Ledger Tracks Explained | Etienne Waldron & Oscar Chaix at ETHOnline 2026

Hello everyone and welcome to the Ledger workshop. We are Etienne Wouters and Oscar Sex. We are both product are both product managers at Ledger and work on artificial intelligence initiatives. So, let's get started So, let's get started . Here is a brief agenda for today.

We'll talk a little about Ledger, but not too much but not too much . Right after that, we'll talk about trusting agents. Oscar will present some tools that you can try. At the end, we will present the areas will present the areas you will work on during the hackathon. So, first a little about Ledger.

Ledger is a French company founded in France in 2014. We create hardware devices that help protect your cryptocurrencies. Today, we protect approximately 20% of the world's crypto assets and have sold over 8 million such devices to date. The title here says "security by design," and if you look at the image on the right, you can see a chip inside the device called a " called a " secure element," with the screen located directly above it. This means that the data on your Ledger device cannot be changed, and what you see is what you sign.

This will be important for everything we cover today. I’m sure you’re familiar with our cryptocurrency security product line, but it turns out our architecture is also perfectly suited for the agent era. Ledger for the agent era. Ledger is also a reliable infrastructure for the agent era we are currently in. This is a very strong statement, but remember: anything that can be used on your behalf can be used against you.

This is extremely important in our time. Of course, prompt injections are possible. Models can, of course, hallucinate, although they do so less and less often over time. And your secrets stored in plain text files can also be stolen. There are many reasons to worry in the era of agents, and we will see how Ledger can be the solution to all of this.

Essentially, we at Ledger believe that agents should suggest actions, but you, as a human, should approve them. The hardware does the signing, so of course the private key never has to leave the protected element, never leaves your device. And you as a person should be able to understand what you are about to sign on the screen. And everything, you know, everything that matters, everything that is critically important, needs human confirmation, physical confirmation from a human. So, this is the model we're talking about here .

And now Oscar will talk about the whole stack behind what I just mentioned. Yes. Thank you, Antoine. And hello everyone. So, in terms of agent technologies, we've already released a few products, and I'm going to describe them right now.

The first right now. The first is command line interfaces (CLIs). They allow you to interact with our platform, whether it's Ledger Wallet, Ledger Enterprise Services, or even enterprise multi-signature, but directly from your favorite agent. So, if you use CodeX, Cloud Code, or any other agents with terminal access, you can use them to automate certain actions that you would normally perform manually in our apps, as well as trade securely using your device. Here are some examples: you can ask your agent to read your portfolio and then ask him to lock up 20% in Bitcoin.

By " Bitcoin. By " clear signing" on the device. If you have specific goals for each asset, you can provide them to the AI, and since it has access to these different positions, it can offer you, as the user, optimized exchanges. You only need to connect and confirm the action on the device. Or you can do things that are a little outside of what we usually do in our apps.

You can check the actual recipients in your shipment history. You can also optimize your gas fees by asking your agent to perform calculations you wouldn't want to do yourself. This is what we started releasing in May. Since then, we've added a few features, which I'll demonstrate on the next slide. You can also teach the agent how to sign.

This is the second pillar. This is a bit of a broader topic, but in short, we have an SDK called the “ device management kit.” We've built skills that allow allow any agent to integrate into your app or app or any page or workflow you can imagine you can imagine . So, you know, what used to take days, if not weeks, now takes 10 minutes. Just ask your agent with these skills to integrate Ledger into my wallet or require confirmation via Ledger before my bot sends anything.

You can also think about non-cryptocurrency use cases, as this supports apps like the Ledger Security Key, which is essentially FIDO2 2FA directly on your Ledger device. Finally, this is a relatively new solution, and probably the one we're most eager to experiment with at the moment—the keyring CLI. Essentially, this allows us to go beyond cryptocurrencies. As a developer, you can configure your machine with Ledger hardware and allow your agents to encrypt and decrypt secrets, like API keys or other data that needs access to perform actions. This is quite convenient.

This allows you to move from storing secrets in plaintext in a .env file to something more secure and encrypted directly through Ledger. This can also be restored via Ledger. So if you lose your Ledger, you can regain access to those secrets, which are stored in a decentralized protocol called the Ledger Trust Chain. So in this regard, this regard, any hacks or experiments to extend the boundaries of this feature or its availability for other settings, for example, for remote agents, would be very interesting to us.

This is one of the things we released. I'll show a demonstration now, but there are two other applications that go beyond cryptography. They are called Open PGP, which allows you to use your Ledger as an encryption device and smart encryption device and smart card. And there's the FIDO2 app I just mentioned, which serves as a security key. I think there are a lot of promising things that can be done with these different primitives.

If you have a workflow that requires access to specific websites, secrets, or API keys. OK. So, that was the general presentation general presentation , and now I'll move on to the video. Let me update this. Okay.

Okay update this. Okay. Okay . Steeply. So, here's the video.

Oh. I press play. OK. We sped up the video by 2 times because AI sometimes takes longer to respond, so it's more convenient. So, I'm asking my code to check the balances via the Ledger CLI and determine if there are any revenue opportunities for me.

He is doing it now. At this point, my Ledger device is connected. I have the Solana and Ethereum apps available so I can check these balances. It found 15 USDC in my first account and several income opportunities, so it's interesting that it offers several options. I do n't have much test money, so this recommendation doesn't have to be followed in full, but I'll still put half of it directly through the CLI.

I decided to take their recommendation and now I'm just confirming it on the device. There are two screens on the device : one to confirm the deposit authorization deposit authorization , and the second for the deposit process itself . These two transactions on my device have a clear signature feature so I can see exactly what I'm confirming, as well as transaction verification—this is a simulation service on our backend that verifies that there are no fraudulent smart fraudulent smart contracts and that the transaction will go through exactly as I expect. Here I confirm both actions on the device: authorization and deposit, which are verified on the blockchain using AI. This is just one example of what you can do with the CLI.

And now the second part is second part is encryption, so I ask it to encrypt my encrypt my Entropic API key. This is done using the Ledger keyring in the CLI. He reports that there is no key fob on this car yet . How can I provide this API can I provide this API key? There are several ways to do this.

The AI ​​will advise you to use the clipboard so that it never sees the API never sees the API key itself. But there are other ways, as you will see, he configured my CLI. As a user, I don't need to delve into technical details. I just asked him to encrypt my API to encrypt my API key. It asks to copy it to the clipboard, but in this case I have it in an open file, so I just point it to that file.

Um, and it goes ahead and encrypts it with a decryption key, uh decryption key, uh , and the encryption keys give access to... it deletes the original file, and from now on I don't really need my Ledger device. My agent can, you know agent can, you know , decrypt and encrypt, uh, access tokens as he sees fit. Um, the password is also protected in, uh, a keychain on the computer, right here, so there's an extra layer of security. Uh, and I just asked him for demonstration purposes to show that he didn't compromise that Anthropic API key or use it in any way.

Um, so, uh, here he is, uh, thinking uh, here he is, uh, thinking before giving an answer. Ahem. decrypts the secret and, um, if all goes well, it shows that, essentially, uh shows that, essentially, uh ...um, the agent is able to decrypt, uh, decrypt it securely. Okay, boom.

So, he, uh, got 200 status from Anthropic. So he was able to authenticate and use that API use that API key, and, uh, temporarily use that secret, without ever, uh, you know, reading it, using it, or exploiting it. Uh, so I ask him to prove it again by just running, uh, this command above, and he confirms that, indeed, he is not in the history, or, uh, in the agent's memory. So, agent's memory. So, uh, essentially, he was able to use that, uh, encrypted secret securely by decrypting it in a chain of trust and, uh chain of trust and, uh , not, uh, consuming its, uh, plaintext version.

This is all for demonstration. It's pretty simple here pretty simple here , because I'm proactively , because I'm proactively , uh, communicating with my cloud agents for, for demonstration purposes, but you know, it gets really interesting when your agent does the autonomous work for you. So you can imagine Open Claw, or Grok bot, or Hermes, they're going to need to access these secrets on a more regular basis, and accessing them and using secrets in a secure way, uh, is increasingly relevant today. Good. So, Yes.

And, uh, that brings us to the directions of this hackathon. So, as you know, there are two directions. So we have one called "AI agents x ledger". So, it's about starting something new during a hackathon new during a hackathon , quite simply. And we have a continuity direction, which is essentially expanding on the project that you already have, but finding use cases for the agent stack and the tools that I just showed you.

Um, so here are some examples. We also have a page, linked to at the end of this presentation, on the Ledger developer portal, developer documentation. We have a brand new page that brings it all together . Essentially, the key message here is that you want to implement use cases where agents don't store passwords directly passwords directly , because we've seen that agents can be compromised, and you don't want a compromised agent to give away all your passwords for every service you've used. So the idea is that you can create something during this hackathon to give an agent permission to use your keys without having to hand over the keys themselves directly.

So this point is highlighted here. In the top right corner we have right corner we have , you know, obviously, the key ring, sorry, obviously, Ledger devices have to be connected to your machine to work. So the idea is to implement a key ring on a machine where you can't do that. This could be a VPS, CI runner, or VPS, CI runner, or anything else hosted on a hosting that doesn't have direct access to your device. This is also very interesting.

We have more traditional use cases below, such as agents paying for API paying for API tools. You may be familiar with X 42, for example. The idea here is how can you add Ledger into flows for standalone payments or payments under certain conditions? This is also very useful and interesting very useful and interesting . And the last point, which is more global, summarizes the whole idea, namely—the key message is human participation in the process.

You want the world of agents to continue its work, but you want a human to oversee and approve high- risk actions, such as transferring funds. There are also many ideas here. You can imagine certain thresholds, having a platform where once a certain amount is reached, you are involved in the process and as a person you confirm certain actions. On this slide, this is done intentionally globally. We don't want to go into detail because we're leaving room for creativity, but there's a lot you can do with the Ledger device , always keeping human control where it matters most.

For this category, we have $2,000 for first place, $1,000 for second place, and $500 for third place. So, we have three winners. And this brings us to the second direction, namely—continuity namely—continuity . We have two prizes here: 1000 and 500. So it's more like a continuation of what you've already created.

For example, you have a product and you want to add Ledger to it. This is also very interesting because there are device management tools that make it very easy to connect Ledger devices to your existing applications. We have simplified the Ledger integration process for you . If you look at hackathons a year, two or three ago, it was a real challenge for developers. Today, it's much easier—not just because of AI, but also because of the capabilities we provide and libraries like the ones Oskar showed for wallets, keychains, and so on.

So, use the wallet CLI to configure the key backend in your environment. We've talked about this: it's important not to store passwords and credentials in files, but to protect them. We also have features like device validation when it's important for your product's typical processes. And finally, feel free to browse our repositories. We have many public repositories and you can create PRs for anything you find useful.

This is also part of the continuity for us and for Ledger packages, but it doesn't have to be just your own code. Of course, you can participate in our open-source initiatives. We like it when the line between autonomous actions and those that require your approval is clear, because we are talking about agents working with human input. It is important that the integration with Ledger is real and makes sense real and makes sense . We need to be able to test this.

We will definitely try. We will run your code. It would be good if during your pitch you explained pitch you explained how the product changed "before" and "after" the Ledger integration, and how Ledger improved your product. Help us easily understand the added value of Ledger in your solution. We also ask that you leave feedback on the documentation, as we are constantly improving our developer portal and your feedback is very helpful in doing so.

So, if you can share an experience, it doesn't have to be long, one page is enough. But, you know, how do you like the tools that Oscar just introduced? Are there any gaps? Did you have any difficulties with this or that? Well, with practically anything .

Feel free to make pull requests, as I mentioned earlier. But, yes, if you add feedback on the documentation or any comments, it will help us improve it for future hackers. And yes, I mentioned the QR yes, I mentioned the QR code. So all the information we presented here is on the developers.ledger.

com/ethonline page. The QR code will redirect you there. There are tracks, all prizes, and quick links to the necessary sections of the documentation. Very practical. You have all the necessary commands.

Almost everything to get started here. Thank you very much. Perfectly. Thank you for this very, very good workshop. There are several questions in the chat .

Of course. I can read them to you if that would be more convenient. How does the check for fraudulent smart fraudulent smart contracts work? From Alex contracts work? From Alex .

Yes. Great question Yes. Great question . I think you mean transaction verification. This is a service that we launched in May of last year.

And, in fact, it uses several third-party providers, including Blockaid and Cyvers including Blockaid and Cyvers , to simulate the transaction. There's a lot more going on in the background, like checking to see if certain addresses are associated with fraudulent ones. But one of the main features offered by these providers is scanning the blockchain and checking for any and checking for any potential fraudulent behavior fraudulent behavior . They have a set of rules in the algorithms, but what we do on the Ledger side is quite unique Ledger side is quite unique because we offer, first of all, parallel modeling from multiple providers, which I think is unique for this industry. Usually you rely on one provider.

So if one of them fails, the other can detect and flag threats. And the second important thing is that the simulation results are displayed not at the software level, but at the hardware level. So this makes the system much more secure much more secure and impossible to tamper with, even in the case of very complex attack scenarios. Um, you know, sometimes the interfaces are hacked by the same hackers to claim that the transaction, well, in this case, goes directly to the Ledger device directly to the Ledger device , uh, your phone, because if there's a real threat, it's going to show up there, not on your smartphone or computer. And, yes, I'll just add that this is also an optional feature, and it's very convenient because you can see the reasons why something was flagged as malicious, and you can choose, "Do I want to continue, despite the risk, or do I want to decline?"

Do you understand to decline?" Do you understand ? So we give control to the user, and it's always up to you to decide whether to sign or not. It's just additional information like, "Wow, I'm about to interact with something that's been flagged as harmful." Yes.

All of this is optional and only serves for user comfort, yes. And lastly, this is only available on our newest touchscreen devices like Ledger Stax, Ledger Flex, Ledger Nano S Gen 5, not on Nanos Ledger Nano S Gen 5, not on Nanos if you are using Nano, but you can also simulate this using our various tools, playgrounds, software simulators. OK. That's all. Yes.

Next question: was that demo of getting an API key based on Marquee from this year's Can? Um, I'm not familiar with Marquee from Can. Do you know me, Etienne? No, I don't, I'm not familiar with that. Um, no, but I'll look for it.

Yes, maybe it was some kind of hackathon project for the Can hackathon. OK. Anorak asks, where is the best place for questions about the Electra keyring protocol and bug reports? GitHub issues, Electra live, Discord or security contact or security contact ? I have two reproducible results regarding permission handling.

OK. Um, yes, actually, on the page that you see here in the background with the QR code, there is a group chat at the bottom. Uh, I think it's, of course, direct messages are also possible, but it's kind of a place where you can talk to the whole team that will help you guys. So I think starting from here would be a really good idea. But thank you.

Class. I think ROM Reactor might have a question. Uh , yes. Can you hear me , by the way? Yes.

Yes. Class. Class. Um, yes, I have one simple question and then an additional one to better understand the project we are working on. So, a simple question: will the Ledger Nano X be able to initialize the Ring CLI, as well as encrypt and decrypt?

Of course. Yes. That's good. And an additional question is this: during the hackathon, this was one of the ideas we tried to implement. Of course, you know that giving agents access to your keys is a serious security issue.

And I wanted to better understand how Ring CLI performs encryption and decryption, because after all, in the presentation you showed me, it seemed like the cloud still had access to the plaintext in the context files for initialization, am I wrong? Does the API really never access the cloud context during generation, reasoning, or anything else? Well, I would say that both the Ring CLI and the OpenPGP plugins Ring CLI and the OpenPGP plugins are two primitives for encrypting and decrypting secrets. They work differently. I'm talking about OpenPGP because it's an app that can be found in the developer settings, and it ensures that everything is encrypted or decrypted only when the Ledger is connected, and sometimes after confirmation on the device, which can be configured through the app.

This is quite useful for certain types of secrets, but the Ring CLI itself works differently. It works on the basis of the keyring protocol, which we also use at Ledger to encrypt Ledger account information in the wallet . And this does not happen with a connected device with formal checks on it. This happens through a chain of trust using a regular web connection. So, what you're really using your device to give a certain machine the rights to encrypt and decrypt these secrets, whatever they may be.

So, your Ledger hardware serves as the root for this application. But once initialized, the agent can actually encrypt or decrypt autonomously, and that's better suited for agent workflows and data protection that are relatively data protection that are relatively ... er, I mean, they're less sensitive than transactions like accessing API keys to make a request to the Cloud or something like that to the Cloud or something like that . I will just add that in addition to this, we have implemented a number of additional layers of security, so for the agent to access this and the crypto- this and the crypto- secret, a password tied to your device's OS is required to your device's OS is required . So, you know, there's a whole bunch of points here, but I'll let you test everything out for yourself.

So, to summarize, the agent simply contacts some computer server. That computer has a built-in CLI ring that connects to the chain of trust, so the agent simply asks the machine, "Hello, I have permission from the user to perform an action." "Here's the signature I got from Ledger, or it could be a selfie verification via World ID, whatever." Then this machine performs encryption and decryption on the ledger, does what the agent requested, and sends a response: " sends a response: " Okay, we sent 5 USDC on behalf of the user with his approval for staking or something." The agent simply receives this response and returns to the user with the words, "Hey, I did what you asked."

Of course, I never saw your secrets because I only interacted with the machine where it was all encrypted, right? So in a way it's like Yeah, right. Yes, and it was also here, I think in Yes, top left , for sure. And one more point: the servers of the chain of trust are actually decentralized, and since your secrets are encrypted through your Ledger on your device, they are not accessible directly through this chain of trust. You can say that you control your own secrets in this way.

It's all connected to your Ledger hardware wallet. This is available through devices you have set up yourself with your Ledger, but it is not available externally. Perfectly. Thank you, guys. Yes.

This is very useful. Successful developments. Yes, successful developments. And I see there was one last question regarding the fraudulent smart contract. Eh, this is only for Ledger users, right?

Ledger users, right? Um, uh, because the simulation takes place on the device itself. Or, more precisely, the results are stored on the device itself. Oh, and as I said, this is only for the newest touchscreen devices.