NEAR AI - What Was Shipped This Week

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

1. 0:17

   near AI what did you ship this week hey everybody welcome back to the second edition of what did you ship this week my name is ceron Dennis I'm a core contributor at near Ai and we're running this space every week or so uh just to give the community a platform to share what they're building uh think of it like a community engineering standup uh for those who have been on engineering

2. 0:43

   teams you know you got to give a quick quick spiel about what you're doing and this is a great opportunity for people to demo what they're building and ultimately see how everyone in the ecosystem can work more closely together we have a pretty amazing agenda packed uh today we're going to kick things off with Eric Wier from near AI who's the CTO he'll be unpacking proof of response and the cool auditor request for

3. 1:07

   proposal that we have out uh Andrew from doppelgangers will be also coming back uh second time in two weeks for those who watched last week uh to give a quick overview on what is happening in the do doppelgangers ecosystem uh we have a special guest today which is very exciting on the doppelganger side and they will be demoing uh some stuff that would probably better for them to to

4. 1:30

   flag so next we have Kendall Cole from proximity Labs he will be sharing a really cool new uh update and type of smart contract AI powered smart contract or smart contract powered AI called shade agents and so more to come there and then after we have Omar from intellect talked about agent swarms for real world use cases specifically for large Enterprises which is exciting because we need agents to do real relevant things in this world and then next uh we got hang in from fala fala is

5. 2:01

   you know building a a pretty amazing agent Cloud uh and they are going to be talking about red pill and daack and how to deploy uh elizab based agents into their te Cloud last but not least we have zahed from Jutsu talking about the near AI typescript SDK he's been making a ton of improvements and very excited to hear from him and so without further Ado Eric I'm going to bring you up on stage and feel free to share your screen just as a quick reminder if you are

6. 2:31

   sharing your screen please only share your screen and nothing else um there's a button to you know share window make sure that there's no other tabs in that window this is best practices for uh views when last week I think we got 1.1 million views because of a little mistake and so um yeah Eric gonna bring you on stage and take it away I mean I think we can top that this week with just fantastic AI content right oh absolutely yeah yeah all right uh

7. 3:05

   thanks for having me Cameron um so I'm excited to talk about what's going on with neari most of what we're building right now we're planning to talk about next week at the eth Denver conference and at the road to a trillion agents event that's happening in Denver around the same time so if you're um if you're in Denver uh go to that event go to the conference talk to us there um we've got a lot more to show that I won't be talking about today but there's a few

8. 3:35

   sort of um pieces that we've been launching beforehand in order to sort of prep for that I want to talk through a little bit of that uh generally speaking what we envision at near AI is a world in which you have uh AI agents that are acting on your behalf and those are talking to other AI agents act that either have specialized skills or acting on behalf of other entities people

9. 3:59

   businesses governments Etc and almost all the infrastructure weu build is about how do we make that happen in a way that's going to be um you know open accessible and user owned so that whenever something whether it's um a piece of whether it's an agent or an AI model is is operating on your data you know what went into that what is going to what's going to come out of it and who's going to be able to access your data um as opposed to this world we're

10. 4:25

    sort of creening into where uh where you are sending all your data out to to um at best a uh massive US tech company and at worst like Chinese government um so we're looking at how we're trying to make that happen we're coming up with a few uh interim things that we that we're coming out with one of the problems is as you get to a world where you've got an a decentralized network of Agents how do you ensure that you're going to get an answer back when you ask for ask a

11. 4:56

    question um one of the um the centralized world you just to you know try another provider and it's kind of a bad experience what we're going for in a more decentralized a agentic world is um something where there's going to be an economic incentive and a proof that you're going to get an answer back uh from the agent that it's online that it's ready to go and that's going to give you a response um we put out a paper about um about how that should

12. 5:21

    work uh I'm going to share just a window for you here oh I can share just a tab that's even better we put out a paper called proof of response uh you can see a blog post that sort of explains the concept behind it and the purpose behind it on the near. blog and then the paper itself is here on archive we also have a team that's working on um on a proof of concept implementation uh directly into near core um very exciting stuff uh if you are interested in in how we're going

13. 5:55

    to have um a decentralized network of agents that actually functions doesn't go off offline on you check out this paper give us feedback we'd love to see it another thing we're working on um when you've got these agents uh that are operating on your behalf um their bind large going to be open sourced and they're going to be working on open models and that's great but unless you're going to get the expertise to hand audit every single line of code and ensure that every um every agent is doing exactly what it says it's supposed

14. 6:27

    to do or what they claim it's supposed to do um when need to be able to ramp up the ability to audit agent logic at scale and so we put out an RFP for the development of an AI auditor agent that can go and evaluate that agents are doing what they say they're going to do um again this is on near. uh just head over to the proposals link and it's the only thing in the proposals section right now um the uh check it out um if you're

15. 6:56

    interested in building this uh submit a proposal here that goes to Cameron hey Cameron uh and uh this this is works very hand inand with tees which you'll hear more about from Falla later um a trusted execution environment says proves that the code that you're looking at is the code that was running but how do you prove that the code was running is doing what you hoped it would do

16. 7:23

    that's where this AI auditor uh will come in so we're opening this up to the community trying to get people other people to build it uh please go check it out submit an RFP um on the near AI site uh we've been shipping very rapidly um lots of little features to prep for the bigger features that we're launching next week uh a couple things I want to point out in particular um we've been uh espe uh in in conjunction with the fine folks at near Dev and at devhub we've been um

17. 7:53

    overhauling our documentation in particular uh secrets and environment varials for agents has been a pain Point uh so shipped I think an hour ago is a brand new documentation about that explaining exactly how everything works um we've made the near AI Dev Hub say about 50% faster in the last couple weeks um we have new agent template oh I haven't been changing my shared tabs wow AI auditor RFP I've been I've been staring at this the whole time I didn't realize you weren't see we can mess up our screen shares in many different

18. 8:33

    ways um a auditor RFP check it out click on proposals read it submit it this goes to Cameron you can always talk to Cameron um documentation overhaul great stuff uh especially our new page on uh secrets and variables that should solve a lot of problems that people are having building near AI agents and then of course the platform itself uh gotten a lot faster as I said and um and we have these new agent templates that we've come out with recently uh helps you just get started

19. 9:09

    with building an um also highlights a few of the sometimes more hidden features of near AI like um most agents are uh all the user all the interaction comes straight from users but you can also have the interaction come from um come from Twitter uh DMS or replies you can have that interaction come from blockchain events so we have ability to start to sort of get you started on the Twitter side of that um we have our

20. 9:32

    neari SDK which helps unleash the particular features of neari but if you want to build on Lang chain or you want to build on uh agent kit by coinbase we have templates for that too um so we to support that and shim that in this will show you how to do it um we also have typescript support uh so if you'd rather build in typescript we can do that as well um we're just sort of building out the breadth of n capabilities uh and then we'll show a little bit more of

21. 9:56

    that of that agentic interaction world that was talking about next week at e th thank you Eric you are right on time so appreciate the update um one last thing that I actually probably should have broughten up is we're running a hackathon in San Francisco called the useful agents hackathon on March 14 15th uh I will drop the link in the thread below but uh definitely check that out

22. 10:20

    Eric always a pleasure hope to have you on again in the future to consistently keep the community updated on what's happening in actually the near AI team so really appreciate your time all right thanks ceron thanks for joining and uh without further Ado we have Andrew from doppelgangers part two exciting welcome back Andrew um be careful when you're sharing your screen this time I really

23. 10:44

    appreciate that and uh yeah would love for you to give a quick Lowdown on what you've been building I know you've been constantly shipping and uh yeah take it away yeah thanks uh Cameron and uh yeah I will be sharing my screen later during the call but uh before that I want to do a couple of updates cuz probably no one remembers what I actually said they only remember my screen so um a bit a little

24. 11:10

    bit of a reminder about what doppelganger is uh doppelganger is your personalized AI uh we have shipped an a mobile app and a web app where anyone can create basically their digital twin our thesis is that the best personal assistant you can get is you basically you you are cloning yourself and uh make uh make the Clone do all the work uh at least that's that's the vision but um since we are building this on top of

25. 11:43

    near and on top of blockchain we we are tokenizing personalities uh so uh it's not only your data that uh is uh being used to train an AI uh but it's your personality too so your basically adding a human touch to the to the to the AI and uh that is actually tokenize and we are shipping new features that will allow users to share their personality traits between each other and between each other digital twins so one of the features that I'm personally excited about is being able to merge

26. 12:22

    different the different personalities into one so creating like entirely new uh fused personalities of different individuals and we are planning an active campaign on that during e Denver next week and after that with a couple of influencers and K KS from uh from the industry so stay stay tuned for that um and uh one of the features that we have shipped this week was actually voice cloning so now you can not only clone your amazing personality but you can add

27. 12:55

    a voice to it and uh I would like to bring a special guest uh uh on the stream uh that's the first person who we officially cloned entirely uh so voice uh personality so we used the all of the data sources that we that that we had access to uh Twitter Telegram and Instagram to create a perfect of my girlfriend Valerie hello hello hello everyone uh so yeah my name is Valerie uh and I'm 100% real I'm not AI

28. 13:33

    generated but but I already have my AI so I'm ux UI designer in web free and in the community I'm more than four years so and I decided to create this AI because I think people in crypto more introverted than extroverted and uh this can be helpful uh for communicate with someone have support and yeah well it's awesome to have you here um you guys have a exciting demo to kind of go through so without further Ado kick it away take it

29. 14:17

    away Andrew you're muted yes oh and I got I gotta bring your screen yeah so what I'm going to be sharing right now is a chat window uh with the digital girlfriend a digital copy of Valerie um what you're going to see right now is not scripted so uh uh it's uh it's it's going to be fully live so if anything breaks just give me a second to reload the page because it's version one but I really hope that

30. 14:51

    nothing will break we love live demos you know especially with this team so y hello how are you honey hey I'm doing great thank you how about you anything thrilling happening in your life or in the crypto realm I'm tired and I want to order something and watch movie let's do it sounds like a perfect plan what movie are you thinking of watching and do you have any cravings for a specific type of food let's make it a cozy night in let's

31. 15:43

    make it awesome what's you go-to movie genre and anything specific you're in the mood for foodwise I'm thinking something comforting like pizza or maybe some sushi I think it's a good idea let's order sushi yum Sushi sounds perfect do you have a favorite role or are you feeling adventurous and want to try something new let's make it a fun Sushi night I like California yeah I think we are running out of time

32. 16:32

    uh but this is the moment where new agent chimes in and actually orders you Sushi uh so stay tuned for more updates from UI now this is awesome um the fact that this is all live on your phone today I think is honestly like I haven't seen anything like this out yet so um I think this is the beginning of honestly the agented Commerce um it's going to start with everything being personal but I actually have a couple questions and we'll dive a bit more into the Privacy component later yeah like

33. 17:04

    how are you kind of working through the the security concerns of like uploading all of somebody's you know in sorry Instagram account Twitter account telegram messages how are you kind of managing that information so uh there are two two points here uh number one we are using x-rays uh you went live uh with them the other day uh they are using a full homomorphic encryption from the GetGo so we don't really have access to your data it's all encrypted but number two the

34. 17:38

    only entity that has access to that data is actually AI so we just feed it into AI it comes up comes back with a response of your personality so yeah uh in the worst case like and the the the the AI is actually running in the te so it's deep seek that NE is using so uh yeah your data is safe and uh the file is actually stored on your device like we we are not we're not interested in your data we're only interested in your personality and uh that's uh that that's kind of what uh what we're doing I'm

35. 18:13

    more than happy to share uh the backstory file that is created uh there is literally nothing there like uh for everyone who's afraid that there is like some personal data or anything like that we number one we are using the data filtering from the go like there is the service from azour that uh Microsoft azour that removes all of that all of that automatically like it automatically detects all the like credit card numbers

36. 18:38

    uh all the names all the uh racist language as well and it just automatically removes it so yeah your data is more than safe and uh this is uh this is not what we are after at the end of the day so yeah yeah exciting I mean I think this is going to be the first sort of agentic user facing app in then your ecosystem and maybe even crypto so oh for sure uh this is actually the first use case in crypto where it's

37. 19:07

    actual personality not a character that was created by someone's figment of imagination or some book or some reference it's legit a personal data like it's created of Valerie's messages of Valerie's Twitter account uh that's why there is so much so so much of crypto in there when you talk to a digital girlfriend because uh uh well it's crypto Twitter guys uh so yeah it's how does it make you feel Valerie how does it make you feel to be the first digital twin on toppel Gamers so I'm so happy

38. 19:42

    because I'm really wanted to have something after me so and this is blockchain this is AI this is Will exist uh like yeah I'm I'm I'm so happy to share some knowledge and uh I don't know support or growth tips because myi talking about that stuff yeah your your legacy lives on uh Andrew is there anything else that you want to add yeah we're gonna be dropping a link uh to test out digital girlfriend in the chat um anyone interested please join like

39. 20:19

    it's an agent we are working on like making it to and running lunch talken like all the all the usual Shenanigans so all the ear testers are going to get an air drop so yeah great awesome guys well it's a pleasure to have both of you on uh we have to move on to our next speaker thank you for joining is there anything else that you want to add before you hop off no no thank you great thanks for

40. 20:43

    coming awesome and next we have Kendall Cole from proximity Labs he'll be talking about shade agents and I do not want to spoil all the exciting stuff so Kendall welcome to the stage and you have about nine minutes and I'll come back on and pest you to get off if you're taking too long yeah no worries that should be more than enough time awesome well hey everyone my name is Kendall uh super excited today is launch

41. 21:08

    day for a I guess it's more of a concept than anything that we're calling shade agents so I will attempt to present let's see how this goes there's a good chance I'll probably have to uh bring your screen on stage uh yeah while we figure this out if everyone goes to near. shade you can follow along in the notion do that Kendall's probably going to be displaying yeah I'll just that's perfect actually I I'll just talk

42. 21:54

    through it we don't even have to to go there just make things easier so tldr on sh agents it's it's actually going to be the first time you can build these fully trust minimized agents um there's obviously a lot of hype right now and justifiably so both in crypto and even in web 2 uh the challenge is that especially when it comes to agents actually doing things like actually executing transactions or payments or anything that involves you know

43. 22:20

    typically a I guess a sensitive action there's there's basically nothing like you know if you use um if you use open as operator which is amazing uh usually you run into like payment walls and then and then it's kind of over uh and crypto it's actually the same if you run into needing to actually send a transaction it's either you have to get prompted which kind of breaks the the flow uh and that's the best case scenario is you get prompted um or you

44. 22:45

    have to basically give your key over to uh either a centralized agent best case an agent that's running in some kind of trusted execution environment but this this key is just within that single environment so you have all the kind of RI R of anything happening to that environment um or you're running it entirely locally and I mean while local local agents are great uh that's often either not a very good experience you're

45. 23:09

    not going to be able to run the best model unless you have like a you know basically uh rack of gpus at home um or some combination of those different things so this is the reason that we that we working on shade agents is we actually want to well address this problem so the way that it works this diagram here which Cameron has nicely pulled up is is exactly it so each uh shade agent has two core components one

46. 23:32

    is the worker agent itself this is what you think of as just agent code right does a few things it probably accesses some offchain data it definitely interacts with some type of model maybe an llm maybe another type of model that llm or that model itself could also be running in a which is the neari team obviously has some super exciting products for that or you could be you know pinging out to a centralized model like 03 Gro whatever um and then and

47. 23:59

    then it does something it basically generates an action that should be completed uh so importantly this is where the second component comes in which is the Smart contract so each worker agent basically effectively is making suggestions to the smart contract of what to do it's basically say maybe it's uh an agent that's going to be making trading decisions so it's like looking at Trends it's looking at trading data it's looking at you know anything it makes a decision off of

48. 24:24

    using some models to kind of decide okay we're going to update a position so now we to execute a trade it then basically um sends that trade request to the smart contract in order for the smart contract to accept that trade request and actually execute it uh it has to basically check that the worker agent is running the correct code and so this is where that trusted execution environment comes into play so using one of the core properties of a te which is basically

49. 24:49

    being able to generate an attestation quote that's effectively showing or attempting to prove that the worker agent has like certain State basically the right code uh acting on the right things making the right sort of the the right decisions it it t it takes that attestation quote sends it to the smart contract and says hey I'm running the right code verify me here's the key I'm going to use to authenticate when I want to make a a request and I want you to

50. 25:14

    let it through as long as that request fits these rules based on like you know interacting with certain contracts interacting with certain protocols Etc and also that it's coming from me who's running the right code the smart contract will either will check the code hash and discernment okay yes yes the code hash in this te is correct or no it's not and either accept or protected obviously if it's accepted that's the happy pass to the worker agent when I'll just submit trade requests the smart contract will say cool these trade

51. 25:39

    requests are valid and then the smart contract is what's actually going to execute those trade requests and it does that primarily using chain signatures so that this uh agent can actually interact with any contract on any chain period or even you know even even non- crypto uh non- crypto applications um so yeah that's basically how it works uh so we're basically taking two Core Concepts one is the decentralized MPC um that is chain signatures which basically means

52. 26:05

    we don't have to worry about sort of key loss or any sort of like kind of Key Management attack vectors can interact with every chain and then we have uh these worker agents running in trusted execution environments which can verify their code uh keep data private and confidential interact with any offchain uh kind of data providers apis websites llms Etc uh and then kind of make this whole run um so yeah the number of use

53. 26:30

    cases here it's extremely high I think like what's really exciting is that for the first time we can actually Empower developers to use agents as kind of core infrastructure in protocols so imagine D5 protocols most D5 protocols have some semblance of humans in the loop today if you think about like a lending protocol is a great example uh Dows play an extremely important role in lending protocols because they have to make

54. 26:53

    decisions about which assets to support as collateral uh what what type of restrictions on those collaterals they have to put whether it's Caps or you know how much like what sort of uh loan to value they allow against a collateral uh what type of uh borrowing tokens they allow you know any any of those kind of decisions uh well it with this we actually expect that um developers will actually be able to plug in these agents

55. 27:18

    to either reduce those humans to the loop or even remove them entirely and and they're able to do that confidently because we combine basically the verification of the entire agent itself within this te with basically checks on the te uh and protecting against any sort of attack that is there with the smart contract um so that's probably the most Crypton native approach if you want to get beyond the the pale on the crypto side then I mean you can imagine these

56. 27:42

    worker agents they can access anything right they can even you know either request uh information from just normal people offchain uh they'll have to submit it to to this worker agent or prove it to this worker agent somehow uh but you can imagine you know any type of automated Commerce I think we'll even see probably you know more traditional Financial type of use cases like you know like normal peer-to-peer lending offchain where people are trying to like take on you know under collateralized

57. 28:07

    loans or things like that against uh you know employment income or or or things like that uh and then you even you can even imagine bosss are able to just like buy you know any type of any type of good or service or provide that uh so yeah it's a very very large design space um expect to see the first agents out next week actually the first shade agents will be coming live there's I think three that are on the lineup for

58. 28:30

    next week and then you'll see launches accelerating from there so yeah definitely reach out to us at proximity if uh if this is appealing to you and you're excited to build we we you know happy to to roll up our sleeves and and help out any teams uh in launching the first truly decentralized agents the first like actual putting the autonomous in uh in agents That's so exciting Kendall um there's also ongoing hackathons for those people who are you know interested in building you know as

59. 28:55

    Kendall said there's a couple amazing examples coming out next week but the sky's the limit there's going to be so many more hackathons coming out there's also an agent fund um that is important to flag so a couple weeks ago um near Foundation announced a pretty much a prop shop fund to it's not necessarily the near Foundation I think it's an external entity that um will just invest

60. 29:19

    in tokens that are launched via agents or agent tokens so this is a really cool primitive to start playing with I think you know there's $20 million dedicated to support uh these projects so uh definitely launch some shade agents and feel free to hit up I'm happy to drop the support uh group as well or just message me on telegram Cameron Dennis if you have any questions um but Kendall

61. 29:45

    thanks for hopping on is there anything else that you want to share before we move on to hang from F uh yeah go to near. shade and start building shade agents for the new smart contracts amazing well thanks for coming super stoked to have you on again in the future and see you soon H bringing you up also bring my how's it going welcome welcome hi great great is so uh nice to be here this is the first time I'm presenting here yeah well we're lucky to

62. 30:18

    have you um just a quick overview it would be great for you to give a introduction of yourself introduction of Falla and then just talk about what you shipped over the last week and and yeah I'll probably ask some questions at the end if we got time yeah sure uh so uh I'm hin uh co-founder and C of f network uh and F network is a pretty old project we have been working for uh five six years uh and our only focus is one thing

63. 30:48

    te so uh yeah that's a long story but basically we believe in that uh the web three and the web two will if Will is uh uh uh eventually you know merge into a single thing you don't need to really distinguish what is web3 and what is web two and the glue between the web and web 2 is TE so that's why you know is so so important because that's uh how we can bring the web2 users in web stre or bring the web3 users to power web2 uh and now we actually have multiple

64. 31:22

    projects uh the main project is uh F Network it's a te based Network so we offer the te service and the te Cloud uh I'm going to show you that pretty soon and at the same time we also have a few other project like red Pew to offer the inference uh and even a consumer uh facing uh project called spr do fund is just a a funny you know AI agent meme thing uh yeah so uh let me uh just jump

65. 31:52

    into the demo so here let me present my screen uh first thing I want to show is uh implementation or I mean just like a use case of the uh near private msdk and we're uh pretty you know uh feel pretty happy to build the private Mr SDK together with near AI uh and as you can see it's basically a SDK that allow you to put AIT uh machine learning models into the te and uh people already mentioned te but just a quick refresh te is like a blackbox you can put your AI

66. 32:35

    model in the te so nobody else can see what's running inside so you get uh the user privacy because input output is purely encrypted you can feel free to send your prompt to some Ser uh te server without leaking any of the input or output the same time T also offer you the verifiability so you can ensure that this is the real uh AI model not you know some model that can be reduced U you know uh a downgrade

67. 33:05

    automatically this was this is what uh is already happening in open AI because sometimes you you use open Ai and you feel oh why the AI is so stupid because they downgrade the model for uh like a more dump model to reduce their cost but here with t everything is probable and uh the thing we I want to share today is is the DPS model uh it's a deeps model R1 uh is distilled into Lama 3 uh Lama

68. 33:34

    3.3 17 billion uh because this is the largest model that can run on a single te and we already deployed this on our server so you can access it through our model serving uh uh pro project called red. it's already live so you can try it out so here uh as you can see uh we uh we open the dashboard I already uh you know charged my account and created the API key and once I have this API key I

69. 34:05

    got access to all the models available in red Pew and today I'm going to show you the deeps one so first I'm going to just uh copy the API key of course you can create whatever API key you want and then uh to use the API key let me switch to another window so I'm using cherry Studio this is is a pretty awesome uh uh uh RRM client so it supports a lot it is not developed by ourself but it supports a

70. 34:36

    lot of models and what I have done is just to create a red Pew provider and enter the API Keys API host and then I configur the model name here of course you can use uh all the other models available on red Pew as well then uh we can switch to the chat uh UI I will just uh switch which the F dipstick and this is run in the te and let's say how it work tell me the difference between te and

71. 35:09

    dkp you can see it's already uh start the uh deep thinking mode and it's going to take some time we roughly counted uh the performance uh it's roughly uh 20 tokens per second for one users and of course uh the the one te can support multip concurrent uh uh requests so uh recently we just did a benchmark uh it's not a serious Benchmark but just a uh a simple Benchmark here uh and it looks like we can easily get 182 tokens per second with 20 uh

72. 35:51

    parallel uh request and we can even bring that higher if you uh uh enable higher concur cury I I haven't tried I haven't tried this limit yet uh and the interesting thing I want to share is the verifiability because now we claim that everything is happening in the in the private Mr is private but how you can convince is really private so uh this is why we need a proof this is called te proof or remote testation so uh we exposed the testation through the red PE

73. 36:25

    API as well so here if we copy this uh this is the testation uh uh uh API endpoint and let me share the console so uh I'm going to replace this API key with the real one and here we go now you can see uh there is the Intel code so this is the CPO te proof and the same time we have the Nvidia payload so this is the uh Nvidia uh confidential Computing uh remote attest station so uh I'm just

74. 37:06

    going to copy this remote testation and show it looks like we lost hang can't see him in the chat rest in peace hang oh here you are great sorry I I mistakely click the leave studio so yeah let me let me present again so if we go back to uh the remote testation explor just copy paste the testation click on verify then you can see this is a verified attestation and you can see all the

75. 37:48

    metrics here I'm not going to dig deep but if you're interested we do have a a tutorial to help you verify what is really the model running insights by checking this proof all right so uh this is the verifiable uh model uh and yeah it take a lot of time so I'm going to speed up for the other things I want to share the second thing we have done in this week is pretty important and pretty

76. 38:14

    technical is the KMS in our underlying uh SDK called dag so dag is our sdk2 uh uh deploy te uh in CPU te and the private MRK is buil on top of that to add the GPU support uh and the KMS uh in uh this deck is very important because uh without KMS you cannot you can only rely on a single uh Hardware to deploy your stuff so say if you deploy your important important application uh you have a lot of encrypted data but if this

77. 38:50

    single CPU uh got you know just a burned or uh if you use someone else server they just uh quit the the exit the network then you lose all the data because everything is secured by the CPU uh the CPU key but with the KMS then we can have we can decouple uh your uh encryption key with the security uh of uh the entire uh the the key management of the entire network so that even if a CPU is that you can still switch that to

78. 39:21

    another CPU uh to continue your work so uh pretty Technical and this is a real giant uh PR uh we're going to merge this soon and it will also be available for the private Mr SDK as as well okay finally uh let me do that pretty quick uh another thing we just launched uh recently is the no code uh deployment of Eliza this is a very popular AI agent uh framework and of course we're going to support more so uh this is a no code one so I'm just going to go through that uh

79. 39:57

    first you can SW you can upload your configuration or you can also switch a preconfigured uh uh character file so Persona like SP SPF and then you just choose which uh client you want to connect for example Twitter Discord but here I'm just going to uh only enable the direct so only API access then model provider of course we can choose open AI answer P or red Pew so I'm going to copy this one here so red pew and default model we can use the

80. 40:30

    dpse and next then uh choose the Eliza version and the character file so uh this is the generated character file you can see uh bio SPF is a totally trust worthy assistant so you can trust it and environment variable means all the important secrets for example the API Keys uh everything is encrypted in the front end so uh it will never reveal to any Cloud providers like us it will go directly into the te so only you and the

81. 41:02

    te can get access to that then I click complete it's going to bring me back to the Fout cloud and as you can see uh the agent s SPF is starting so you can see the logs like here is already uh putting up the uh virtual machine uh and uh since this will take a long time because Eliza has a huge uh image uh so I'm not going to just uh uh walk through everything but uh here as you can see uh it's very simple uh process

82. 41:36

    you can easily click click click and then deploy your agent without any code amazing all right yeah that's all from thank you so much I cannot wait for the day where you know these agents that are hosted on Falla are visible and doppelgangers and you know verified liveness with proof of response and it's also a shade agent so you can actually transact on any chain like all of this

83. 42:00

    stuff is coming together in real time and so I think this is the whole point of us running these spaces is just to kind of let the community know that you know everyone on this call is pretty much working together in one way shape or form um and this is how we build the open web and US own AI so thank you for joining this has been incredibly enlightening uh next up we have Omar from intellect they're taking it a little bit

84. 42:25

    out of the retail eye and more into Enterprise world and so um Hong thank you for joining and Omar welcome to the stage if you can just give everyone a quick overview of yourself what intellex is who your main Target customer is and yeah what you ship this week it'd be great to hear from it all right awesome thank you thank you camera for having me today um I've been hearing with a lot of

85. 42:51

    excitement everything that is been developed um shortly and for those that don't know we we just rebranded from brain stems to intellects like I don't know like a couple of weeks ago uh so we I might still say brain stem by mistake but you know brainstem intellect is the same thing we just rebranded to to intellect um what we what we are working on and by the way I'm CTO and co-founder at intellect um what we are working on

86. 43:23

    is on creating what we what we call the agent interoperability and coordination for multiarm architectures um we are building a protocol um and doesn't stop just um with a fancy word of agent that everybody is speaking about um there are a couple of things that are funding fundamental for for a protocol in one side is um everything around assets registry when it comes to agents uh

87. 43:53

    autonomous agents the models that are behind data sources um and anything that is related um there is of course the the what is about the building of those models and and that knowledge and that intelligence uh we have been speaking about today about te uh ml private ml um we come from the decentralized thir learning concept um which is the way that you build um common knowledge or a

88. 44:24

    common model without exposing private information and also ensuring that we have the right environment um for an open ecosystem that brings and contributes um to the to that intelligence so and speaking a little bit about the customers that we are serving and the kind of companies we are working on if you take the food service industry in the US you know it's a fragment in the industry you have a lot of different actors from the biggest

89. 44:49

    to the smallest and all are important to understand how the ecosystem Works um for having the best and most comprehensive uh model around how the food service industry works I need from Coca-Cola company information but I also need the mama pops information you know on the different products they have the different marketing strategy they have you know from the smallest to the to Big to the biggest one so for that to happen

90. 45:15

    it's really important to have um the the the right contribution incentives the right um the right way of managing all that intelligence the right way of contributing uh data validating and all the different actors that that you need to have and to keep this open ecosystem because the biggest also is going to try to establish the rules and the only way to ensure that that's not going to happen is using blockchain and using of course the AI methodologies that that

91. 45:48

    that we are using on our on our framework um there are other components like consensus mechanism and validation that is also a way of bringing the web tree and and and and token holders to all these um environment which means you know like I can own an agent I can manage that agent that can decide what tasks and we will speak about that they will they will be solving and and stuff like

92. 46:12

    that um I I will quickly move on because I already spoke a lot um but speaking about what we want to show today we are working on a use case for a a customer on the food service industry that is part of that you know fragmented industry where you can have a manufacturer like Nestle Coca-Cola um you will have Distributors you have restaurants you have Broker sales and they all have the same kind of problems they they have different understanding of what the future is going to be so when it comes to um plan and and and act

93. 46:47

    in an coordinated way because they all need to work together they fail uh because they don't they don't really trust each other don't put all the information on the same part they don't they don't train the models with that same the same information so one thing we are working towards that is on on uh using swarm of agents and here we have a different approach or at least what we think is a different approach uh from what we can see out there uh which is um

94. 47:15

    a more probabilistic approach so we can see a lot of agents that are being built those agents have a task to be done but it's it's somehow they they're defined the path is defined for them so we are really expect that what they are going to do and how they are going to do it the way we we we solve those problems um on this fragment in industry is from a probabilistic approach where we know that there is a problem to be solved but we are leveraging the um the

95. 47:43

    probabilistic part of an of an agent and everything that is behind that agent to solve that problem that means that I'm building swarms of Agents with different roles and and and and objectives and I throw them up problem and they will solve that problem in ways that I'm not expecting to happen um so the demo that we have today that is as a shorter part of what we are working for this customer uh when it comes to well particular we

96. 48:12

    have ferate model um that given a a a product will give you or given menu items will give you the different products and then you can establish a a sales strategy for selling that product um uh depending on the place you know industry Etc uh but we what we have is um a bunch of different near AI agents with particular tools you know representing different roles as analyst consultant data scientist datas scripter the coordinator search representative and there are some others that I'm I not

97. 48:43

    including um for time purposes on on on this demo you know like agents that are representing a product that that are representing restaurant so you can you can have an entire not just a plan but a a verified plan um and we have two three different models that are behind this and this is important also to achieve different ways of thinking uh sometime from the same agent but sometime from

98. 49:08

    different agents with the same roles so we can compare and choose the best strategies um so without further I will try to make this work and again you know like agents and and probalistic stuff might go wrong uh so the fun part is that and on the left side you can see um the agent framework running on the right side you can see um the the the near AI agents and at some point we will see um some of those agents that are showing what they are doing what I what I have and hopefully it's going to work

99. 49:46

    sometime I hitting right limits um we will see those agents establishing depending on the tools they have the problems that right now I'm asking is build me um a sales strategy for selling um uh baby ribs in Austin Texas so you have a manager that is planning and establishing you know well I will need to search I will need to analyze I will need to check on the menus I will need to do a lot of different tasks to achieve this marketing plan now sometimes it might happen that it thinks

100. 50:19

     that it has to go you know into into one route sometimes why think that it has to go into another rounde and that is part of the beauty but also part of the problem that we are trying to solve when it comes to have agents that are interacting and this could be neari agents could be crei could be whatever platform and framework um agents we need to establish a standard where they are

101. 50:43

     going to Define who they are what they are doing what what is the role what's the kind of information they are managing you know because we will need to uh ensure that that's the right agent in one side but at the same time and connected to what we have seen from the other teams you know having proof that this was executed that it was fairly executed that it was done correctly and so on and it's going slow as hell but it's it's moving it's moving so those

102. 51:14

     are those are kind of the the the the things that we need to understand we can see here that is already calling the different agent some some cases for scraping some cases this is an analyz is uh agent that has you know a particular task um so understanding and looking at how how it behave it allows us to understand you know like what we need to take care how we need to watch out you know like when we are going to have three agents for the same task we need

103. 51:43

     to agree or to get to a consensus um so those are the kind of things we are working and next week we will see next week we will show some some cool ux that we are working out for the activators management platform I know I'm over time let me go through the latest part of it uh I can show this or share this later um on what is next for us and what is important um establishing what I was

104. 52:10

     saying collaboration communication star mechanism reward and payment system this is highly important this is something we we are working uh using near reputation system based on consensus this is Rel to what I was saying just before and you know an on chain call for agent that I think it's cool for having different agents that are out there you know on this open ecosystem so we can throw up a task you know this is something I have to solve and looking for the agents that

105. 52:34

     are listed that can fulfill that task um and that's it yeah beautiful this is again incredibly exciting actually building agents for the real Enterprise if agents do end up automating the backends of businesses we're talking like a multi-trillion dollar opportunity here and so really excited to build together I also just want to again like how all these things sort of Stack upon

106. 52:58

     each other um and at the end of the day we do need that Enterprise adoption in order to like hit Mass market so keep up the great work and hope to see you on a future episode thank you awesome actually before you hop off is there anything else that you want to tell the community follow you guys anywhere I did drop your uh Twitter account in the uh in the thread but yeah absolutely uh check on

107. 53:21

     our website check on our telegram we are we are going through a lot of different things uh hopefully on on the few coming weeks we are participating on e e Denver next week so hopefully anybody that wants to discuss about agents Enterprise reality use cases you know happy to help on on on on some good meetings awesome well thank you Omar and last but not least we have zahed from Jutsu AI to

108. 53:48

     talk a bit about some of the open source contributions that he's been making to the near AI ecosystem and zah if you want to share your screen I'll bring it on stage and you can take it away yeah can you hear me yeah okay perfect uh present let me share I'll share entire screen but I promise there's nothing please please please no yeah okay so hey everyone I'm zahid I'm founder of Jutsu so Jutsu uh we are

109. 54:26

     working with near to kind of build a better typescript experience uh today my goal is to kind of a share the current state of uh typescript and then show some demo and then later we can have some Q&A if we have time but before I even start I think most of the work that I'll show it's uh the I think the team that deserve the credit is uh near AIT and especially uh vladim and Anon and all the people that helped me to kind of getting uh into learning this typescript

110. 55:06

     SDK and kind of a start building uh agents so thank you near AI team and thank you Vadim for your great work so most people probably know about uh near python SDK and python SDK is the mature SDK so if you want to know about near a SDK so you can go to doc.ai where you can find very uh good documentation about uh how to build a agent in Python uh but recently uh this week uh vladim uh kind of a created this uh typescript SDK that get marched so you can start kind of building uh typescript

111. 55:54

     SDK uh typescript agent today so this is kind of the this this is kind of whole thing is kind of the anatomy of a typescript agent so all you have to do to build a typescript agent is kind of a include the import the typescript runner and then that environment exposes few function which will which should be similar to someone who already built a python ENT so you can write a file you can uh ask for a completion from LM uh and you can add

112. 56:27

     environment variable and get uh latest messages and thread ID I'll come back to kind of the um uh kind of the demo later but this is a document uh if you want to get started with typescript uh SDK today so this is a document I kind of a I'm going to share here so if you want to kind of start today you can start today uh now let's go uh share you some some some of the typescript uh agent that I build so previous speaker was talking about kind of a multi-agent

113. 57:03

     communication and how do you build a more sophisticated agent so uh I will share some of those demo so one of these agent for example so this is the simplest agent so a simplest form of agent can be a agent where you have a system prompt and you your us as a user you ask something to uh that agent to do it so in typescript it is just a adding a message and from environment variable you call this completion method and then

114. 57:35

     you pass your prompt and that's that's a simplest agent uh and then you the other thing that you have to do to kind of a deploy this agent in uh near aiub is adding this framework so if you uh this is a way to signal that uh near AI have that this is a uh typescript framework so that they can run the typescript runner so this is the most basic uh agent that basically given like a it can make a joke so if you go to the uh if

115. 58:12

     you go to the um Hub and if you try to access that tell me a joke so it will um sometime the joke get called because it takes a little bit time but I know that near AI team is working very hard to kind of make the uh environment run faster so hopefully pretty soon it will it will come back with a joke but the second oh so it come come up with a joke uh so nice and then the next thing I want to share is like a what

116. 58:47

     happened this one simple thing is great but like you probably want to build a more sophisticated agent where it can communicate with each other so this is another example of of a agent which is more kind of a sophisticated so uh I always wanted to write a book about Ai and book about AI agent and everything so I thought uh basically yesterday why not kind of build a agent that can write a book about whatever thing I want to write so this is a agent that can write

117. 59:20

     that can help you write a book so you can see the first thing I do is I ask this outline outline agent to kind of a generate a outline for a topic and the goal that I give it to uh that uh agent uh So based on my topic that I'm interested and the goal it will kind of create the outline and when I have the book outline then it goes through each chapter and then ask another agent to kind of a write that chapter for chapter

118. 59:50

     for me so the uh the outline agent gives me few things it gives gives me the name of the book a name of the chapter description of the chapter and kind of a key points that that chapter should be and then go and run each chapter so uh I'll run this demo so this is uh you can see it's a typescript uh project so if I run the demo so the first thing it is doing is like it is trying to figure out the um the book outline and sometime

119. 1:00:22

     because the book outline have to be in Json format so it does not f it failed to kind of destructure so you will see like it will retry so this is one of the thing when you are building a agent you should remember it's like a sometime agent need to be retry so you need to have a kind of retry mechanism so you can see it already start writing the first chapter so introduction to

120. 1:00:45

     aien uh and then it will write in markdown so then it will start writing the second chapter so if I go to the here so here is my book so so you can see introduction to AI agent different type of agent so like a s reflexive agent model base reflective agent so you can see like it keep writing that book in uh in a markdown format and now you can extrapolate the idea so maybe you

121. 1:01:14

     want to write a uh article or whatnot so you can you can do all those things using this agent going back to my slide so uh this is the two demo and if you want to go to the typescript uh documentation this documentation will be in the GitHub uh in the uh in the quick start guide pretty soon so like but for now if you want to start now so go and check uh I have a step by step so you

122. 1:01:46

     can really follow through and then it should just work and yeah and then like the next a road map for us is like a to improve the developer experience so right now it is still little bit of work on like if you go to the through the documentation you will see like you have to rename the uh python uh agent. Pi file to t uh agent.

123. 1:02:14

     TS and you have to kind of copy paste some files so there are a lot of these kind of a developer experience that we will work on and we'll kind of improve uh in next few weeks um so that is exciting so and I am really really excited about typescript SDK because I love typescript um and if you want to kind of a reach out to me you can reach out to me obviously but there is a uh there is a channel where you can ask for help so maybe Cameroon you can

124. 1:02:47

     uh share the uh the Channel with everyone so U so that anyone if anyone have question they can ask in the chat Channel yep I'll do that but I'm also a little cognizant to post things on Twitter because that's how you get a bunch of bots so um yes we'll absolutely post that to anyone that needs any support um but zahid thank you we are up on time we don't have a lot of time for questions but uh definitely reach out also just make a like issue on GitHub like they

125. 1:03:17

     get responded to quickly so um again thanks for hopping on hope to see you next week we'll be be doing this every single week so uh obviously more awesome well that concludes Our Guest for today just as a quick recap uh just want to walk through there was a lot shared so you know shout out to all of those who have been actually following since the beginning um I did drop a lot of the links from today's uh call on this Twitter thread just as a quick

126. 1:03:46

     recap uh the near docs got a pretty big Improvement uh devx is improving thank you zahed and the team for making that happen doppelgangers launch laed their uh digital twin in their mobile app so definitely give them a follow there's the auditor agent RFP so if anyone is a security engineer and wants to help red team agents in the near AI Hub check out the RFP it's added here um again made the near AI Hub much faster uh Hong and the the f team definitely shared some

127. 1:04:15

     exciting updates about their Cloud about red pill definitely give this video review and absolutely give them a follow they're by far one of the best teams focused on tees in this the entire actually AI space as well so definitely give them a follow tees are going to absolutely change the how all AI operates like this is going to become a standard over time this is why Nvidia is adding tees to all of their new a100 and

128. 1:04:39

     H h20s and h100s and b200s so like this is Way Beyond crypto and fall and near at The Cutting Edge of this so be sure to give them a follow as I highlighted we are organizing an in-person hackathon in San Francisco on March 14th and 15th that is in partnership with SF compute and Electric Capital Electric Capital is amazing there if you do or if you're lucky enough to get investment from them

129. 1:05:03

     uh they support a lot of incredible teams with uh these types of events so shout out to them uh private MLK we talked about that intellex give them a follow uh again they're mostly focused on Enterprise adoption again a prerequisite for Mass adoption we definitely need it and with uh shade agents is a huge announcement as well I actually need to write about it again so here's actually the Twitter thread by

130. 1:05:27

     the near protocol account um explaining all what this is so exciting stuff and yeah type script we just talked about it that's all we got for the week and shout out to near week for post hosting this space and putting this all together it's been a pleasure to work with you guys and oh actually lastly next week is e Denver there's a lot going on we do need to confirm if we're going to do this space we'll talk about the async but

131. 1:05:51

     there are some really great events if you do find yourself in Denver um there's an entire list of all these events that new year will be at first off is the one trillion agents day it's a two-day conference uh actually the first day is followed by a happy hour with Deli digital good friends of ours focused on the decentralized AI space and also crypto of course and then you know I'm not going to go through all these events it's a ton the one event

132. 1:06:15

     that I'm co-hosting again with electric capital is eth rising this is mainly around uh decentralized Ai and defi so come attend this event for what's kind of on The Cutting Edge of institutional adoption and the role of AI again there's so many things here not going to unpack all of them definitely give this video a rewatch if you want to dive into the specifics but that is it for the

133. 1:06:39

     week anyone have any let's just double check who's gonna send this nope okay good we're good guys appreciate you all tuning in and give near AI a follow see you next week thanks for watching subscribe to near week