Excited to finally share what we've been working on.
MemWal is a durable, verifiable memory layer that lets agents remember, share, and reuse information reliably.
End-to-end encrypted. Portable across apps. Ownership enforced on chain.
Live on @WalrusProtocol.
Building a memory layer for agents takes time. Redis for caching. S3 for storage. A vector DB for retrieval. And somehow it still doesn’t work right.
Today, we’re excited to announce MemWal: a single, verifiable memory layer for agents — persistent, shareable across systems, and no more fragmented infrastructure. Just memory that works. 🦭
Learn more about MemWal, now on Devnet 👇
Crypto Tumbles as Iran Conflict Escalates, BTC back under $70K, and we're Talking NFTs, AI, & Prediction Markets with Managing Executive of Walrus Foundation @RJ_Simmonds
https://t.co/IJ8BTUWEeo
Why does verifiable data matter for AI?
short answer:
we're handing these ai systems more access and more autonomy than we can currently verify
and i think verifiable data is the cleanest answer we have to that
Rare to find a builder who tested every decentralized storage protocol in production
rarer still to find one who wrote the comparison without flattering the winner
@oceanscarr did both and congrats on the win
There is a huge difference between an LLM that generates responses and one that can actually act in the world.
@GDanezis shares why current infrastructure is already falling short of what comes next 👇
.@karpathy wrote this in 2021, and I think it was fair then
five years later I would say AI and blockchain technology are on course for a grand unification, as @pmarca would say
i'd point at @WalrusProtocol as one of the early signs of it
I like blockchain tech quite a bit because it extends open source to open source+state, a genuine/exciting innovation in computing paradigms. I'm just sad and struggle to get over it coming packaged with so much braindead bs (get rich quick pumps/dumps/scams/spams/memes etc.). Ew
I agree with this take, and I'd push it one step further.
AI systems are shaped by the data they're built on. Models are trained on it, fine-tuned with it, and increasingly retrieve from it at inference time.
If that data still sits on centralized cloud infrastructure, the sovereignty argument breaks down before it reaches the user.
Sovereign AI needs a sovereign data layer underneath it. Otherwise we've just swapped one gatekeeper for another.
"When social media first started rising, we didn't necessarily realize how much of that was actually just foundational infrastructure for the future of humanity.”
@benfielding of @GensynAI on why AI is about to repeat the mistake social media made, and how we can still fix it:
"This isn't just a product that you chat with. It's actually new fundamental infrastructure for humanity, and we should build that learning from the lessons from before. If we let a single company build AI, they hold an enormous amount of power in the world."
"If you actually distribute that out and provide it in a kind of sovereign way — it's on all of our own devices and we own it — then actually you steer the world in a very different social direction."
💭 @GDanezis: "Memory is not anymore just a casual trace of what the agent has done so far. It really is the soul of the agent.
It is its personality, its professional qualifications." 🦭
Facebook, X, TikTok, and every other social platform makes money by targeting ads on data users generate on the platform.
So if users control that data instead, can you still build a business that runs on ads?
George's answer is yes, plus the new paths @WalrusProtocol opens up on top.
Provenance, integrity, availability. These are the 3 properties @WalrusProtocol was built around.
And now in the agentic era, they're becoming the load-bearing properties of everything that comes next.
AI agents are now managing hundreds of millions of dollars. Very soon, it will be billions.
Right now, agents rely on centralized & outdated cloud infra, which has no authenticity proof, a single point of failure, and virtually no control over the data.
Nobody asks where the data came from, whether it was changed, or if it'll be there tomorrow.
The more you research this topic, the more you start understanding @WalrusProtocol's vision of a world where, instead of storing data on centralized servers, it's split up and distributed across multiple independent nodes.
No single entity controls where your data lives or who can access it. That's a major bottleneck for agents that nobody seems to really talk about for some reason.
When agents are making autonomous decisions in the real world, that data needs to be as trustworthy and provable as possible.
The next gen of AI agents is being built in front of our eyes. The data layer powering them will determine if they're here to stay or not.
This is tough to read. Losing the AI skills, memory and context of a 60-person team is something no company should experience.
We're building Walrus so agentic memory is portable, rather than tied to a specific vendor. Yours and available, no matter what.
DMs open, @patomolina–let us know if we can help.
There’s no universal legal framework that treats AI agents as criminally liable actors. We’re heading toward cases where no identifiable human or legal entity can be tied to fraudulent activity, and it’s inevitable that some will try to offload responsibility onto autonomous systems they claim not to control (“that wasn’t me… it was an agent I couldn’t control”).
Combined with deepfakes, we’ll face a broader wave of “deep obfuscation,” along with human–AI hybrids and even enhanced animal cognition. The result is that legal systems and criminal investigations will need to be fundamentally rethought, within a single generation, our human element could look entirely different.
@MaxLensherr This is honest feedback and I appreciate it. We've been having this exact conversation internally.
We built something technically strong and the next phase is about removing every point of friction between a builder and the infrastructure.
That's where our heads are right now.
In the next few years, most financial trades, medical screenings, and legal reviews will be handled by AI systems.
Today, there is no standard way to verify that the AI behind any of those decisions produced the correct output.
That's not a minor gap. It's the single biggest unresolved risk in AI deployment.
Now @OpenGradient is solving it.
Verifiable inference, cryptographic proof of execution, model storage on @WalrusProtocol.
This is what verifiable AI infrastructure looks like in production.
The OpenGradient Model Hub has surpassed 4,000+ models.
Over 4.5TBs of model data is stored on @WalrusProtocol. Each model executes within environments that enable verifiable inference, ensuring outputs can be validated rather than assumed.
Key components of the system:
• TEE and zk-backed execution environments
• Onchain verification of model outputs
• Privacy-preserving data handling by default
This architecture establishes a reliable foundation for deploying and scaling open intelligence systems.
We appreciate the support from Walrus in enabling verifiable, scalable data infrastructure.
We're starting something new: Walrus Sessions🔥
Every two weeks, a focused challenge that puts builders closer to the ecosystem.
Session 1: build your personal site onchain with @walgo_xyz.
A $1,500 prize pool in $WAL to kick it off.
🔥 NEW: Build your personal website onchain with @walgo_xyz and win from a $1,500 prize pool.
Whether it's a portfolio, resume, or interactive experience, if it shows who you are and what you can do, it counts. 🦭