NEAR AI Ecosystem - What Did You Ship This Week? #7 by NEAR AI & NEARWEEK 🚢

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:16

   Near AI, what did you ship this week? Hello everybody. Welcome to episode seven of What Did You Ship This Week? Uh my name is Cameron Dennis. We are, we do this every single Thursday at 11:00 a.m. Pacific where we really just want to create a space for the near ecosystem to share well what they show this week.

2. 0:36

   It's pretty pretty straightforward. Um, we have an amazing set of speakers today. Uh, Jordan from public AI, Chris from Verita, Matt from Asimov, Zahed from Jutsu, Matt from Proximity, Charles from Divvy Agents, and then at the very end I'm going to give a little bit of a lowdown on what has been going on in the near AI team. So without further ado, uh going to bring the first speaker to stage. Jordan,

3. 1:02

   welcome. Great. You are muted. And if you unmute and just quickly explain a little bit about what public AI is, introduce yourself a little bit, and then spend about seven to eight minutes or so talking about what you guys shipped this week. Absolutely. Thanks for having me. Appreciate it.

4. 1:20

   Um, so what we're doing at public AI is crowdsourcing data for AI training. Um, I am the co-founder and CMO of Public AI and our that's like the the what we're doing is is crowdsourcing AI data. Um, and the why we're doing it is because we w we want to see a future where we have um humans actually being able to make their expertise and experiences and everything it means to be human

5. 1:54

   equitable. So to do that, what we have is basically a crowd sourcing platform where as our clients come to us um with specific needs to improve their AI models, usually they've done an LLM experiment um or something like that and need to get it up to production grade in terms of reliability and cost effectiveness. We're able to go and source the new ondemand data that's required to get that to production

6. 2:19

   grade. Um they've already scraped the internet for what's available there. um or through synthetic data that's already been trained on the huge Reddit, Google, etc. data sets and they need fresh data to be able to get their models performing. And we're able to go do that. We have close to a million data builders, we call them, um contributing all different modalities of data. And then when it comes to making it equitable, that brings us to what we ship this week. And let's see if I can share screen.

7. 2:56

   do the window where we are paying people out using um stable coins and points. So here we have a recording of what it looks like to sign on with your near wallet and after you've earned on the platform can go in and claim your earnings and get paid for both the uploading of high quality data and verifying each other's data. So because we have so many data builders showing up like I said close to a million total accounts right now um when we go and run

8. 3:37

   a campaign generally some fraction of those show up like say 20,000 people come and are actually contributing data and then the next video walks us through what that looks like to contribute data. Currently we're doing a campaign where uh everybody is recording audio samples in their native tongue. So this shows the process of what that looks like in our data hub app. And as everybody is coming and adding data, um we say read

9. 4:06

   the rules, not everybody always reads the rules. There's going to be a great diversity of people that are contributing. Um that's why our platform is valuable as we've got people on every continent, different demographics, um all of those things, but then we also get a very uh high high diversity of quality in that data. So as another step after this uploading process, we use the

10. 4:30

    web3 primitives such as staking um onchain uh reputation and then uh voting to be able to get high quality data out of uh the data sets that are being uploaded. So, um, I showed you how you can earn on the platform, what it looks like to upload, and then let's get into an actual live demo of what it's like to vote. So, um, the voting process inside of Data Hub after you log in is to go and look

11. 5:05

    at all of the different campaigns that are going on. And the deal with voting is you don't actually need to be an expert in, for example, speaking Chinese or one of these other languages to be able to vote and say whether or not the the data sample is good. And that's what we're doing with the the verification process is just making sure that the AI will be able to use this data and that it'll be meaningful for AI training. So, the requirements are that the voice is

12. 5:40

    clear, that it's natural flow, so it sounds like an actual speaker, like it's not a robot voice. It's not just an AI recording of a voice. Um, that the volume is okay, it's not totally blasted out or too quiet. Um, and that the voice is steady and and smooth. So, basically looking for a native speaker of the language and that the audio quality sample's okay and that it'll be useful

13. 6:02

    as a sample. And that's what we're voting on. Not voting whether or not we like the content or not. Um or if we understand it. It's more just to ensure the quality of the data. That's what the voting is about. So to actually go in and do that, then we hit start. And then we can listen to the data sample. Not sure if it's going to come through on the recording, but if you can hear it, this to me I can't hear it unfortunately. Okay. Well, it sounds like an actual uh Mandarin speaker uh after playing the audio file.

14. 6:37

    And if it didn't, then I would say no. Like if it um was all distorted or it was choppy. But it would sounded like somebody just saying a natural sentence. So we go ahead and confirm and then you get your notification that you voted and can go on to the next question. So that's what we shipped this week is um both the ability to claim with near um your near wallet both stable coins and points and then uh we shipped this new uh data

15. 7:11

    collection campaign. So we're looking for people speaking all different languages. Uh we currently have 10 languages on deck. Like if you go into the DAP at um beta.publica.io, io. You can see if your language is listed there and if not we will be cycling through 44 languages in the next couple weeks. So your turn will come up. Um join the telegram and discord to connect with the community of other experts and then be sure to give us a follow on social media so you hear about new earning

16. 7:42

    opportunities um and everything else exciting that we've got going on. So that's it. Thanks for having me on. This is this is amazing. Um curious are is there like a premium of like un like you know rarer languages like if you Absolutely. Yeah.

17. 7:58

    A lot of people when they were um coming to the campaign for the first time were like hey I I speak this niche language in Africa it's it's very regional and um should I put down English because I know that English is a popular language or should I put down my regional language and 100% regional languages is what we're after or what the the client is after.

18. 8:18

    And we see this as being like a huge opportunity. Um just data we know is already valuable. This is something that you know was kind of the start of the story of of near to begin with. And um it's something that it's it's not a new insight either. Like I I came from the advertising industry where market research and so on is is valuable to this day. As long as people are spending money then human data will be valuable.

19. 8:44

    And the more diverse and niche the data is, the more valuable it would be. Um, and and through web 3 primitives, we're able to get folks of all the different walks of life actually contributing and make it equitable for them. And it's still cheaper for our clients than having to run their own like warehouses of data labelers or, you know, in cases where we're gathering new data, which is

20. 9:07

    our main focus. There's just there's no way to get people to who are working in a factory to go take a picture of a bird on another continent. That's just not possible. So yeah, totally. That's that's what we're excited about. Hey man, this is so cool. A real use case of crypto. Who would have thought?

21. 9:26

    Uh super exciting. Thank you for joining. It's always great to have you here. Um and you know, gonna bring on Matt from Asimov. Matt, how you doing? It's good to have you here. Hello. Thank you very much. Hey everybody. Uh I'm Matt from Aszmoff Protocol and uh yeah, let's look at what we have done today.

22. 9:46

    So it's been a lot of little incremental bits and pieces this week that we've been working on. Uh so let me just uh present here, share screen, and here we go. Right. Uh cool. So, um, the Azimoff protocol, uh, for those of you that don't know, we're working on a public decentralized knowledge marketplace. Uh, so the big like beach head use case we're solving for at the moment is when

23. 10:19

    you go and ask an LLM a question and it comes back with a response, uh, you have to double check that response. Um, if you were on a few weeks ago, you remember us chatting about this and uh Cameron mentioned about uh going to Thailand and asking uh Chat GBT, are there any dangerous animals here? Chat GBT saying no, you're fine. And then him finding out that uh the uh hotel had warned that um uh elephants had uh had

24. 10:45

    been trampling people. So, um, that's just a random example there, but that's the thing we're trying to solve is, uh, LLMs often hallucinate information and so we're trying to give them verified knowledge to work from. So, uh, within our repository, we've got a whole bunch of, uh, sort of packages. Uh, some of these public, uh, some of these, uh, not quite yet, but I think all these ones are public here. Uh but just to

25. 11:10

    highlight two that we've been doing some work on at the moment. Um Azimoff CLI that you can go to and Azimoff CLI is our main CLI that allows you to interact with the Azimoff uh platform and you can install that directly now using Cargo um or Homebrew as well. Uh plus we've got package managers for other uh platforms as well, Scoop, Nyx, um Flatpack, etc.

26. 11:35

    um depending upon what platform you're on. But that's your main entry point as a sort of enduser um and a producer of information to the Azimoff protocol. What does that allow you to do? Well, one of the things is to use the data set CLI which is a a addition to this. And the data set CLI that you can install again through similar means allows you to actually publish data to the um to the network.

27. 12:02

    Uh I was going to do a live demo here but was having a few issues just beforehand. So here is a screenshot showing what was going on. Um but here we have the ASMO data set CLI and we are publishing to test net um to a particular repo here two uh different um RDF data sets. So they're taken from open street map, right? So uh these are two data sets from the open street map data set uh that describe countries,

28. 12:32

    boundaries, roads, all of that kind of stuff, but in a uh knowledge graph format. So that allows you to load those up. Uh so you can use the CLI to publish them. Uh it will go ahead and prepare them, chunks them into batches, and then uploads them to the test net. Uh on the other side of things, uh we've got a few widgets here we've been working on. So we have a map widget here um that allows you to render some of the the the information uh that you might

29. 13:02

    see. So again, open street map information and a graph widget that lets you uh do network uh network graphs for example. So this is the the the graph widget running here. You can see here we've got uh an area of France here, Breton. And uh we can actually click on this and it will bring up the original open street map data. So this is just a widget on its own but this is just one of the parts that we are working on

30. 13:27

    towards uh the whole visualization of this marketplace. Uh so yes uh that's the main thing. So, ASMO CLI, Azimov data set, ASMO data set publish, and the two widgets um that we have there for rendering stuff. Just as a bonus uh to throw in here as well as a demo, I've been working on something to automatically transcribe um messages, voice stuff, and put it into a knowledge graph. And as with all

31. 14:00

    live demos, I've been trying to demo it right here to pick up everything I've just said and it's come up blank. Demo gremlins, but just to show you. So this is uh taking a query from the last uh update that I did in which I was talking about our trip to San Francisco. So this was live coding uh live uh creating a knowledge graph from that. So you can see on here bits and pieces about you know problems with LLMs, hallucinations, artificial intelligence um you know we're in the US talking about blockchain

32. 14:31

    the near ecosystem and you can then query it. So what problem is Azimoff solving and actually referencing the graph with specific relations. It comes back and says that Azimoff develops the Azimoff protocol. The Azimoff protocol aims to address hallucinations and it defines what hallucinations are. So it says therefore Azimoff is developing the ASMO protocol to address the problem of hallucinations in large language models.

33. 14:54

    So that's just kind of a demo of some of the stuff that uh we're sort of using just as a uh just to sort of test and kick the tires of some of the bits and pieces we're working on. There we go. Thanks very much, Matt. That's super exciting, man. Um how can you and public AI kind of work together here because it seems like they have a lot of people who can, you know, provide data and help address these nations. Yeah. No, that's great. I was just I was watching that just then and I was thinking we should be collaborating on this. So, um yeah,

34. 15:22

    we'll we'll uh um we'll reach out to them and and and chat and see because they've got all of that, you know, language information for example that they were demoing there um to be able to surface that through uh Azimoff would be uh would be fantastic. So yeah, definitely we'll uh we'll be in touch and that is one of the reasons why we run these spaces is like we can all actually work together to solve this

35. 15:45

    giant issue of open source AGI. So Matt, thank you so much for hopping on. Uh Chris bringing you on. Welcome Chris. It's good to see you. Can you hear me? Cannot hear you. It did work in the uh in the like green room, but let's try to uh maybe just change the output or input of your uh testing. Hello.

36. 16:16

    Beautiful. There you are. How you doing? Very good. Thanks for having me. Of course, man. So, uh just real quick, uh can you please explain one who you are, two, what Verita is, and then three, what you ship this week? And I feel like uh you're sort of a special case.

37. 16:32

    Marita's been doing a lot of stuff over the years and so it would be great to kind of maybe talk a little bit about the journey to where you led today, what brought you here today and um yeah take it away. Yeah, would love to. Thank you. So I am Chris based here in Australia.

38. 16:48

    So it's currently quarter to 10 to 5 in the morning. So um nice and early. Uh so Verita is effect started out really as a decentralized private database storage um platform. So effectively allowing users with their private blockchain keys to um take ownership and control of their personal data. Uh and of course with the advent of AI um uh you know the ability to connect your own personal data and services you use to AI uh is a

39. 17:19

    really interesting emerging um use case and particularly doing that in a in a self-s sovereign way, privacy preserving way uh where you as a user are in complete control. So what we've done is uh built out a um a whole set of uh tools for developers to easily get access to user data. And when I'm saying user data, I'm talking about things like you know accessing someone's Gmail or

40. 17:43

    their Telegram, um their Google calendar, their drive, you know, their Discord server, um Slack, notion, etc. Um, and so our vision here really is to make this data available for AI agents um, and create these sort of personalized AI agents. And so, um, I I'll show you some stuff today, but um, this is a little bit unusual. It's a bit of a call out as well to the near um, ecosystem and community because what we've built we're really keen to integrate, uh, into the near AI agent stack. Um, and uh, I'll sort of talk to that a little bit um, in a second.

41. 18:19

    So, uh let me just share my screen. Uh get the right one. Okay. So, effectively um how this works is a user creates an account. Um so VIA has its own network specifically designed for um encrypted private database storage. Uh a user then connects their different accounts so like a Google account, Telegram, Discord etc. And we have a a deep pin network of uh private um TE nodes and they

42. 19:02

    effectively synchronize the users's data. So that data is retained private. Um we no one can see that data um except the user with their private keys. Uh and then we allow thirdparty apps. So this is where you could be a builder building an AI agent on near um and a user can click a connect button and effectively authorize your agent to have access to different types of um data. At the moment um we support things like chat

43. 19:30

    messages, emails, calendar entries, um files or or documents. Um and in the future we'll also support actions. So being able to send a message, send a Telegram message, send an email, order an Uber. um you know whatever those those capabilities are, we want to have that full read sort of write type capability. And then moving forward to what we shipped this week uh we um have deployed or or um yeah released a set of

44. 19:57

    langraph um uh SDKs. So you can effectively e easily use these tools within a langraph environment. Um and the next step as I mentioned is that we want to um support integrating this with the uh near AI um agent um infrastructure. So what that would look like um is as a user uh if I sign in here, this is the what we call the var.

45. 20:21

    This is like the home for a user of their data. So I've just got a test account here and I've um connected some data. Uh if I I'm going to just jump across now to a different screen and show where is it? Here. Okay. So this is um a little command line chatbot that we've built and this is just a demo. It's all open source. Um this is a part of our SDK and effectively this is um using those langraph um uh um I guess modules that

46. 20:59

    we've developed um that has a whole bunch of tools and this is just a simple interface to interact with those. Now behind the scenes I've actually generated an API key from my um my data and I've put it in an environment variable but this is just modeling what you could also do in Near AI in terms of putting it in an environment variable and then integrating that into the agent infrastructure. So I can do something like um list my recent

47. 21:23

    emails. Um obviously with this you can connect to um integrate this with different LLMs and things like that. But as an example here um you can see it's invoked a tool and it's queried my email to generate that response. Um I can then follow it up with something like uh when is the demo day? Um, this account's a very just a test account. So, only got access to a few emails. Um,

48. 21:53

    uh, sometimes this happens because Llama doesn't always do the right type of query. Uh, it's not doing the right type of query at the moment, but that's okay. You get the idea. So, you could do all sorts of things here like list all of the crypto telegram groups I'm in or um, you know, anything that you want with with user data.

49. 22:11

    So if I jump back now to my window again. So as I mentioned this is like the vault where a user has their their data. Um it is possible as a user to come in here and click on authorized apps and you can effectively create your own um API key. Um, so you can create your own or you can actually have other applications uh request them and then you consent to those keys. So what we're looking to do next is take um allow

50. 22:48

    people to create their keys and then within the near AI assistant put them in an environment variable and then have AI agents in being able to access user data. So, for instance, you could do something like um you know, as the example before with Telegram, list all of the um uh cryptoreated Telegram groups I'm in and then do something on chain relating to those or um provide some suggestions on um uh you know, some

51. 23:14

    projects that that you might want to invest in um things like that. So, really thinking about how you can connect this sort of stuff. You do the same thing with email. you can sort of search, you know, query someone's email about all of the um crypto communities that someone's a part of or what types of um crypto projects is somebody interested in and the AI will do a lot of the the heavy lifting to kind of um answer those types of things. Um so that's that's a very quick

52. 23:38

    overview. Um we do have a whole bunch of documentation here at docs.vver.ai. Um and we do have some uh documentation here uh around all the different types of AI um queries and things that are available to devel to developers. This is awesome Chris and uh thank you again for taking this at 4:45 in the morning which is insane. Um so I appreciate that and can you just drop all these links and these docs and stuff into our shared telegram group where we can share them after? Yeah, we'd love to

53. 24:10

    mate. No problem. Awesome. And if uh you can kind of had a magic wand and wish you know then your community or anyone watching this uh would engage with Veruda in any way, what would you want them to do? Yeah, I'd love to collaborate with anyone that uh would would want to take um user data and actually deploy an AI agent on Near that's actually incorporating that data and doing some really interesting stuff with that because um we're huge fans of

54. 24:35

    NI and and huge supporters of of Near and we really see that um being able to create personalized AI agents is is going to be a huge leap forward. So we're trying to enable some of that infrastructure and we'd love to work with people in the community to to enable that.

55. 24:47

    Well, I actually think what you're building is an absolute prerequisite for people to give a about AI because if it's not personalized, then like it doesn't really matter um to like my mom and my dad. So, love it. Um and fortunately for you, we have a person who is building a uh an AI agent that can post on LinkedIn on your behalf and will need your personal context in order to make sure that the tweet the posts are actually, you know, sound like you.

56. 25:15

    And he's coming up next. And so Chris, thank you for joining. Zahed, you are up next. Um, see you later, Chris. Zahed, welcome back. I think you're seven episodes, you know, you've done every one of them thus far. It's pretty impressive. Yeah, it's it's the consistency that counts in life. So, I want to be consistent. Yeah. Thank you for for the invite. Yeah.

57. 25:40

    So, you work on a bunch of stuff. Um, you have about eight minutes. uh take it in whichever way you want. Uh but before please just for those who might not know you introduce yourself, introduce what Jutsu is and then talk about what you shipped this week. Okay. So hey everyone, I'm Zahed founder of Jutsu and Jutsu is a AI agent platform. uh and we are working with near AI to build the TypeScript SDK and our main goal is to build a uh build a developer experience

58. 26:11

    where from zero to production will be in minutes not hours. So most framework today is like a takes a lot of time just setting up the infrastructure but we want to create experience where people can uh get from zero to productive in few minutes. So we have been working on this SDK for uh one and a half uh almost two months now. Uh and I'm happy to say that now we have a AI agent uh kind of

59. 26:42

    TypeScript AI agent platform or agent SDK where you can build a uh AI agent locally and deploy right now locally the the demo that I will show but everything with with using just TypeScript. So you don't need complicated configuration or anything. You just say npm install uh near AI CLI and then it install that and you say npm create your agent name. It will create and then you say npm run it will run the agent. So I think as far I

60. 27:16

    know because I pretty much use every single agent framework out there. So I think near AI SDK, TypeScript SDK especially uh will have the fastest productivity time. Uh let me prove it to you. Uh so I'm going to share my screen uh where let me share my screen and I think you can see my screen. So this is uh this is a demo that we uh kind of instead of live coding just recording. So in in the left side you can see the CLI and in right side we are running the entire uh near

61. 27:56

    AI hub locally uh and we are working uh with near AI team to push it to production pretty soon hopefully by next week we'll have something uh but you can see all you do is near AITS create and then it create give it a name u that create a u typescript agent and in in the agent.ts PS you add your code and after you add your code it's pretty similar it's one on one matching with the Python SDK so we wanted to keep both SDKs same when you have the agent then

62. 28:34

    you can just say upload so now we are uploading locally but uh in the future you'll be able to upload that agent in uh uh in the hub so when you upload your agent that will show up in the website in the hub and it's pretty similar experience when you are building with with Python then you can run your agent uh from the hub. So let's see uh if the hub can answer uh the query user query. So in

63. 29:09

    few second uh it will come up with the answer. So one of the other thing we are doing is like a trying hard to improve the performance. Uh so you can see like this this is not a kind of a demo setup. So it's it's something that's the output is coming from uh from the hub. So right now we can create so the current status is we can create a a using one command we can run a a using one command and we

64. 29:38

    can deploy. So the next step for us is like a configuring everything so that you can deploy uh from local to the near hub and then user share it with your user. So that's kind of the um SDK side update and those who know me they know that I do a bunch of stuff. So I was uh there was a hackathon uh this weekend uh in SF. It was about uh generating videos. So how do you generate a video? So some of you know

65. 30:14

    that I have this I worked on this uh platform called post where you can create a LinkedIn post um easily with um just just providing a prompt uh and then it will produce a LinkedIn post and some of you already know also like our platform produce really realistic images and videos. So some of the images and videos are realistic. So these are all autogenerated video and then I thought like it will be an interesting time to kind of a figure out like a if I can

66. 30:47

    take this video generation to the next level. So allow me to introduce this uh these videos. where this whole videos are generated using AI. The caption the story line caption and everything is generated by uh by uh AI. So I know that uh in homebrew they organized a vibe coding uh kind of a event. What if like they could also add a video without any any kind of a work? So this is I don't know if you can hear this video. In this

67. 31:22

    video, we're diving into the world of Vibe Coding, where creativity meets technology. If you've ever felt lost in the coding jungle, stick around because we're about to unlock the secrets to coding with style and flare. I don't know like if you hear the kind of the audio, but like the audio and the tired of boring drinks. Let's shake things up.

68. 31:47

    Today we're diving into the refreshing world of Spin Drift, the sparkling water that's anything but ordinary. Get ready to discover why this drink is making sparkling water and real fruit. No artificial flavors, no added sugars, just pure delicious. You get the idea. So like a what is possible is kind of extraordinary. Now like a with um let's see if I'll stop the sharing.

69. 32:18

    So now imagine like a if you want to create a promo video for something for your product launch or maybe you are running events uh you don't have to do anything. So from like a just a simple topic or a prompt we can gener generate a story line. we can generate all the video stitch together the video and then the caption uh and the voice so all in one place so that's the future we are going uh after and yeah privacy is also

70. 32:50

    very important so like um uh that's something we will work in the future but right now uh we are using um LinkedIn API so your data we don't host your data we don't keep your data so if you want to give it a try go to post.ai um and give it a try. So like all I'm asking at this point is give feedback. That's it. Thank you.

71. 33:15

    Thank you so much. Um we've gone we've come so far like since a year ago when I remember seeing like a very similar like AI generated ad where everyone's faces are merging together and the hands are you know doesn't make any sense but this is really cool. Um so people could go check out post.ai but where do they actually give that feedback? Is there somewhere in post.ai AI where they can provide that feedback. Ah that's a that's a good feedback from you. So

72. 33:40

    thank you for that feedback. So yeah right now just DM me in telegram or or just send me a message but this is a great feedback. So this week we will add a feedback form or something that people can give feedback. Thank you. I got more feedback for your feedback and that is what tell what's your telegram or you know open up your Twitter account so people can actually like send you notes.

73. 34:03

    Okay. Yeah. So I can I can share I can I can share it with you and then you can share uh with everyone. Perfect. Awesome. Zahed, thank you. Hope to see you next week. And up next we have Mr. Matt Locker. How you doing, Matt? It's good to have you here. Hey.

74. 34:20

    Um I'm good. You You got me like I'm coming in I'm coming in off off of some severe like release PTSD here. Uh, so we had a really we had a really fun time releasing I know I got seven eight minutes. Yeah, you're you're pretty good, man. We It's pretty fair. Just let it just let it run. Yeah, just let it run. Charles, you're up next. We're coming off a very exciting uh release.

75. 34:47

    We released an agent for um for base names. Um, so these are basically ENS uh names that you can get on uh on base and um what we thought would be interesting is that you know I I really wish I I can't get through to x.com right now. So do do you mind sharing your screen Cameron and uh yeah sure I can do that.

76. 35:12

    For some reason it's just spinning for me. So, um, I'll let you know if it clears up later, but I just wanted to show a couple of a couple of accounts and interactions and stuff. All good. So, essentially, um, uh, let's let's check out Bankerbot if you can pull that up.

77. 35:33

    Bankerbot Banker with just bank R then bot. Oh, bank R. Yeah, that guy. Little TV one. Yeah. So, a bankerbot, really awesome, uh, bot created, uh, by this guy, Zerox deployer. Um, it's an awesome concept. So, basically, it gives everyone with a with an x.com handle, uh, an EVM address, basically. And with that address, you you can go to the Bankerbot website and use this thing called the terminal, or you can just tweet at it. So if you click replies for

78. 36:06

    bankerbot, we can check out the the terminal is not the the most exciting part of it, but if you look at replies and you scroll down, you can see uh some people there we go at bankerbot sell 0.01 ETH for uh you know this other token. I'm I'm presuming this contract address on cow protocol. So you can even just put in a contract address. It'll figure out how to do stuff. Now um right

79. 36:32

    now it only kind of sends ETH. It swaps tokens um and you can choose the protocol you want to swap on as long as it's on base and it'll also issue tokens so you can get it to like launch new tokens. Pretty cool stuff. So we thought uh with shade agents it would be cool to create an extension for something bankerbot wasn't doing. So if you go to uh just at base names based names uh base names.

80. 36:59

    Yeah. So this is our agent. Uh and if you click on replies here, um you can see that basically uh people are like if you go to Kendall's uh hey at base names grab Kendall. Yeah. So uh basically we can now uh purchase these base names. Now what's really interesting is that you know our bot kind of kicks off the flow says hey I'm on it. And what we do is we use near chain signatures to create a

81. 37:29

    one-time deposit address where users can send the funds to that address. And that address is specific for their ex uh user ID plus the name that they want to register. So basically it's a it's a it's a complete one-time address for just that name registration for just that x.com user. Um now that's their custom onetime deposit address for that name purchase. So then uh Kendall tells Bankerbot, "Hey, send the above funds." Bankerbot sends the ETH to uh to the

82. 38:03

    one-time address and then our bot goes and looks up the uh deposit. And you can see at the bottom it says done. And then it also um in the case of kind of overages like uh in case there was extra ETH left in the account, um it'll automatically issue a refund from the onetime deposit address too. basically leaving nothing but a tiny bit of dust in the account. So, we basically have

83. 38:28

    created like this Bankerbot extension bot or another bot. Um, you don't have to use Bankerbot, but the flow with Bankerbot plus base names is now super seamless because you do not have to leave X.com at all to register base names. Um, so that's the good that's the happy path. That's what that's what happened, you know. That's that's what we say what happened. But obviously when we release something like this, we get all sorts of crazy stuff that goes on.

84. 38:59

    Um I won't get into Yeah. Yeah. I kind of want to know. Well, I mean like there there's some really fun stuff that happens. So for example, like um we were using like we're using a bunch of different technologies here. We're using near chain signatures. We're using the Twitter API. We're using uh base scan which is basically ether ether scan for bass. We're using their API. We're using uh we're using regular uh base I think

85. 39:27

    it's like public node RPCs right so if you put all these things together you have a lot of things that can go wrong basically you know exceptions can be thrown and you got to handle like oh okay so I get an exception how many times should I retry this you know because retries obviously would count towards rate limits and you know you're rate limited on various APIs especially X right so we did run into when we first launched, we ran into a X rate limit.

86. 39:56

    Um, so we just had to increase the account uh and pay, you know, the the Elon tax. I hope I don't get shadowbanned for saying that, but um yeah, basically, you know, you you just gota you just got to handle these things as they come. There were some really interesting edge cases from users, though. So people were sending base ETH from smart contract wallets which was causing basically an internal transfer instead of a normal transfer of ETH. So we kind of had to handle that edge case as well. So there were a lot of things

87. 40:27

    that we had to jump in and sort of uh just handle these interesting kind of cases. Um different you know different pricing for different lengths of names. Um, and then also like when we originally launched, I think people people were trying to uh say stuff off of replies. The bot was getting confused. It was it was like double it was like double responding to uh to replies and stuff like that. So, we had

88. 40:53

    to shut off all replies and quote tweets and stuff, you know, much to the chagrin of some people. Some people just they see a thread and they want to try something too, so they just kind of post at the end of the thread. Fair enough.

89. 41:05

    But I mean um we we had just launched so for us to kind of jump in and disamiguate that case of like replies and quote tweets and all that stuff. It was just easier to shut off any reply quote tweet or or or retweet. So anyway, long story short, there's a lot of uh interesting lessons learned what we're doing now. Uh because we just learned all those all those lessons in uh in

90. 41:29

    sort of um it was like a baptism by fire kind of thing. like we we really got uh we really got scorched on a few things and but we we're taking those lessons and we're going to uh basically distill those down into the repos for the the base names agent uh which is all public.

91. 41:50

    It's all out there and open source. Um and again this is built on top of shade agents. So, it's built um it's built on top of the shade agent stack which is a completely non-custodial um and and the worker agent code that talks to X talks to the EVM uh you know prepares the transaction and then eventually calls the smart contract and gets the signature that code base that's the the worker agent kind of brain that's verified. Um, so we have

92. 42:23

    blown away the key to the near smart contract. There really is no way for us to change that code base on people. Um, and you know, it's it's uh it's a truly non-custodial agent just kind of floating out there in the cloud. Now, we're using Fallah as and they've been a great infrastructure partner and and the code for the worker agents deployed in a trusted execution environment. These are great guys. um they've been awesome with

93. 42:50

    supporting our work with shade agents and if the if if the infrastructure like if if the agent were to have a problem or kind of get stuck or there's some wild exception was thrown and it was to be thrown offline, we can redeploy the same code, but we can't change the code.

94. 43:08

    So we can redeploy the agent, boot it up again and and sort of put some funds into it and get it going again so it can make those um those near chain signature uh requests like to to get to get EVM transactions signed, but we can't change the code. So we can't change what it does. So it's just doing it's doing what it does and it's going to keep doing what it does. And as long as we keep paying the x.com API fee and uh and we keep uh you know we keep the we keep the

95. 43:39

    fall instance online then it's just going to keep on going and the near smart contract obviously just sits there and uh and just expects that that a worker agent um that's currently deployed on FA uh talks to it with that that verified codebase. So yeah, so lot lots of stuff going on, lots of moving parts, but we're we're still we're working very hard on kind of distilling a lot of the shade agent learnings and knowledge um in getting that out there for people. And uh yeah, we're in the

96. 44:10

    works. We're working on some small mini hackathons to try to get people interested in in sort of building agents like this. Um so just look for that in the future. Hey, uh, how do I top off my like top off my account so I can press post and it'll buy for me? Uh, how do you top off your account?

97. 44:31

    Yeah, how do I give the the actual like agents the crypto to get me the name? Oh, so f first you'd have to post this and then it'll respond. It usually responds in like a minute. Um, so it'll just take a sec and then you'll see a reply to that and then um and then and then it'll tell you which onetime address to send the funds to. Um, so I don't know if if you're not using Bankerbot that that's fine. uh you can still send base ETH to that address and

98. 45:01

    then uh and then within I don't know within 30 seconds or whatever of you it depends because it sort of it has to check for the deposit then it goes and it prepares the transaction gets chain signature comes back broadcasts it and then you get a you get a success and it kicks off the uh I think it it responded to you already.

99. 45:25

    All right. Yeah. So, I got to do the next 10 minutes. All right. Uh, actually, I want I want Cameron near Cameron. Okay. Um, I will Yeah, you can. You don't have to. So, you don't have to send to that address. It'll just eventually check for the deposit will just time out and then um it gets it gets kicked into a refund list and basically if you send way late uh there'll be an automatic refund kicked

100. 45:55

     out. But um but yeah, we don't we don't kind of keep you in the the live flow for very long. Cool. I'm doing this. Um so yeah, that's the story. We're really excited about how this went though. It got it got a lot of um it got a lot of attention, got a lot of eyeballs. Uh people like the flow. I think you're going to see more agents like this. Um, yeah, and and uh like we said before, uh I was on here a couple of weeks ago, I think, for the hot wallet service and things like that.

101. 46:27

     We're still looking for teams that want to uh potentially use shade agents and chain signatures to create kind of a universal uh you know, wallet substrate for for various web web two kind of services. So, um, for example, like you know, C like Cameron, your your your X handle would be used for, you know, maybe one wallet on Bankerbot, but then if there's five other alternatives, you have different wallets, right? Um, we're trying to kind of see if we can maybe

102. 46:59

     tie those together. Um, at least for some some of the bots um so that developers don't have to like reinvent that wallet infrastructure every single time they want to make a bot that does different things. Um, yeah, and other ideas, hackathon stuff, uh, coming up shortly. Um, that that's really all I got to share. Um, but we're really excited about it. We think it's pretty cool stuff and, uh, we're hoping that,

103. 47:23

     um, you know, that people want to hack on shade agents. Um, there's there's more you can find out for shade agents at just near.aishade. Beautiful, Matt. Yeah, thank you. Um, we'll be sure to share that with the ecosystem. Um, is there anything else that any anywhere that people can go learn more about you, learn about what you're constantly shipping.

104. 47:50

     Well, yeah, like I'm on I'm on I'm on X. Just Google Google the name and you'll find me. Uh, and then also, um, just follow, you know, follow what, um, I think it's ProximityFi is doing on on X because we we got a great, uh, you know, we got a great marketing guy. He's he's pumping out lots of stuff. Um, Vashal does great work and and we're we're plugging into a lot more other ecosystems here. Um, you know, some some people maybe that are that are sort of

105. 48:20

     in the the near ecosystem might say, "Hey, why are you doing all this stuff on base or this or that?" Well, we really just want to show off chain abstraction and chain signatures. We want to show off the fact that we can build these agents where, you know, the brain the brain is in a trusted execution environment. the smart contract is on near. So all of the transactions are are basically a near

106. 48:43

     transaction to get that signature. Um and and that's where you also add that extra smart contract guard rails. So you can have like an LLM running in a T, but you don't know what the LLM's going to do. And if it does something really kind of silly, you probably want your smart contract to kind of add the additional guard rails there so it doesn't kind of it doesn't do anything really uh stupid with user funds. And then that's all

107. 49:07

     happening on Near, but then sure the ultimate uh the ultimate broadcasting or or maybe token issuance or or swap or whatever could be occurring on another chain. And that's what we think is really exciting about this tech is that the verifiable agent and the sort of the um the like the the secure kind of stack that we've built around shade agents is a near stack. Um, it just happens to have the added bonus of being able to execute on any chain.

108. 49:40

     Yeah. Um, everyone asks me like, "Oh, you're doing all this multi-chain stuff." And I'm like, "Nir is the new internet." The blockchain was just the database to maintain state. We checked that box. The second box is all chain abstraction, a prerequisite for, you know, the user owned internet to actually work because users don't actually give a if their app is running on AWS or Azure. They just want something that's easy to use. And why do

109. 50:03

     we care about all this? Because AI is the future. And AI is not going to care about what blockchain it's interacting with. So, um, that's sort of my framing of it all. And it it it just makes sense. I, um, you know, poo poo to those who, uh, you know, saying only do stuff on here. That's kind of silly. Um, so Matt, thank you for joining. Um, gonna be bringing Charles on. Charles, welcome aboard.

110. 50:30

     Oh, can you see Charles? Hey. Yes. Hey, thanks for having me. How's it going? Thanks for joining. Um, quickly just brief intro on yourself. Uh, intro and what you're building and what you ship this week. Sure. Uh, I'm Charles. I'm one of three contributors at Benevio Labs, uh, including my co-founder, Jacob Null, former guys from Pagota, and we are focused on building DeFi apps for mainstream consumers. And right now we're working on a project where uh we have this idea for an overall app experience that we want to

111. 51:02

     build but as we're working towards that we want to release core experiences of this app as soon as we can. So we're one of the things that we shipped this week that I'll share is the the designs for um the the divvy agent that we're building. Uh these are the overall designs but we're also using the interface that we don't have to build that already exists the chat interface with near AI. So let me share my

112. 51:37

     screens. Yeah. So generally we want to onboard mainstream consumers non-cryptonatives so that they build comfort and starting to build wealth in digital assets and we want to allow them to use automated conditional swapping of tokens to stable coins as a means to regularly provide financial support and for crypton natives as a means to regularly trim off your portfolio as the market drives it up so that you can always buy the dip for example. Um, this

113. 52:07

     is the UI that we the designs we landed on this week in addition to the UI that we're using with Near AI. Um, and I just like how we found this kind of sweet spot between using the the chat interface and then providing additional context with these quick buttons and actions that you can do to interact with with agent. So, that's one thing that we shipped this week. And then if I can

114. 52:32

     switch and share um the terminal uh another thing that we did this week is we improved uh the LLM on near AI that we're interacting with with our agent improved its ability to match user input against agent functions. So as we're writing Python functionality to grant the LLM capabilities to do things like save user information to be able to save conditional swap conditions um how do we persist that between messages and

115. 53:11

     and just kind of found it found how we navigate between saving information into the class where the agent is written versus saving things into the file context that the LLM uses in between commands. So now when I give it different information, I previously gave it my mirror account. So now I can see what the agent is doing and see that it recognizes, hey, I should use a certain tool to fetch the near account ID. I know that I

116. 53:36

     saved that in a file and I can get that file to grab that information and then return it to the user. So now we're able to move forward with prompting the user to add all the information that they need to add in order to set conditions that would that the agent will use to autonomously issue a swap in the future. So I can tell it, hey, I want now that you have my near account, you know, all of the Omni

117. 54:00

     tokens that are in my portfolio. I want to set a growth goal of $10,000. So I want to set a growth goal of $10,000. It can recognize what functions it to use. So save that goal. I want to set an allowance goal of $3,000 such that when my portfolio grows by 10,000, I want to do a swap of some combination of stable coins, some combination of tokens, excuse me, that the agent will determine into uh the stable

118. 54:28

     coins. So, we improved the LLM's ability to recognize what functions to call. And now when you give it different commands, it can appropriately read all the different files that it saved in that info to determine how uh how to satisfy your requests. So I then asked it, okay, now that you know my goal is 3,000 in USDT and you know all the the omni tokens associated to my near account,

119. 54:51

     recommend a set of tokens and quantities that I could sell to realize that value. Um, so now we're getting these values back and one of the one of the things that we need to improve upon here is we can see it returning the value. It's calling the correct function. It's suggesting here's the combination of these different tokens to sell to realize your goal in USDT. Um, but the LLM is still not not parsing that as

120. 55:14

     appropriately to tell the user exactly the best way like what that means and still the LLM is doing it a little bit differently than just the raw output. So that's one area to improve upon next. And this all fits into the context that we want to design for trust and allow an LLM, allow an agent to act autonomously on your behalf using um in an approach that users will feel comfortable with. So, one of the things that we were thinking about this week is how do we design to improve trust such

121. 55:49

     that one would be comfortable giving an agent the ability to do this type of trades on its behalf. And the design that we landed on, we shipped a PR this week that experiments with using a proxy account that a user would have access to that they would use to send a smaller quantity of tokens to. And then we can leverage near's NPC network and near intents such that the agent can use the

122. 56:14

     NPC network to request a signature to be able to leverage the proxy contract with restricted access to certain methods and restricted number of tokens to be able to act autonomously on a user's behalf um at any time without giving an agent access to your entire account. So this is one of the ways that the way that we're right now experimenting with how to design for trusting autonomous agents

123. 56:38

     and we want to leverage NPC and near intense to do this and thanks to the near1 team and near intents that chimed in on the telegram chat because we found that with this experiment this the signature that we're able to generate from the MPC network is of a format that near intense does not yet support. So, Near Intent supports ED25519 signatures and NPC Signer generates the SECPK signature type. Um, but was happy to see that there was a PR that was merged just last week to add ED25519

124. 57:13

     signature types into NPC. So really excited to be able to test that out to be able to generate the signature type that we can use to allow an agent to act to act autonomously on your behalf with your intents and really tie together what we shipped last week of doing this all manually with uh with your intents.

125. 57:35

     Charles, this is so cool. It seems like a prerequisite, another prerequisite to uh just ensure everyone in the world can actually understand DeFi and you know not have to worry about constantly updating their portfolios. Um this is super cool. when mainnet not to be that guy uh when mainnet and um yeah I guess where can people learn more about what your like the project and you yeah so

126. 57:59

     we're open web economy on Twitter and all that's where you can see all of the everything that we're shipping including the experiments that we want to get feedback on different pieces of this app we'll launch those there um the I'll also add in the notes a link to our GitHub repo where you can build with this agent um if you're doing something you're building agents, you want to do autonomous trades. We're also doing a

127. 58:23

     lot of this work open source, so we can build faster and help others build faster as well. So, you can find everything we're doing there and maybe that helps you move a little bit faster, too. I have a lot of builders in our network. Um, what kind of builders you looking to work most closely with?

128. 58:40

     I'm really curious about people. Yeah, people interested in DeFi, people that are solving similar problems of how to bring users into this experience. um uh one person that's contributing with us has a lot of experience on the LLM side.

129. 58:53

     So, it's been really great to see his knowledge of LLM SDKs from other environments and and bring that over to the near side. But really, anyone who's building or using DeFi apps and has questions around common patterns that work in some context, but in DeFi there's there's friction like we're dealing with those as well. So, I'd love to chat about that. Totally. Um you should absolutely get in touch with the Delta Trade team if you haven't already.

130. 59:16

     Uh, happy to make intros. Yeah, thanks. Yeah, awesome. Charles, thank you so much for joining. Hope to see you on a future one with more progress. Uh, you know, it's amazing to see your journey thus far. Uh, we've been chatting about this for a bit and so it's cool to see you guys. Thanks, brother. Good thing.

131. 59:34

     Of course, man. All right. Um, I'm the last one. So, I've been focused on hiring. Uh, we are building the team at your AI, which is very exciting. uh takes a long time to get all the budgets and people and interviews and everything going. Um but uh making amazing progress on that front. Also making great progress on the auditor agent RFP. Uh found an amazing security team to to kind of take that project and move forward. Um there's a

132. 1:00:03

     lot of things that you know my my role is primarily external integrations. Um so I don't want to be the one to break the news of some of these exciting external integrations that happen. So I can't talk about a lot. Something that I've been thinking a lot about are governance agents. And I'm actually just going to drop uh a doc I wrote about what kind of governance agents I would like to see. Um because I don't know if

133. 1:00:27

     uh many people here probably aren't familiar with a very interesting experiment that uh the near ecosystem ran last year called the near digital collective. Uh this was an experiment to slowly decentralize the near foundation by handing uh funding power to the community. and we ran an open election to determine who should those people be. Um there was a lot of amazing learnings.

134. 1:00:49

     I have a whole presentation dedicated to that but I'm not going to get into but um we pretty much created this like incredibly complex DAO that uh had all these you know uh levers of stopping funding and approving funding. So the problem was that there's a lot of bias in governance and uh I want agents to help address that bias and help people make better decisions. So uh I'm probably just going to post this on Twitter uh with view access only for

135. 1:01:19

     people to check out some of these governance agents that we like built. Um we'll probably put out a request for proposal for some of them. Uh Charles and everyone else like if you guys are interested in governance and especially for DeFi protocols like might be helpful. uh and also people sometimes you know can make money participating in governance and so there could be some interesting things that we could all do together. So I'll drop that. Uh other

136. 1:01:43

     things I'm working on are like encrypted memory uh solutions for the new AI hub. So like the varita uh example like how do you personalize your agents with like it's your actual context. uh been focusing on that and focusing on getting uh compute for the new AI hub specifically H200s with TE support which are very hard to find and so uh scraped the earth found some it's been pretty great shout out to our partners who are

137. 1:02:11

     were doing that exhibits specifically um and then also let's see oh managing the Twitter account that's something that takes much more time than I you know originally thought so these are just some things that I'm working on personally, but the new AI team is shipping a lot more. We actually have next week uh Alex Koma for come back on to talk about the progress happening on the MCP side of things and then

138. 1:02:35

     hopefully we'll also get PR on to talk about the agent cloud. Um I don't want to be the one that breaks the news. It's not my news to break. So you guys are going to have to hold off and wait. Uh but a lot of things coming. Super excited. That's all I got. Um, thank you for joining and if you want to, you know, keep follow the near AI account, the Twitter account, follow your week.

139. 1:02:58

     Um, check out the near jobs jobs jobs board. Uh, we'll be kind of posting all those open roles there. And that's about it, guys. Hope to see you next week. Thanks for joining. Thanks for watching. Subscribe to Nearweek.