Permissioned Memory for AI Agents: What’s New in NOVA on NEAR

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

1. 0:04

   Yeah. Uh, thanks everybody for joining. So today we have uh Julian from Nova who was on a few months back now talking about the product Nova and the SDK, but you're here today with some uh exciting updates to Nova and some different directions as well.

2. 0:23

   Hi Joe, thanks for having me. Let's talk about how what NOVA is and a little bit of when you came on the first time, what you were talking about and then kind of where we are today with everything. Yeah. So, I think last time we spoke together in this podcast was probably just before the near con. So, it was probably January uh of this year 2026.

3. 0:48

   And I was presenting the first MVP of Nova. At that time I was building a secure file storage system for AI [snorts] and um my main use case was building uh a decentralized data layer for user owned AI. I could identify at that time that the whole ecosystem was driving towards private inferences to be able to run AI models in secluded enclaves gated gate like computation.

4. 1:29

   And um I was wondering how could we complement this by proposing a solution for the data and the files that will get in and out of those enclaves. So that two AI model could communicate and exchange some data sets um while being sure that those data sets will not be read by anyone else. And so that was the beginning of Nova. and um working onto it um I started to private uh so let's say the whole principle is

5. 2:08

   the same the the main uh statement is the same it means that no server ever holds plain text or keys and um so I I'm I'm still betting on the zero data retention uh principle and what I do now is that I change my target of customers of users. I was targeting B2C. I was targeting like individuals wanting to use AI privately.

6. 2:40

   And now I understand that there is probably more concern coming from businesses. So I'm proposing an update on Nova with more features that will fit uh what professionals would expect when they are running multi- aent systems as a business. Yeah. And so the new let's say ba baseline for Nova Nova is a shared permissioned memory for multi- aent systems. This is where we are.

7. 3:12

   Nice. what prompted the switch from B TOC to B2E because I've heard this actually a lot from different AI companies uh in all different capacities. So what was it for you? For me honestly I think it was customer adoption uh because I'm speaking I can discuss with uh let's say my friends and my family people who start to use AI and start to be aware that okay maybe this data will be leaked somewhere but there are not necessarily buyers of a service of a solution that

8. 3:45

   will you know uh encrypt their data because they they don't really need it that much I believe like professional users or power users they are more concerned about that and so they have they have another level of expectations that that's the reason nice um for Nova what is the the what does it start to look like now that you're going towards enterprise adoption does the product change drastically um because I

9. 4:17

   think like obviously you said like the data retention piece is a really strong selling point um of all of this but what changes in in the product and you did a demo last time walking through how easy it was to kind of start to connect to it. How how do you see it differing?

10. 4:34

    So, what we've seen last night uh last time was um is still uh the same. I would say it's maintained. It's the is the backbone of my product. I don't know if you remember but we opened two Nova sessions and in chats and and we saw that they can authorize one each other to read uh data they they upload on the on the shared layer and so this thing is still uh let's say the the core of Nova um what has changed

11. 5:06

    are a few things I have migrated the file storage I was using IPFS back and and now I'm using fast for two reasons. Uh the first reason is that it I mean it allows me to integrate more um tools from the near ecosystem. So I tighten Nova to to its closeness with near but also the real reason is that you cannot really delete a data from IPFS and with fasts it's the file deletion is engineered a bit differently. So I can claim that yeah if

12. 5:43

    I say that your data will be automatically deleted in some days it is it is actually like uh deleted. So that's one thing I migrated the storage. I also migrated the MCP server. I was doing it on um fast MCP and uh they have been in integrated in another brand I think Horizon I I started to okay I use the opportunity to migrate my server.

13. 6:10

    Now the MCP is stored in the same TE at fellas the same TE as my shade agent. Uh I also actually I keep using shade agents even though it was uh not maintained anymore. I'm uh I am now self-maintaining the shade agent v2 which uh matches well what I have to do with Nova. Um I improved the key management system. So I realized that um yeah every time I push the code updates uh on the beginning my my keys were lost and that was of course a big uh I mean that that

14. 6:53

    was a big down part. Uh now I re-engineered the whole key management system. Uh so I'm I'm I'm storing a master key on a on another contract a KV contract. And so now I can I mean I can push as many code updates as I want and I can even migrate from one TE provider to another and the whole Nova systems re reconstructs uh all the all the keys all the the data access keys and the and the user keys every keys is derived from the

15. 7:31

    from the same master key that that's stored outside encrypted of course uh but so that's Another that's another improvement because now let's say at some point if I if I want to change my TE provider or even like use another private inference system I could do it without hurting anything of the existing files at Nova.

16. 7:54

    And uh I would say the most visual improvement is that I built a file viewer dashboard. So in my early days I was only um interacting with my data layer from the chat. Uh and now you can directly like open the dashboard see all your groups and and see all the data that's stored here and all the useful information. You don't have to query the chat to know um which groups you own and and which data you you you still have on those groups.

17. 8:29

    Nice. Nice. That I It's a It's really cool to see being able to take open source software and like kind of keep it your own, you know, keeping Shade agents alive for what you need. And I think that that's really it's a really cool benefit. Uh do you have something you want to show today? Do you want to show anything? Um yeah, we can overlay. Yeah, I will show you the what it looks like.

18. 8:56

    Great. Uh so this is um the front end. Uh it's a bit updated. I'm I'm showing like the new definition of Nova. Um but also I'm proposing three ways to build with Nova. My flagship product is the SDK. There are two versions, the JavaScript and the Rust SDK. Uh there is an agent plug-in that was um pushed to to clude uh community marketplace.

19. 9:24

    And uh and then you have the the legacy chat here uh which we which we will access soon. Um we can connect. So something else I I don't know if you remember but uh back in the days when you were connecting your Neo wallet you had you had to create a Nova account in the background because I'm I'm using Oro and so all users were redirected to using uh a Nova account which is a sub account of the Nova SDK main account. Now um

20. 10:02

    near wallet users have real non-custody so they can use their own near accounts uh to pay fees and to store their data and so Nova is really blind on on those accounts. Obviously for all the convenience those who want to connect with their email will have an account being created. So a new wallet's being created in the in the background. Nice.

21. 10:28

    And so you see now when you connect your account, you have a new dashboard tab that's appearing here in the header. And this dashboard opens to four um pages. It has a light version or dark version. Yeah. Um and then here you can check for your account first.

22. 10:52

    This is where okay you know that you are connected with this account. It gives you the balance. It gives you the balance. So the the near that you have in your account. Um you have a convenience pink pay link if you want to buy near tokens for your uh to to uh top up on your credits. And then this is where you will create uh generate an API key. The API keys um allow you to use your Nova account from

23. 11:23

    the SDKs or from the from the clo plug-in for example. Uh as soon as um someone have seen it here, I would rotate this key. Yeah. And uh yes, I confirm this. I want to rotate the key so no one else um can use it. Yeah. Yeah. Um then uh you still have the chat the chat that we used to to use uh on the first days for example we ask the chat which yeah so this one is only me uh but so yeah that's that's a basic way how how

24. 11:59

    you would interact with Nova from a chat but now what is really uh cool is the the files the file viewer dashboard here you can have a direct look at all your groups and uh the different data that are into it. And so you see here for example this group I'm both an owner and a member on the on the engine test evt group. I'm just a member. So it means okay I have access to the files in this group but I'm not owning it. I don't have the the owner rights.

25. 12:34

    And when you click on the group here you know who are the members. So it's it's uh my email and and the Nova SDK and you have uh the file that are into it. On the back end I can see it's the legacy back end which means it's on IPFS now. Uh but the status is active. So the file has not been deleted. Sometimes we'll see deleted here and we can check the integrity. that that's I would say

26. 13:03

    that's probably a big uh highlights of the of these updates is that uh we can prove the bite ident identical data that you're about to retrieve. So when you click on verify you have this uh this stamp uh that tells you okay the file that you are downloading by click on the button is bite identical to the one that was uploaded by the owner. uh this is very important on some use case. Imagine you are I don't know a hospital and you are training uh an AI model on some cancer

27. 13:43

    data for cancer research. You want to be sure that the data you are training your model on uh has not been corrupted uh or even has not been leaked. You know, you want to be sure that the data is the exact same data as the one that was uploaded by your your um your other hospital partner, the one who let's say sharing the data with you, you know.

28. 14:08

    [snorts] So this is what we have here. Um this is very convenient I would say because we can directly check for example here you see on the back end active you you you can see now with fastest when uh the data was uploaded you can still verify this you see that it's active and there is for example on this one there are some files that has been deleted so this one was my test for automated deletion

29. 14:43

    And um yeah when you click on deleted uh this is the real tombstone it tells you okay the the file was deleted at that date and it can prove that um I mean it's an onchain proof that the file is not available anymore. So of course the encrypted byes stay in the in the blockchain archives but the key has been shredded and um and the routter uh has been is not serving anymore that route.

30. 15:15

    So it means yeah we cannot claim real data deletion on chain. of course the the the encrypted byes stay there but there is nothing else that can retrieve it and nothing else that can decrypt it. So that's the solution I found to be able to claim for example if you have um if you are a company you are you are a company that that's selling some um some coins or some products and you tell your

31. 15:47

    users okay we don't own your user data and uh it will be automatically deleted let's say 30 days after your purchase. So this is how you can prove it. You will store the user data. Let's say the email or the wallets that that make the purchase. You will uh you will maybe store the metadata of the product that was that was uh purchased for your accountancy or maybe because the yeah the user wants would want to download a bill uh of his purchase in the next 30

32. 16:20

    days. But after after the after the time like your your data is auto automatically like deleted and no one can access it anymore. So I think this is a really good solid argument for their VR. Uh you can we with Nova we can really build apps and websites that claim okay we don't own your data. we never do and and and yeah and we delete it automatically after after some time like the retention points.

33. 16:52

    Yeah, that's really interesting. I mean I love the addestation part of one like the integrity of the file you know I think that is certainly overlooked right now you know like especially we talk about so much with like TE's yeah the add astation is with the processing but like is the output staying consistent or whatever whatever else you're uploading here is that actually the right thing uh probably eliminates some hallucinations I'm assuming as well when you're using

34. 17:23

    the chat feature to make sure that like they're not pulling in something else. Uh, which is really interesting. And I do I think I can see why you pivoted to enterprise here because this makes a ton of sense for a corporation, you know, or research or, you know, academia. Like I think all of these areas like this makes a ton of sense of like how you would start to use this in that in that setting.

35. 17:53

    Cool. Thanks. And yeah, so I started to build a few use case with this. Um, two or three months ago, I think uh I built um an automated hackathon submission engine. So I use Nova for hackathon participants since now we are using like I mean aents are are the main like um thing that you build let's say in a in a hackathon and so I use Nova so that agents can directly uh uh join uh a hackathon uh submission

36. 18:35

    group a hackathon um competition and uh and they can at at dates they can automatically wrap up their code as a metatata file and send it to the to the submission. So only the judges will be able to decrypt and analyze the submission of that hackathon. So that's one way I use Nova which I think makes sense and could be used elsewhere. The other thing I'm also using Nova in um in an accountancy context um when your

37. 19:14

    agent is used uh in a commercial um context it will buy and sell stuff maybe data or other things and um I use Nova so that every month there is a crown job that wrap ups all the purchase and the sales. that the agent did and uh it sends it to the accountant uh accounts.

38. 19:41

    So only the accountant can receive like every months all the activity the commercial activity of the agent. So this is the kind of automation that I think makes sense. Uh last time we discussed but okay will you store all data all kinds of data on Nova. Well I don't think so. I believe that we still will use local databases and database for other purposes but there are some data that works being at another level

39. 20:12

    of of persistence I believe uh suppose you will you will if you crash your your app let's say after two or three years you can still have like someone auditing your company or the state saying hey by the way uh what did you bought and sell with this old project and you need to come back with with accountancy. So these these are the kind of things that I'm thinking about.

40. 20:37

    Yeah. No, I I think I think that's that's great and uh hopefully people see this and as a great thing to start to build on top of. It's another great feature of just leveraging some of the things that are on near some of the things I think that you have come up with here to build this in. Um, I love that it both automates but kind of keeps the human in the loop if needed to some degree. I can always verify uh myself which is really nice. Uh, so this is

41. 21:08

    great. Uh, how can people get started and what do you in in uh is this live in production for everybody? Yeah, it's live in production for everybody both on mainet and test net. If you want to switch to a test net uh to try it, you click on the test net peel here in the header. It redirects you to test net.nova SDK.

42. 21:32

    Uh, of course on testnet the file storage is uh mocked so you don't but you can still you can still test u your SDK. you can build your app for free and then when everything is is well settled where you will start to pay fees to store your data. Uh Nova is pretty cheap uh but it still have to pay for the storage. Uh this is a cost for me. So you will pay for this cost yourself. Um

43. 22:02

    you pay a few cents uh in near to upload and retrieve file made all all they say the the onchain operations. Um and yeah now you can start with your own near account. Uh you connect your wallet and and then you can handle your files and uh you will pay uh small fees every time you want to upload data and make file operation. That's mostly about it.

44. 22:31

    It's ready to start. Um I'm happy to provide some support if someone has a special need especially for companies. I have been contacted some time ago by different companies uh that that were using a fleet of agents to handle some operations. So maybe if it's not clear for you, we can sync together or how can Nova help you like for your Asian communication within the fleet and all these kind of things.

45. 23:01

    But otherwise it's pretty straightforward. Um the SDK is well documented. That's great. That's awesome. Yeah. No, thank thanks for for showing this. You're welcome.