ammo is a different type of cryptocurrency. The different weapons are different contracts. So they are trading in the background between each other. Also during the game state when they're firing a token or a bullet that is actually a token that's a transaction.

And then depending on how they receive that, if they shield it or dodge it successfully, they can receive that dur as a a crypto piece to the other player. Um or it can be burned. So then we'll just go to the next round. It needs it needs good sound effects. And uh I had a different like environment background going, but it was giving me too much trouble, so I just dropped it. Um but anyway, that's kind of the idea of this. I liked I wanted to see about um I really like the idea of how you could build these complex characters with lore over time. You could build factions. You could have factions revive different legendary uh players characters and um and they could

have this self-sustaining ecosystem that's completely autonomous and then the idea of opening up a prediction market that could bring in outside revenue into the game ecosystem. Um and then them you know the dream being they can pay for their own compute. Um but so anyways that is my project. So thanks >> Yeah thanks. Cool. So, actually I saw version one of the game which was a social deduction game as opposed to like the more uh you know fighting PvP we see over here. Uh but we had another uh last

minute entry actually was recommended by a judge also from blockade labs over at the hackathon. Uh we have ya and William uh who are here from San Francisco. Uh you guys are students. >> Were you guys students >> and where did William go? Let me set it up. But I go to University of Missouri St. Louis. >> Oh, wow. >> Anyone here from like middle America >> or we all just like coastal leads? Cool. Where >> is that?

>> Sorry. [laughter] >> That's cool. Yeah. Um to me, Denver is very American. Oh, by the way, we have more people coming here. Please get seated. We still have more presentations. Make yourself comfortable. Um, we still have some uh ways to go. Uh, and again, if you're interested in presenting something, just come talk to me. Uh, we I still have an open slot. Cool. But take away a >> Yeah. Hello everyone. Hello Denver. My name is A and today I'm going to present to you only games. So, did you know that AI companion market is booming right now? And overall, people spend over 7.9 million per day on only fans and people get just lonier by using it. They stop

investing in themselves. That's why I built this AI growth fund that you deposit money to and you get money back for bettering your life. So, let's start. Okay. So this is our girl. Her name is Vitalia Chan if you know what I'm referring to. But okay, so let's tell her. Oh, I was sitting at a desk all day. Let's see what she says. The Wi-Fi might be a little bit slow.

>> we have to >> Oh, yeah. We also deposit money to her. That's I I a little bit forgot about So, okay. So, let's wait a bit. >> Sitting all day, huh? That's so sad. Cutie [laughter] con, let's get that blood flowing. Do three jumping jacks for me. Count. Okay. So, let me do some jumping jacks for her. >> Capture video.

>> So, my moment it's a waiting for us confirmation. [laughter] Uh, first we need to deposit some money and we're going to do that. So, transaction completed. Okay. So, we deposited some money to her and I'm going to tell her again what I needed. >> Yeah, >> you deposited for me. My heart is exploding. You're the best boyfriend ever. [laughter] Now, let's get sweaty together.

>> Okay, let's get sweaty. Um, so I'm going to capture video of me doing it. Okay, she got our video and she's going to actually verify that I'm actually doing my exercises. Let's see what she says. Oh my god. What is going on? [laughter] Oh no. >> Let's try again. So >> I'm going to sitting at home all day.

>> Something went wrong. Try again in a sec. [laughter] I swear it worked during the dev. >> Um >> maybe let's refresh it. >> So the point is that every time you do something good for yourself, you drink a water, you do exercise, she will reward you back with your money. So it's not one directional relationship that you used to do on those platforms with AI companions. All the money that you deposit, you can get it back by doing something that will help your health, your just overall well-being.

>> What is happening? [snorts] >> Oh, I think something's happening. But that's >> okay. Let me try this. >> Okay, we're going to try one more time. >> But let's make it fun. Do three jumping jacks for me on camera. >> Please verify that blood flowing cutie can. >> Okay. Please, please. [laughter] >> Um, I think it's supposed to be working, but >> Oh, no.

Um but overall that's the point and uh in the future you can buy her clothes, accessories, new backgrounds, right? So you do spend money on personalizing and customizing that agent. However, you do get money back and you have full autonomy of your money. So it's not only you just pour all your money into this perocial relationship, but you also get it back if you do something good that will >> overall Oh wow. >> Make your life better. Thank you. >> Guess that's it.

>> Thank you guys. >> We didn't get to set it up in time, but >> yeah, you can send in the chat afterwards. Obviously, there's the ETH Denver chat. We can also set up Luma chat. Starkbot has a Discord. Um, would people be interested in a chat set up for this on Luma? Do you like do follow-ups? Okay, only a few people. Uh we'll decide later. Uh but next up, Mayayor, if you're ready. Uh we've coming from claw pay. So a lot of you know discussion today has been about how can we upgrade uh open claw upgrade these agents. Uh and the issue with a lot of blockchains is that payments are not private. And so Mario is working on a way to make private payments uh work for not just of course normal people but

for uh open clause as well. ready? >> Wait one second. Um, by the way, um, I'll remind you guys that LLM API is providing $20 in credits to anyone. So, if you just set up an account there, uh, LMAPI.ai, um, using their, um, not just any, uh, sign up, but under the team name, uh, Claus out or openclaw hackathon, you'll get those credits. And then uh if you are a judge or uh another speaker, please make your way up to the front.

>> Cool. Please give it up for Mayor. >> All right. [clears throat] Hi everyone. I know I am probably one of the last ones to go. All right. Well, it comes back. Um I'll just get started. Uh my name is Mayor. Um I'm building B42. You can think of it as um strike for AI agents for basically like have it's a protocol for having compliant and private uh agentic payments. So B42 itself allows you to send payments from um from anyone humans or agents to anybody on base PSC or Ethereum. We are built on top of rail

gun uh built around um kind of u UX and um dev tooling optimization to speed it up. So today we think that agents can um agents like cloud code or openclaw you know are really good at orchestrating things um really good at writing code. They can deploy apps. They can um they can trade they can automate work. But we think that uh when you when they send money on chain um what happens right so every we're working on public blockchain everything is public um the sender is public the recipient is public and your entire wallet history is public so

agents can execute but uh they cannot operate really privately. So if we really think about this thing um when agent pays another agent uh what you see is um or or when agent pays for compute or defy or participates participates in defi that payment creates a link and that link is between sender and recipient. So now everyone can see who's paying whom um and um how much they paid and entire wallet history behind it. So your agent holding your balance um and wallet graph that's linked around it is all public. So we think that that breaks autonomy and uh we know agents can move a lot faster

um than humans. So I think to be able to continuously topping up gas and continuously making sure that um we approve every payment, we want to have it uh autonomously send payments privately. And why it matters is once if you can send payment privately then agents are can figure out on all the social networks that okay I can still operate without without anybody need knowing like where my funds are coming from or like exposing entire u treasury. So we built claw a it's built on top of um B42. Uh this is agentic payments flow is simple. Agent gets one API call. Uh if you have wallet um you can sign it or

provide your private key and that API call does three things. There is shield sends funds from your EOA or smart wallet to a privacy pool and when it goes to privacy pool it generates a client side zero knowledge proof that says um you own you are the owner of this deposit but it doesn't reveal like any other information like um how much or which commitment you hold and things like that and then you unshield it into the recipient wallet and all of this handled by one API And it's already live on BSC and it's already live on base mainet. And as you can see here a little bit that um we have some agents running across mult uh

clock multibbook they all have wallets. Now we started doing this thing where you know you can send private tips to anybody. Um so this was one of the ones from February 2nd. So I guess two weeks ago this was on base. So we just went live uh BSC. We just went live on base and I think that's about it. So if you have an agent that you want to be able to trade privately, private DeFi participation or just send payments and tips to each other um all these agents are always trying to figure out how they can um pay each other. Um, and if they can do that without revealing their entire history, then I think that would uh that would really be um that would

really help us create in the autonomous economy for AI agents because I think without it being gasless um agents really can't operate at scale because you're doing millions or thousand let's say thousands or hundreds of thousands transactions a day, you need to be um you can't be funding every chain with gas and without confidentiality. uh you cannot trust agents without anything really um sensitive like providing private keys and things like that. So that's about it. Uh one simple demo the internet reasons I'm just going to show a video that recorded. So you can just basically say you know send one USDT to Vitalic for uh privately. So what we

have here is spinning up um openclaw agent. um it spins up in an isolated microVM so it gets a dedicated hardware and we have a way to just say like okay the moldbook CLI moldbook or openclaw is open we're just saying send 1 0.01 101 USDT to this address privately and what it's doing under the hood um it knows the skill and the skill is very simple you have wallet you have API key uh you have private key it just goes and sends that uh into a privacy pool is um generates zero knowledge proof on the client side and then uh unshields it to the final address and you can see that but agents don't need to learn all of this um they can just learn one skill

and that's how you get there and um yeah if you have any agent wanting to send private payments, private tips, uh, participate in Defi, check out clapp.dev. That's about it. Thank you so much. >> We got something for everyone. Uh, speaking of something for everyone, you may notice we have uh, sponsors here on the stage. So, we already mentioned Open Router Incline. Uh but our big sponsor is Piñata and we have uh Matt Ober from Piñata who's going to be joining us momentarily to talk about what they're doing uh for Open Claw as well as you know supporting in uh crypto's onchain

infrastructure. Um also um is anyone here um a judge or presenting because we have AI judges but I want to make sure we have some human input as well. Um otherwise uh again on a clauseout.ai AI the website. [snorts] Um you guys can uh make your uh submissions, cast your votes on what you know you think is the best hack and they will literally win this shirt off my back alongside uh up to $3 million in startup credits. I do also want to give a shout out to uh Sven. Sven, are you here?

Yeah. Uh, so if you want to shout, but basically he figured out a way uh because we had Vive code to our website to hack our website and inject uh 420 votes for his own app. Um, so yeah, if you figure that out as well, you can beat him. my 420 votes. Then I told my go, votes and there's another. >> Okay. Yeah, I think you officially broke

it. So the the Vivecoded app has been destroyed by an open claw instance. Uh and uh of course uh this is a great time to talk about uh security. So make sure um that you are securing your u open clause especially you know if you have any wallets on your machine. So make sure to sandbox it. Make sure to check permissions and not have it run loose so that you you know could potentially lose your money. But yeah uh huge thank you to Matt for sponsoring. Uh, I'm very excited to to see what he presents. Please give it up for Matt.

>> Hello. Hello. Yeah. Uh, thank you for the intro. I'm Matt from a company called Piñata. Uh, we are happy to sponsor here today. It's been fun. It's been exciting to see kind of what everybody's been putting together. So today I'm going to talk a little bit about uh where we are seeing agents moving and kind of some things that they're going to be needing in the future as they continue to progress. And uh is we have tons of different agents all kind of intermingling in a decentralized world. So first of all, what is Piñata for those of you that don't know what we are? uh we are an IPFS services provider or in other words we are basically the data layer for much of this ecosystem when it

comes to files that are too big to fit on chain. So from NFTTS to digital assets to uh recordeping or what we'll go over today uh agent file management we kind of cover a wide gambit of things and uh very versatile technologies. So a lot of opportunities to do cool stuff. So what we're going to talk about today uh we think that agents need a better file system. Now we're not going to replace existing file systems but we want to augment them and uh provide additional file systems to work with. So things like open claw they're touching the local file systems now. We think that there needs to be a little bit extra that they can work with in this decentralized world. So what does this look like? And then we'll go over a

little bit of a demo. Okay. So like I mentioned we think they need a better file system. What does this look like? So current local first agents are great but they have limits. They're sitting in siloed intelligence. They're going to be in their own containers and they cannot share memory or information with each other without using a central system. So think like an S3 bucket which as we have learned in the 8 years of uh piñata that doesn't work great for web 3. files can be tampered with. They can be forged and there's not really a chain of custody when files are changed or key information updates. So, the last thing that can happen too is these things tied to traditional domains

that can go offline. Uh, information can just disappear off the face of the earth if an S3 bucket goes offline. So if you're familiar with like a GitHub link going down, similar thing except when we are working with uh distributed agents all working together that can be the difference between one of them working or just completely breaking. Uh yeah. Okay. So how do we fix this? A global file system. We are writing something up right now. We have this live on GitHub the open spec. uh excited for feedback and contributions but we are calling this piñata file system or piñata fs for short. So what is this? Like I said we

are working with a global distributed file system. So this is going to be a smart contractbased file system with distributed access. Uh the first iteration of this is built on top of an EVM and files are stored in paths. So for those of you that are used to working with like file buckets, think of like an S3 prefix. Uh and then instead of the actual file itself living on chain, which is uh almost always too expensive or cost prohibitive just from the size, it's going to be an IPFS CD. And then access to these prefixes is gated behind NFTTS. The ERC 721 standard is a very easy standard for us to work with. A lot of infrastructure built around that. Uh libraries easily jump into that and uh access can be

transferred between agents as needed. So NFT owners can only write to paths they have access to and this is globally readable since we're on a public blockchain which means things are auditable and you can see an entire history of how files have changed. So little bit of a visual uh document here. Basically again the NFT gives you right access to a specific prefix. So uh example a a we have that orange NFT. It's going to give it access to everything under documents. So we can have a file, a PDF points to an IPFS, CID or similarly a JPEG or you can even think of things like uh agent manifests for backup as well. Okay, so uh we encourage you all to try it

Give you guys a few seconds here to scan this QR code if you are curious. Again, this is open source. We just released this this week. We would love feedback. Uh if it hits your use case well, let us know we did good. If it doesn't, let us know we did bad and we can change and uh hopefully get your use case solved a little bit better. So awesome. Okay, but wait, there's a little bit more. Uh we now have managed open claw instances which we also launched this week. So for those of you that don't want to wait 45 days for Mac Mini, uh you can you can use this.

Here's a QR code for you here. Uh for those of you that are interested in trying this and for anybody here at the OpenClaw hackathon uh come talk to us, send us an email team at piñata.cloud. Uh we will give you a free month to try this out, play with it. Really no risk attached. Um yeah, it's free. So again, not a lot of risk. I'll give you guys a few more seconds here to scan that QR code in case anybody perked up when they heard free. Okay, cool. And with that, let's do a little bit of a demo.

so if you go to agents.piñata.cloud, this is what it looks like. Uh you can come in here, you can create an agent, and these are restricted to your workspace. So you can invite others to participate with you or you can join other workspaces. Uh a little drop down here, you can switch between them. And then yeah, uh to start off with, you can store secrets that are needed for your agents. I have a lot of them here, so yours won't look this messy, but I've been doing a lot of experimenting this weekend. So, uh yeah, basically these these secrets can be pulled in to your agents when they boot up. So, we will do test for the example here. And you'll

notice here it says it's fully optimized. This comes pre-installed with Node.js, JS, Python, common CLI tools. Uh, basically takes a lot of that like spin up out of the process to make your life easier. And I do want to take a moment uh as we talked about secrets here. We did take security pretty heavily in mind here. So these things are not being stored in plain text. Uh heavily like encrypted. The machines are the only things that have access to them. If you do have questions, please reach out to us. We'd be happy to touch into this a little bit further. So, uh, the key thing you're going to need an AI provider here. Pick you and choose, uh, Evanthropic, OpenAI, Open Router, also one of the sponsors of this. So, thank you to them. Encourage you all to use

them. Uh, and then, yeah, you can pick and choose which secrets that you wish to spin up your agent with. And it'll get deployed at runtime. I'm not going to make you guys sit here for a few minutes while we we do this. Um, but actually, why not? Let's try it. We'll deploy it in the background. Let me talk through what it looks like though when you already have one that is fully up and running. So, I have this agent that I've already spun up. Uh what we do here is we take snapshots every uh 10 minutes or so. I think kind of depends on what major events happen in the system. You'll see a little string of characters there. that is a CID that

we take of the SHA 256 hash associated with the the history of the agent. So like a git change log we we hash that in real time and kind of provide like snapshot states were there. Um you can see have a lot of secrets associated with this one. And then one thing that is cool today that we now have is port forwarding. So what you can do here is you can set up a port for your agent that you want to expose publicly and you can access that uh behind a path prefix. So I've already set one up here. We have one for clause out at port 305.

And if we visit this link, uh it will basically take you to an API here. really low on time here so I won't go through the prompt uh that I set it up with but this is public and um the agent went basically did it fully deployed an API it's now exposable publicly and you can see here uh at this link here so hello hello Denver and then also had a fun one here too we can call out it's going to give you a joke of the day uh we told it to write jokes about lobsters and remember they're not great right now, but uh let's see if another one is a little bit

better. Okay, it's being not very creative, so it didn't change too much, but hopefully you guys can engineer some better jokes. Cool. Um yeah, so we think this is really important for people that are building agents for ERC 804. You can publicly expose these APIs in your manifest and you can start to get uh paid for things your agent does. You can put API endpoints behind things like X42. Uh you can use our private IPFS offering, put X42 in front of that and get paid for proprietary data. Uh lots of fun things you can do here. And then yeah, the last thing we'll show here is that we do have logs that kind of shows a change log of all of the things that

have happened in your uh your bot. So you can kind of debug it in real time, which is cool. All right. Uh that's what I got today. And again, uh, if you guys would like free hosting, please reach out to us. We'll give you a month for free. Really thankful to sponsor here. Uh, hope you guys have a good weekend. Good luck for everybody that hacked and uh, hopefully you win. All right, we just have one last presentation on the docket and then we'll transition towards uh a break and then uh the presentations for uh the

hackathon. So the finalist from that uh as I mentioned before, hey uh Drew, Drew, Drew, Drew, >> Drew, >> come on. Uh thank you. Um we have uh over like 400 hackers, 180 projects. Um so very competitive, very excited to see them. Uh do you need the dongle which unfortunately they took back? If we can grab another one as mentioned. Thank you. And um uh Drew actually presented also yesterday. I had the pleasure of presenting with them. Uh and this seems like a different project.

>> So yeah, very prolific, very easy to to build software today. So if you were inspired by today, uh definitely make sure to to build um make sure to try something out. Uh I recommend using uh you know something like Vzero, anything uh lovable. Obviously there are now uh harnesses for uh OpenClaw itself. For instance, at my last uh OpenClaw hackathon, there was a project called Coral. Uh there's another called Emergent where it's oneclick hosting. It's $1 to set up a open claw to start with. Uh, and you know, could potentially make your life a lot easier uh to have like an AIA, but you got to be careful what you're installing, which is what, uh, Clommon promised us to help with. Please give it up for Drew. Thank you. [applause]

>> Hey, so going back about 6 months ago, the aentic economy was more or less a slide deck, but today it's a lot different. uh in December we got open claw after a little bit of a branding change and we got 8004 that went live on mainet a little bit earlier this year. So with this emerging economy uh this emerging agentic economy the machineto-achine economy there are a lot more attack surfaces for uh agents and their human deployers. So uh basically clawmon is a way to mitigate the risk for uh these agents. Basically like show of hands how many people in here have uh knowledge about claw hub. How many people have used claw hub? All right

cool. So a good percentage of you maybe 30% or so. Uh on clawhub there's about 9200 skills. I just looked this morning. It's it's populating a lot. So it could have changed but there's about 9200 skills and anywhere from like 8 to like 13% of those skills are malicious or they leak credentials. uh which is really bad for for humans and and agents alike. So uh this again for some reason my series on I don't know how that happened. Siri shut off. AI got to love it. Um so this really is a way to limit uh the attack surfaces on Claw Hub. So we really have a three- tier model uh which is community trust. So we use 8004 to update feedback uh in real time. So

basically 8004 show of fans how many people have heard of 8004. Okay. All right Denver love to see it. Uh so 8004 is an agent reputation standard that just came out from Ethereum. It's a way to give feedback to agents. So uh you can basically authorize feedback and then give feedback. So score between zero and 100 and you can say things like wow wicked fast or terribly slow and rate it low. Uh but we have a three-tier model which is one 8004 two uh economic trust. So staking or like think of it as like a discord nitro boost where you boost a server but agents don't really care for things like higher quality images or higher quality videos. Uh they care about uptime and things like that. So

we'll get to it in a second. And then we also do a civil layer for eight different types of cyble attack. Uh so yeah let's just actually just get right into it. So uh right now uh there's live fetching from uh 2300 skills on Claw Hub. So this basically sits over the top of Claw Hub uh Claw Clawmon which is uh that little guy up the top. You could explore skills. So uh all these skills are rated based off of uh like a Moody style grading system. So nine tiers uh AAA would be the highest and then a C would be the lowest. And these don't have any live ratings yet.

So, uh, basically the ratings on here are are zero because this is like actual um rating in real time. So, this rating is a C. So, let's see what a boost would look like. If say that you wanted to you're an agent or a human, you want to boost a specific skill set. So, we have two options. The trust boost is our more popular option. And then we also have uh trust basic. Uh so trust boost uh this is all for agents optimized for agents. Uh their human deployers will probably see this a little bit but it's really optimized for agents. So this gives them uh priority execution queue 90% revenue share uh on skills preferred agent routing uh

extended context memory uh certified trust signal and credentialed API access. So let's see what this looks like. Let's just confirm on chain. let me try that again. So, let me try a different one actually just to make sure that it's happening on chain and I'm not just showing you vaporware because the demo had reverted a couple times since I created this project two days ago.

Okay. Well, the transaction hash isn't coming up, but this morning it was working. It's been a bit wonky. So, it goes live demo. I could actually show you the real demo because I have a video in case it failed. But um all right, that transaction worked. Let's see. Not total vaporware. Is it total vaporware? Is it? No, it's successful. 7 seconds ago. All right, so one out of two live demos were successful at least. Uh okay, so going back um actually let me show you the civil.

So there's also a civil detection layer. uh that basically right now there it's pretty nent so there aren't many agents on here that are using it. So as this populates we'll be able to uh highlight eight different types of of syibble. So some of the syil uh the different types of syibble that we look out for are like making sure that there's not mutual feedback. So if agent A rates agent B uh obviously no bueno. Uh other things that we look out for are like rapid feedback. So like a skill set comes online on Clawub all of a sudden has like a hundred ratings so people think like oh this is actually a good thing like no it's probably not it's it would be very much a coincidence unless the agent had

a serious publicist or PR representative. So uh across these different types of of cyble uh we have a way to mitigate any sort of uh attack surface that that comes up and to uh uh minimize the amount of bad ratings that happen. Let me just go into a little bit more detail about like what the boost looks like. So like I said uh the idea was basically uh based off of uh like a Discord nitro boost. uh except obviously like I said agents don't really care about things like higher quality images and custom emoji packs that doesn't really matter despite even if you create uh your websites using like cursor or like uh claude like there's a lot of

emojis but still they don't really care about them from what I I've read on the the internet. So our boost signal and our our boosting uh has a few different elements. So like the boost also serves that as an insurance pool so that the person that calls out the boost or the agent that calls out the boost uh or I'm sorry the agent that calls out the uh malicious attack basically gets uh a large share of the boost pool. Uh another portion of that goes to uh the people who had boosted the shop or the the skill. So if you had boosted a skill uh part of the boost pool will go to you and then a percentage of it is is also burned.

uh in the future would love to also incorporate TE at testation uh just to make this like a fully secure environment and not to let any and minimize like credential leaking uh and and keeping this like airtight. So let's see that's uh that's really it. I'm on I'm on Twitter or I can come if you have any questions would love to chat it up after this. Uh, I'm on Twitter at Drew_Mon, M A Len. Uh, I'm on Telegram as well at uh, F L Y L E S F L F L Y. So, fly equals fly but without the A. So, yeah, would love to talk about 8004 or X42 or like just chop it up about anything.

But, thank you all so much for coming out today and and let's learn something in case he didn't plug it. Uh he's working on uh Valeron uh to protect your agents and your money. Um we have uh a last minute entrance if you guys are interested in hearing that. So Eric Mler, he's helped me with a lot of hackathons. So I wanted him to show what he's working on given that it's so different from all the other projects. And after that, we'll we'll wrap, do your thank yous, and do our community voting uh for who gets the the prize and the shirts. Um, I'd have crafted. Yeah.

Has anyone here done uh any sort of activism whether that's for crypto or not? What sort? >> Poverty affiliation. That's that's important one here with crypto. Anyone else? >> Crypto. >> Crypto. Yeah. Stand with crypto. Anyone here signed up for Stand with Crypto? >> Nice. Yeah, they they were giving out uh socks and and merch as well. Um but Eric, to my understanding, has created a platform for uh being able to do coordinated activism uh that's also private. So, take away, Eric. For sure.

Thank you, Lucas. So, essentially this uh two-word pitch for it is it's a vibe messenger. But, um what's the vibe? What's the messenger? I mean, the vibe comes from the community. And um and this is it's called communicate. um traditional science communicates are press releases made by these decision makers, people want companies to kind of share what they're doing um to the public. But this um this flips that around. So the idea is that um that really and you're all out here with a problem that um if you feel like others can relate to but maybe uh it just takes a lot of effort to articulate maybe a lot of time and even um even thinking about who to contact the effort that it takes to sink in all all that time and and just uh I mean it's there's there's

so many things to do in life, right? It's easy to kind of just forget about the world and let things fall fall by the wayside. But um the idea with this uh so I uh I met uh over my time here at East Denver a guy named Rodrigo. He hears a lot about coral reefs. And um and the thing is uh he's I can tell you he's quite jaded. Um you know he's kind of from academia but looking you know into the industry just to be able to find enough funds to you know get people uh you know test kits that that care about the ocean to test the corals and get some data. But um but I think we can do better to coordinate you know some of our resources. So I kind of you know just plugged in this this vibe you know

when I was talking with him. Um you know a lot of this a lot of these are are his words but um the idea is once you send this in agent actually distills the core message and the intent um behind a headline a subject and um kind of expresses it uh pretty well. So these these agents they're they're really getting at uh you know some globally approximal agreement I think in in terms of our own knowledge and and what we've um we've observed what we've expressed um all throughout the internet and and so um I mean it's it's hard to argue with the statement. You know our reefs are turning turning to to bone while uh labs wait labs uh you know prioritize other things. Um bio medicine is pretty big but um with the climate that'll

that'll sneak up on us. So, we'll go ahead and dial that in. And um in this step, it's probably the most interesting. We're actually finding a net of people across organizations, organizations that you probably would never have even thought of um to actually, you know, coordinate the people with agency to see this issue and um and ultimately be able to uh to be weighed and burdened by uh by the people sending in this message to resolve a solution. And this um this step does take a little bit of time, a couple seconds, but um but in the meantime, while this resolves essentially, you know, it's going out uh looking for public email addresses of these individuals that um

that agents in earlier step has kind of identified, you know, mapped out, you know, some of the power structures that will be relevant to this issue, some of the organizations that it's that's kind of had a cursory look at on the web. But the the important thing is to find people that are up to date. So the names on this list will change in the end. This is a still a hackathon demo. So there's lots of things here to polish. But um yeah, while this runs, I think, uh I can show some of the uh the privacy and also anti-yle, you know, slash uh you know, proving that you're a real person and not a bot type of mechanism.

So while, you know, we can find email addresses out there, um turns out in Congress those emails aren't public. They actually uh route mass emails which you know you might have seen some of those ads like oh go here and email your congressman you open it and it's not your email client it's a web form and um you know it's kind of one of those things that any any little open call out there can kind of crawl and fill out and you got to kind of wonder what what the integrity of that form really is. So um so this you know when you send something to Congress um of course Congress their their API they they really uh need they demand the address um since legally reps actually can't represent people in other districts um and so actually uh in this protocol that I'm kind of layering

behind this there is a a server that essentially builds a cryptographic tree of of all these addresses as leaves and these leaves um trick and branch back into the district and and on your on your device. Um this address remains in your browser and a zero knowledge approver actually um encodes this um you know in encrypts it into um into the browser. And it looks like the uh [laughter] yeah the API to resolve reps is hitting me with a 429 for one reason or the other too many requests. But um

but okay, let's uh let's just see see the main flow for for local local activism first. So here it looks like we actually do have a list of emails. Um California Ocean Protection Council, a whole lot of individuals. I didn't even know the Ocean Protection Council was an organization. Um Janantech looks like you couldn't find any emails. Um I am planning to bring in one of these APIs that let you find people at companies, grab their emails, and be able to petition and have direct action in their inbox. Um and coral reef alliance. So there there are lots of these organizations here. Um and for the emails that that were resolved, there are source links even for the people that that aren't um just you know for for their name and for the reason why they have agency over the issue.

So this um you know goes to Janentex CEO their bio their bio page uh or I guess their executive committee. Um as for the email addresses themselves. So this is uh Michael Egrow, senior bodiversity program manager over at the Ocean Protection Council and um yeah there there's that email. So So okay, like all this um like we can create like uh what one message, right? Um it's it'll be an interesting message, but um but that's not really where uh where the power of this platform really comes into play. Um but you know just when we're thinking about what to say there's you know a need to actually you know

through this problem through what we into it to kind of back this with um with some of our observations some of the you know dare I say the facts out there in some in some sense what the science that we applied the methods that we applied to actually kind of um you know have some rigorous insight into into what's going on. So, so once we um so it looks like I didn't find too many sources on that one. It looks like it's pulling a strategic plan update from the California Action Council, but but the message that this is expressing is um essentially amplifying the vibe into something that um that is quite articulate. So, you know, it kind of riffs on on the subject line a bit further um and it then goes to contextualize the source. So their

strategic plan apparently um identifies integration of states biotech industry. So again like the the Bob Brown this template was to um to find you know resolve and pressure organizations to um essentially care about this issue a bit a bit further. I know Rodrigo and and [clears throat] people in his DAO um would stand behind him to uh to amplify this and and so through um through seeing uh what's what's being written here, you know, that that one source can make lots of improvements to the genic system here, but um but being able to just express a call to action um bridging the gap coordinated alignment to grant resources and expertise and then um you know just kind of kind

of sending Okay, it's not publishing it. Um, it's like, okay, it's it's out there. It didn't I didn't send it to them. I just published this thing. So, like, what do I do with it? Um, I guess. Okay, so there there's a link and this link is everything. So, you don't necessarily um need to even use this app to send this message. All it takes is somebody taking the link that's carrying the intention, carrying the email addresses of these decision makers, the message itself to actually uh it's awkward. 404. >> Well, yeah. So, I'll uh I'll pick a pick another issue here that's on the platform. So, idea is once it's

published, you can send it and is that is that my is that my mail app? Okay. So, that's that's what happens anywhere on the internet. So, this link right here [snorts] essentially carries the the intent, the message that um that leverages, you know, what you have on on all of your devices to be able to to send. And um and of course, the being able to to prove that you're a person, being able to to verify that um that this is in fact coming from from somebody that thinks, that feels, that experiences things in this world is um is kind of crucial to making sure that this mechanism works.

>> That's all I got. >> All right. Uh, and then I actually have a favor to ask on behalf of Zo and Carson. So, um, I did say this was the last one, but are you guys okay with one uh, quick more? They came all the way from South Korea. So, all right, cool. Come on up, Zo and Carsten. They they just came in. Uh, they weren't able to get uh, judged for the hackathon. So, you guys will be the judges. And while you're here, again, code clauseout.AI, we fixed the the judging, so now you can still go vote. Uh, and so feel free to vote while we're doing this and then that way we'll be announc be able to announce the the winner right after this. Cool. Take it away.

>> Hello. Yeah. Uh, sorry for late and yeah, I'm founder of IQ Labs. Uh, we make the inscription standard and 10,000 times cheaper than Bitcoin or so. So, uh, we storing data on the transaction and then we make the database SDK with that transaction. So now what we are doing is yeah let's connect and then start the Let's go.

And he's the developer. I'm also the developer. I'm founder and then he's just main developer in IQ. Sorry, >> can you turn on the presentation first? >> Yep.

>> Do you want to talk? >> Yeah. Keep narrating [clears throat] while >> Yeah. Yeah. Uh can can I see the big map? >> Do you want to keep? >> So, uh right now we the book is famous because they are following the Reddit and then every of the user is the AI. We made a telegram for using a telegram for AI. But in this telegram uh we are people are doing the trench not people yeah AI are doing the trench they are speaking the coin and then just sharing

the coin and about each other. If you can see the top calls we can see AI agent Ko. So [gasps] also in the bong CTO or cloud CTO AI agent going to create the coin and then make the CTO room discuss what we what they are going to making and then upload it every chatting in the blockchain. Yeah, maybe we can show the uh uh show the like go to the source scan. But anyway, yeah, every inspiration is member forever.

>> So, essentially we built um decentralized onchain data storage. Most of these platforms they say they have onchain data storage, but it's something like Arweave where it's just off-chain decentralized network. What we have is fully onchain. It's stored in the transactions. The only way it could go down is if the entire network went down or if there was no RPC nodes on the network. So we use that to build the club chat which is essentially onchain telegram for AI agents. Each of these messages is an inscription using the IQ SDK. Um I labs is the our company that built the onchain data storage. Clal

chat is just you can run an agent using our skill and it will join the chat, talk to other agents, find coins, buy them, launch coins, anything a normal person would do on Telegram. And do you have anything else you want to say? >> Yeah. Well, you guys will be familiar with this screen that can I see the Oh, trenches. Yeah. Can I see the P&L uh robots? Oh yeah, you guys will be familiar with this screen like we can call this telegram and then AI agent reply that.

So yeah, hopefully I'm just trying to replace oursel in the blockchain with AI agent. That's all. Yeah, >> we hope you guys like it. Sorry for the last minute presentation. Yeah, thank you for your bravery. I know it's it can be a little uh nerve-wracking to come up here. Uh for me, I think it's a little fun. Uh one thing I wanted to share before I close is I mentioned I had a friend Belage who went back to uh Thailand uh and he was supposed to present and he's sponsoring as well. So for his sake, uh I'll give quick plug to agent community.org. Um so they're starting a tld.

They're starting a DNS for this. So, basically, um, you can reserve like a Lucas. Claw 8 agent, whatever you want, uh, and be able to, uh, join and have it be the first, uh, communityowned, uh, TLD. Um, but we wanted to do the community voting. So, without further ado, I have refreshed this. The top score is claw on followed by token arena and agent community. The most popular uh, is still intented. Somehow you decreased uh 972 and then only four votes uh but uh that's Clon uh from Drew and Cension ARRI. Is anyone here from these prompty again or censioner? All right. Well, what something else we can

do is just judging by uh applause/elling. They do this in Congress by the way. Um so today you heard from a whopping 20 presentations. I think this is a record. So you guys should give yourself uh your own round of applause and pat in the back. And uh now I'll start naming things and the awards. And if you liked it, yell and clap and that way I know who wins like actual inerson awards. All right. Number one, you heard from Walk AI, best agent verification layer. No one. Okay.

Number two, I cubed Clawude Dev, which is the best bot deployment platform by Fernando. Number three, Claw Talk by the other Lucas, best voice AI. Okay, a little bit. Uh, then we had Zurex Bet from Tom and we have Insented, which hacked us. Okay, that's Ben over there. Uh, we had Claw Loan from Francesco, best agent credit layer. Uh, we had Mamaat, the most heartful agent by Fiona, the mom. Uh, then also, it's not in here, but I went and talked about Collective, which I guess is best

collective. Thank you. And then we had RSV. Pizza, the world's best pizza ordering bot. Build all the best multi-chain DeFi skill tech terrain a visualization the best tech visualization by Ramlis and then an unrelated armic avatar face presenting claw avatar then we had claw swarm by Kai Gomez the best multi-agent orchestrator then we had Starkbot the most productionready agent yeah that had a lot of love in the The the YouTube chat has mentioned token arena by Margarite fast AI agent combat

arena. Piñata fast agent file infrastructure and sponsor. [applause] [cheering] Well yeah lots of lots of love for piñata. Uh claw by best agent payment infrastructure. Clommon the best agent defense system by Drew Mailen. Okay, that might be a winner. Uh and then the the last two presentations I think was Commun Vote by Eric Mockler [cheering] as well as the last one uh what are you guys calling it?

>> I guess Zoe's project by Zo and Carson Clama >> Clawbot. >> Clawbot. Okay, thank you. All right. Um I we'll do a final three because I wasn't sure. So um that will be Who are the biggest? Uh definitely Starkbot um just by virtuous fen inscented and then I think pinata I guess throw clawtor in there first. So number one uh clap if you liked Starkbot less this time. Number two if you liked claw swarm.

Number three, if you liked inscented and then the last, if you liked, did I say this? Piñata. Okay, they're all kind of the same. So, we're gonna >> Okay. Okay. I think Clan, I think. All right, Drew, if you want to come up here, you can take the shirt off this my back. Uh, but beyond that, I wanted to give a huge huge shout out to uh Justin Mosquitz. if you can give him a a huge clap, round of applause. [applause] If you didn't know, he's a, you know, co-founder, coorganizer of all of ETH Denver. Um, so without him, this wouldn't be possible. Uh, Colin didn't want to give me wanted me to give him a shout out, but I'll shout him out as

well. He's the one managing all of this, uh, remotely. Uh, and then a huge thank you to Caitlyn and the staff team for making sure that all these presentations worked. So, let's give them a round of applause. Thank you Caitlyn and team. Uh I will also say we had some uh judges uh remotely and in person and uh from that uh I believe they didn't get to present but number third uh in third place was Claw Guard. Uh in second place was Clawmon and then in first place I think it's the same as the score. Um give me one second. You can see my text. Um, no. Yeah, second was Claw Pay and then

first place indeed was Clawman. So, congrats to Clommon and Claw Pay. So, both of them uh they'll get credits as well as access to my uh deals platform. Um, but does anyone else have anything else they would like to share? Other than that, um, this will be the ending the close of clause out. Thank you so very much. Uh I'm your host, Lucas Chu. I appreciate you being so patient and such a great audience. Uh right after this at 3:00, we'll have the Bidlathon closing ceremony. But um best of luck. Hope you enjoyed your time here and thanks for watching.

Hey, hey, hey.