Gaia, A Decentralized Ecosystem to Support AI Applications

Transcript

GM, welcome to what's going to be another great episode of Devantel. So, if you didn't know, Devantel is a 30-minute podcast held every week allowing founders, hackers, and anyone in between to come on the show and uh tell us about their product. And today, I'm ecstatic to welcome Meowi and Toby who are devs from Gaia. So, if you didn't know, Gaia is building a decentralized ecosystem to support AI applications that learn, improve, and grow over time. So, if you stick around for today's episode, you'll see Meowi and Toby give us an overview of Gaia, what features it packs, and how you can get started using it today. All right, let's get into it. But first, a word from our sponsor. Today's episode of Devantel is brought to you by our friends at Arbitum. Arbitum is a technology suite designed to scale

Ethereum. You can use Arbitum chains to do all the things you do on Ethereum, but your transactions will be cheaper and faster. With over 3.8 million ETH saved, 1.69 69 million active wallets and over 1.7 billion total transactions. Why would you want to settle for any less? Get started today by exploring Arbitum's comprehensive docs and tutorials linked in the description below to help you get started launching your next DAP today. Now, on to today's [Music] show. GMG, welcome to the show. Meowi and Toby. Uh, pleasure to have both of you on today.

GM super excited to be back here and uh yeah this is one of the OG show from developer that is the longest and still running till this day. Indeed it is. Indeed it is. And yeah, as long as people keep building really cool products, um there's always going to be a stage here for them to uh showcase it. And uh yeah, uh Gaia is uh certainly an interesting one uh at the intersection of AI and uh crypto. Uh so really excited for this one to hear uh uh basically learn about it. Um, but before uh we get into all the great content you have uh packed for us, uh why don't you give uh the crowd a introduction for

yourselves in case folks aren't familiar? Okay, I can go first. GMJ, this is Mary. Um I started my web3 journey from developer DAO. Then uh I it was like over three and a half years ago and then I joined consensus and I was uh on the rapid application development team or devil team later and uh now I'm at Gaia. So Gaia is a decentralized AI infrastructure. Uh so yeah we're going to talk about that but as this is I love building and uh talking to people and how I can help things easier as a dem.

So yeah, I'll give the popcorn to Toby. Hi. Hello everyone. I'm Toby Loba and I'm from Nigeria, Lagos, Nigeria and um first time here on Dev and Tell. Excited to be here. M has told me a lot about this um uh you know this podcast and the show. So it's actually an honor to be here. Um my journey started um my journey in tech started like in 2022. uh learned to code um uh learned to build on the front end mostly because that was what I used to like you know learn how to like actually write software and then from then on yeah I progressed into you know open source uh in open source I did a lot more things made a lot of friends me is one of the great friends that I made from you know contributing to open

source and uh from there he sort of like uh sprung into me now building on um you know AI agents you know building on building on the blockchain and you know building web 3 Cuz when open with open source you never stop learning. So as you build you learn. So that was essentially the process of how I you know was able to delve into this tech spaces and you know you know me now working in web 3. So uh excited to be here. Yeah. Amazing. Amazing. And uh yeah thank you for the kind words. Um uh I was telling telling you both before we went on uh as long as people keep building cool products um we'll keep uh having this show.

Um, and I I I suppose um from all your contributions from open-source, uh, Toby, um, as well as yourself, Mi, uh, I know you're a big open source contributor, too. Like, um, what made what made both of you kind of stick to web 3? Um, like what what keeps you what keeps you both, um, in this industry? I think I can go first. It's like when you first joined developer to it, it was all about learning and building with friends. So it was like we joined here for a movement like web three. We want to build with friends and I think we stayed for me I stayed for the community all the friends that I made along the

way like and having to wake up every day and build something exciting is really fun. And if it's with your friends, it's it's more fun. So, yeah. Yeah. Uh yeah, for you, uh sorry, the question was um uh why um could you repeat it? I'm trying to like my response. So, uh yeah, you you do a lot of open source contribution. um you can choose to do open source contribution for a regular web 2 project or web two projects, but you specific specifically um stuck with web 3 and found yourself at a um at a company um that does web 3 and AI. Uh like what what made you come

to this industry and and made you stay? Oh yeah. Um as with most of the ethos of like open source, you realize that open source sort of like um uh prioritizes um uh you know collaboration. It prioritizes community and it also prioritizes um you know perspectives. Uh if you think about open source is just um uh you know you know a garden of people with several different perspectives that are you know coming together to then build something beautiful or maybe not beautiful in some other cases but at least as long as it works you know you use it. Um so yeah that sort of like perspective just like you know drew me a lot more to web 3 cuz uh with web 3 it seemed like super niche. Uh it was you know was very

complicated. A lot of people didn't know how most things work there me included back then. But then I was like um yeah why why would any technology be complicated. So my goal in web 3 uh since joining has always been like you know tend to like always you know you know every day to like lower that barrier and be like okay this is just another software stack that you should learn. It's not something that you should you know uh you know break your brain over. It's just something you can still use like everyday life and um you don't even have to know like the blockchain is powering this piece of software. You just have to know that okay you make a call uh you get a response uh you send your regular you know HTTP request and you get a response back maybe you're calling a protocol so

yeah you know those things just like yeah I mean the fact that it's niche and um the on boarding um is something that I think needs a lot of work was was one of the things that like drew me to it and then you know AI agents just came along the way cuz I was like I mean the Eliza boom sort of just drew everybody to AI engineering I guess So yeah. Yeah. Yeah. Indeed, indeed. Um and and yeah, I was going to um also uh say like um web 3 has its own um uh complicated rabbit holes to go down. Um and AI is like a whole another uh a whole another rabbit hole in itself. Yeah. Um so uh what what um in particular um made both of you

gravitate towards Gaia? um as we mentioned um a project that is both um at the intersection of of web 3 and um AI. I think I can maybe take a lead on this then I know Toby also might have something to add. I think what drew me in about Gaia is that okay let's let's talk about what Gaia is first. Gaia is decentralized AI infrastructure. So what do I mean by that? So think about this in a centralized our centralized AIS like chat GPT if you don't pay for a month do you have access to all the premium features you kind of don't that's that's not something you own and what drew us

towards this web three is the ownership of it and that's how we I for me I also drew towards like Gaia okay this is this is my AI this is my data this is all I could own and I train it however I want. It's not going to be biased towards anything. It's not going to be like have these own sets of uh data that's just comes within or even take my own thing. So with this it's it's my own AI that I could have as long as I want and no one can access it. It's it's the privacy.

It's it's the decentralization. It's it's all the parts of it that drew me in for me. I would give popcorn to Toby. I know he has something to add. Yeah. So, so with Gaia, it was actually a very like fun story to be honest cuz I actually found out about Gaia in this, you know, small community of ours that we call Akeru. Uh, so the idea behind a was that we wanted to actually build, you know, actual open-source AI, you know, like not Yeah. not not as a brand name. So, um, so then we sort of just started the idea and then um, you know, decentralization came in cuz most of the folks there were like a web 3 folks. So um put web three then put AI together and then you start thinking um how can AI run on on defense decentralized

infrastructure? How can AI guarantee data ownership and how can AI guarantee you um you know um actual earning um would I how would I put it actual earning um you know incentives such that you're able to use your own knowledge to be in a sort of network of or a system that you're able to use information to exchange or something like that. So, um, Gaia just sort of ticked all those boxes for me cuz, uh, finding out about it and then I found out you can actually still then use Gaia like every other regular, um, how would I call it, every other regular like, um, uh, inferencing provider, but now it's actually on decentralized infrastructure, which means that you have some access to like parts of the

blockchain, and then you can also still like um, you know, use a lot of onchain things. Um so it was just like um yeah like honestly this is like the this is the um the the perfect combination uh of tech. So so I was like yeah this is this is the something I want to be a part of. And then the guy sort of um how would I pos the guy sort of waved then moved along and and now I collaborated on GAA with a few uh you know sort of side projects and here we are today. Amazing. Yeah, a wonderful answer and uh certainly open- source AI is um is pretty wild like once you start actually like working in the working in the space and like exploring all the different

tooling and um open source versions of closed source software that folks are making. Um it's it's it's really really awesome. Um basically you can you can run a lot of the AI um like the core set of AI um either locally if you got the hardware or um if you don't have the budget for the hardware you have these decentralized um uh AI networks like like I and all the other ones um to to kind of help you out. Um and um I think this is a good segue uh to get into the uh content you have prepared and and basically give us a a rundown of what um all the all the neat features that guy packs in. Um so uh Toby, should I uh bring up your screen or

uh wait one second? Sure. Um uh yeah. Okay. All right. And uh folks uh folks watching us live um if you have any questions for uh Toby or Meowi as they're going through this content, please post and we will bring that up and get that answered. Uh stages. Okay. So why don't we go to our network map first cuz we recently hit like 600k nodes. Uh yeah, for sure. Network map. Yeah.

It's coming up. Yep. There we go. Yeah, we recently hit like 600k nodes and like throughout port over Yeah, we're going to hit maybe 700k soon in this. Yeah, 600k. Yeah, 600k nodes. It was like 611k like yesterday. 611k. Yeah, now it's 659. So, I'm guessing 700k is going to be super soon. So everybody's actually like you know you know putting more knowledge to the um to the network uh people deploying know it just makes us uh sort of see um you know that momentum just gives us the uh the energy to know and and it gives us that assurance to let us know that okay people actually

value what they see in um you know what they've you know gotten to experience using the centralized AI. So um definitely a motivator there. Awesome. Now let's dive into our dogs and uh let's talk about what we can do builders can do with Gaia and yeah awesome do you want to start? Oh you can go ahead. Okay, so as I mentioned earlier guys, decentralized computing infrastructure that lets everyone to create, deploy or scale your own LLMs and you can use that for agents or for anything. So for me, what can I do with Gaia? I can go ahead

run my own LLM node and I can train it with my data. But for example, if I'm running a really smart LLM, it doesn't is it's not going to have some context unless I give this LLM some knowledge base, some rag models and then it it has all this context and it can go through it. So if you click on go uh for devs uh topic yes and then uh setting up your own node is the way to go where we have some system requirement we can ask for like 16 16 gigs RAM at or if you don't have any GPU or like it's just CPU uh you I think like can we pull out our

node config repo first and I just want to show all the llm uh that we have let's see node config yeah so this is like all the llms that we have so far so now there are all the time we get questions from dev like which node we we should start running so let's say if you have like a computer that has no CPU or no no GPU but only CPU we suggest you to run like maybe smaller models like really small like 3 billion parameters or something. But if you have 16 gigs RAM or like any Mac M1, M2 or anything, just go ahead and run

something like 8 billion parameter models. So before we jumped onto this call today we were helping a dev to do build his like own uh smart wallet application and he was facing some problems with tool calling with one of his application and our main suggest was to run this llama 8p rock tool that one tool call yeah is uh so this so also like there are sometimes there are some dev who goes who doesn't know this system requirement and jump straight into 70 billion parameter model and that sometimes shuts down the whole system. So yeah, be

careful like I I never run any any par LLM that is over 8B yet. I've only run like 8 billion parameter module. So that is where like if you have like 16 gigs RAM and a MacBook I would suggest running. But yeah like Llama is really good. One of the best lm that I have been running a lot. There's also like Quen, there's also like Mistro that is good for tool calling and of course there are dips and bunch of others. We'll share all these note config link uh with nerp so that anyone can take a look at that. So yeah this is one thing and uh now let's go back to our docs as we have already talked about uh all the LLMs. Let's see uh

docs. Awesome. So as you know as you already have these options now you can go ahead uh there are three ways there are only three step for you to run your own node this just you just install the gy cla you pick the model and go ga in it and then it's just ganet start that's it and to stop it it's just ganet stop it's just only three step and uh if you want to update your like knowledge base there's option for customize your Gaia node that's how uh you go from It's over here. Yeah, exactly. Customizing your guy node. So, but let's say that you want to give your node we we kind of give it like some pre-nowledge base just to get started

with. But let's say you're like, "No, hell no. I want my node to be clean. I want my note to be a super smart freaking intern about web 3 and define. I want to do my own thing." You can go to knowledge base. Uh Toby, can you click on that on the left knowledge base section? Yeah. Yeah. Then you go to knowledge base and create your own knowledge base and you config your model with that. But let's say that uh yeah you can u to create a knowledge base. There are like multiple things to consider. You can start your you can like update the rap prompt with txt file maybe a markdown file maybe PDF maybe CSV whatever works

best. So it's kind of up to you. So and we already have videos and like we kind of try to keep all the clear docs for this. We also have a section for uh all the examples and like um what is it called templates. So we kind of keep we kind of keep building this agents and templates. So to showcase like our guys capabilities and um yeah we also have this live stream every week or something where we build agents life to showcase like how guy could work and all that. So yeah uh this is like all the minimum way to explain what we could do but also let's say you are a dev you don't want

to run your own node you want to try out Gaia for that case we have our public domains you can get Gaia API key which is if you have uh you you have to connect your wallet and like you will just connect your verify that you with your email we'll give you 10,000 credits for that yeah I I think Toby is kind of showing that like how that would work. So we'll give you over 10k credits which you could use and for your application. So when you run a Gaia node it comes with this uh link uh let's see yeah there is your API key you can apply for developer free trial and we'll give you

10k credits right over there and you could use that. So you might be thinking okay Gaia is difficult might be difficult to you know like configure into your application. This is the fun part when you're using Gaia you don't need to have some different sets of code or anything. You just swap out your just like how you're going to use open AI config like instead of the your URL you just put gas URL and then for the API key if you're running your own node you don't need any API key. If you're using one of our public domain, you need some API key just just so that no one like abuses that. So yeah, so this is like

one stop all the explanations. So I'm open to questions. Uh so if we go back to the nodes um section for people who want to run nodes um are they only able is the node only capable of running one model at a time? So it depends on I is it one more I think there's a option to run multiple node but uh it's there's a problem with hydro requirement most people do not meet the criteria for that so we always love to work with the multi- aent models and all these but it's the hardware that kind of stops us. Yeah, it's always it's

always the hardware and like and when you were saying oh uh people want to jump straight into the 70B models like they don't realize you need like quite a bit of uh RAM and or VRAM um to to run that um even even on a local machine. Um yeah, so uh yeah, decentralized um compute networks like this are are really um really helpful for for folks who don't have the money to spend on um like three or four 5090 4090 cards just to be able to run a a 70 billion model.

Um and uh as as you mentioned uh you also make it accessible through an inference API. Um you mentioned that uh people get 10,000 credits to use against that. Um uh is there also a paid tier if people go over that or or how can people basically pay for more uh more usage? So we are not uh currently there's no way to pay for any so we is we are not monetizing that. So we are mostly focusing on running the node. So there's some fun part that uh Toby do you have any node running currently?

Um I can stop my local node. Just give me one second. Okay. But you can continue while I stop my note. Yeah. So there's fun part is that when you run your own node, it gives you a link which even I can interact. For example, if Toby is running a node, he can give me the like link and I can interact with it. It's it's not only local. It's it's not totally local. Even you can interact with it. Anyone can. So, but this node address kind of comes with uh I think there he goes kind of comes with this string of address just like wallet address. And that's that's the where's the fun part is that we have Gaia

domains which lets you uh just like for wallets we cannot identify whose wallet this is unless we have some ENS or something. That's what Gaia domains are but for our AI nodes. Gotcha. Gotcha. And we have we have multiple domains listed on our website that anyone can go and try it out. a lot of people are running. But yeah, I I want to bring up the um the chess the chess domain Meowi and I worked on for our stream that we did on Saturday.

So like she was saying um you can actually totally run any of these nodes um on your own. You can even deploy them as domains and um the fun part about domains is that you can have a domain that um you know sort of acts at an interface for your node or acts as an interface for a bunch of or a cluster of nodes. meaning that um with that with a specific domain you can um create like a sort of knowledge expert or a sort of um you know super um contextualized LLM um embedding for any model or that we have currently running on GIA and what I mean by that is that say you um are somebody that is trying to you know you have your company's products currently running

maybe you need like a sort of AI agentic um sort of layout to your product you can come to Gaia um make your own domain. Um deploy your own guy domain. Uh give it a ton of uh give it a maybe a bit of uh context here and there. Have a bunch of nodes running there and you can sort of have this um domain serve as the you know source of truth uh you know for your product. you can integrate that domain in any products you want because it has an LLM that runs with it and it also has your embeddings that you've created like Maui mentioned earlier to um you know sort of grant you that um ease of access you don't have to always um train every model every time whenever you're trying to use this in the product just you know one API call um because you already have

that sort of backend service that is powered by GIA like like she actually was saying uh so uh you can share my screen Uh share again. I don't think you're sharing. Okay. Uh let me share this. Let me just share this window first. So you want to continue. Okay. Yeah. So yeah, this is like we Okay. Give us the link so that we can interact with it. Okay. Awesome.

Uh there's the link or you can click on that and show us. So this is like we kind of customized one of our LLMs to be a chess expert. So this was for one of our like chess agent that we built live the other day. And uh if he gives us the link, we all can interact with it. It's it's just I think the demo gods are trying to get at me. It it's it's almost a tradition that has to be uh this is uh how do I what can I change here?

Same URL. Uh let me stop and start it again. Could be something. Um so yeah, you can just stop your notes. Uh you can continue. I'll just stop sharing. Yeah. So I think like Jawe you see when he started the node it it comes with all the strings and then Gaia domains ga dot likeet.com. So when you have a domain it can be okay devantel.gaya domain it can be just one source of truth with this and like anyone can identify that you can have things private if like living on your place with data safety and all of that.

So yeah, I think that's the whole point we were getting at, but yeah, if you have any other questions. Awesome. Yeah, uh it looks like um uh there's a lot of great uh content you both um have prepared for a lot of devs that might be watching today uh who want to get started with um Gaia. Uh, I guess from um all that you've seen, are are there any particular AI agents or or any applications that have been built with a Gaia that have kind of caught your eye like, "Oh, that's cool." Or um maybe something you wish people would build with it.

Yeah. Awesome, Gia. Yeah, we have this ripple for all the cool projects that comes within. Let me share the link with you which is called awesome guya. So all the cool projects that people are building in the community or us building we add into it and um I think one of the greatest cool project that I have seen when I before I joined Gaia was this search agent that searches and like gets all the answers and then you kind of like get all this. So it was kind of before like the deep research came into the picture. So this is kind of fun to see from that point. And then there we had this uh what are the other other agents that we have bunch of agents we

worked on. So Toby, do you have any other favorite? Uh I think um uh PR review um A research was definitely very cool. So research add this kind of um uh perplexity kind of um vibe where you can uh you where like you have an inferencing layer that you know behind that Gaia is powering uh somebody called um a very great developer GU worked on this. So he sort of created that sort of um GIA layer that powers this and then he added some tool calling capabilities to the agent such that um with GIA you can also use G inferencing to get that information you know pass it and digest it then give the

user back back a response and on this note um this is also a very nice segue to mention um the integrations that Gaia has um there's integrations with Eliza currently integrations with um lang chain integration ations with um land graph uh integrations with um cursor even your your ids you can integrate GA there you know instantly and you won't have to um do anything too crazy because GIA has an openi compatible API eliz also was built I don't know if you saw you used at the time when it was like very live and it was live then a lot of people were using that was GIA powered so um uh yeah there's so many of these projects that are up here on like um the

awesome GIA repo that people can come to check out and um you know test uh agent kits also from CDP and DTK there's several other integrations uh uh that Gaia currently has at the moment and um um projects that you can you know build and scale using GA's infrastructure. Yeah. Yeah. Also like this is maybe one of the perfect time to mention uh Gaia node currently have MCP support now. So you can also have MCP to do bunch of work like cuz MCP is like super this layer cuz we don't have bunch of tool calling you just go ahead it's just it's also open MCP it's just using that with your node and giving this aentic

capabilities to do bunch of stuff with it. So just wanted to mention that I was going to ask you about it but that's good. Yeah that's good you brought it. Awesome. Yeah. And um I guess for people who um might be starting to learn um like AI and Gaia or or what have you um for the first time? Like what would you recommend um them as a good starting point to start trying to use Gaia? Um like is there a particular hello world template or something that you you can recommend?

I think I would highly suggest is to go through our doc and trying to understand how you can set up your own node because just like we mentioned guys is all about opensource decentralized AI or MLM that you can use and like into your application. So try to see how you can have set up your own node, have your own perfect intern with it, what you want to automate or a identify with that. So I would suggest that's how to get started with.

Awesome. Awesome. And yeah, um folks who are watching uh this uh today or whenever um we will have links um for all of that uh in the description below. Uh so definitely uh you'll want to check that out. And uh I guess uh before we we end off um I just always curious to hear um from from the folks who are building in the space like um from both of your perspectives uh what do you think are some of the most interesting and or important trends um that are being built today or will be built um at this intersection of of crypto and AI.

I think we can all agree that MCP is yet to stay. So that is one of the thing uh because when you see this one tool calling for this other tool calling for this and then we have this MCP doing bunch of layer layer for bunch of things this is something here to stay also I'm a really advocate for zero proof knowledge GK ZKP and stuff so I think these are the things that are here to stay and you know just all general web three stuff I'm always advocating for web Yeah. Yeah. Also, um um aentic networks um uh what do I mean by that?

Uh think um decentralized networks that that are indexing um AI agents. Um AI agents executing on task, AI agents delegating tasks to one another. Um AI agents paying each other for tasks that that they've worked on or something. Sort of like an internet of agents. Uh that's something that I definitely see um you know web 3 vibing towards also cuz um there's a lot of protocols in web 3. There's a lot of um SDKs. There's a lot of um projects building out there. Uh but a lot of a lot of these projects could also be um themselves just have an AI agents that sort of executes you know what the project itself does and um having these agents put on a network um with that is

powered by say something like GIA for as an inferencing layer is something that I see um uh you know part of the community uh drifting to um in terms of you know web 3 and um yeah decentralization also like I think we can all agree that web UX still kind of on a difficult frustrating position. So it's not that we need really cool UX or something. Maybe the agents could fix those and for the mainstream. Yeah. Exactly. Yeah. Exactly. Just get just type, oh, I want to get started with Ethereum. And the agent makes your wallet you funds with your credit card. Bang.

Ready to go. Exactly. Yeah. Uh yeah. Yeah. I'm wait I'm waiting I'm waiting for uh for that moment uh because it doesn't look like anybody wants to to solve the UX problem in crypto. Uh but anyhow uh that is a show for uh another day. Um but uh with that uh unfortunately we've come to time. Uh Toby Mowi, thank you again so much for taking the time out of your day um to give us that wonderful overview of Gaia and uh looking forward to seeing people build um their own projects um using that. Thank you so much for having us. Super exciting. Thank you for having us. My pleasure. And before everybody leaves, um those of you watching us

live, I've dropped a link to um in the chat uh for you to go claim your uh live attendance badge um through bit badges um for watching us live. Um so if you're into uh digital collectibles, um definitely check that out. You'll have about an hour or so to be able to claim that. And uh with that, I just want to wish everybody a very happy Friday, happy weekend wherever you may be. And we will catch you back here for another great episode of Devontail next week. All right, all have a good one.