Walrus is one. The agentic web is what’s next
How Walrus grew in 2025 — Mainnet, new features, and the beta launch of Walrus Memory for AI agents.
Walrus is the Verifiable Data Platform built on trust, ownership, and privacy, giving users and developers verifiable control over their data. In 2025, that foundation extended to AI agents through Walrus Memory — a portable, verifiable long-term memory layer, that lets agents store context, recall past decisions, and share memory across apps, persistent and under builders' control.
A year ago, Walrus went live.
What followed suggested more than a successful launch, where the technology worked reliably from day one. From the start, we saw builders doing more than just observing: they were experimenting.
But if the first year was about showing what Walrus can do, the next phase is about something bigger: what it enables for AI, finance, and more.
We spoke with Rebecca Simmonds, Executive Director of the Walrus Foundation, and Adeniyi Abiodun, Co-Founder and CPO at Mysten Labs, about what’s changed since Mainnet launch, what surprised them most, and why Walrus is poised to serve as core infrastructure for the agentic web.
Verifiability is the way forward
The cost of creating content is approaching zero, driven by AI and increasingly powerful developer tooling. Anyone, or anything, can generate massive amounts of data instantly.
But as data becomes cheaper and more abundant, it also becomes riskier. As more systems and AI agents start making decisions on our behalf, bad data stops being a nuisance and starts becoming a real cost.
As Rebecca put it, “payments are going to be made agent-to-agent with no human beings in the middle of them. That’s great, but also scary.”
In the future, verifiability will be more than just optional. It will be a foundational part of all new systems and structures going forward.
“The verifiability that Walrus offers is absolutely crucial,” she adds, “and the world can’t really progress without it.”
Planning for an agentic future with portable memory
Over the past year, Walrus has supported a wide range of use cases, from prediction markets to wearable technology. But as patterns emerged, one area started to stand out: AI agents.
Adeniyi points to a broader shift already underway: "The majority of internet traffic is largely automated and robotic, but now we believe we're in a world where most transactions on the internet are going to be autonomous and via agentic workflows."
Agents today can act, but their memory is fragmented and hard to rely on. What they remember often lives locked inside a single platform, where it can't be verified, moved, or shared.
That's a gap Walrus can fill.
With the launch of Walrus Memory, we introduced a portable, verifiable long-term memory layer for AI agents. Agents can store context, reference past decisions, and carry memory across apps and workflows — persistent, verifiable, and under builders' control. This becomes critical as agents take on more complex roles, from interacting with financial systems to coordinating across applications.
Without reliable data and memory, those systems break down quickly. And as agents become more capable, Walrus is ready to become part of the foundation they rely on.
The Walrus Protocol: shaped by builders
Walrus was initially built with large, unstructured data in mind. But as builders got their hands on it, a different pattern started to emerge. Developers weren’t just uploading big files. They also wanted to upload many small ones.
At the time, that wasn’t economical. So the team moved fast to respond.
Quilt was introduced just a few months after launch, giving developers a native way to bundle small files efficiently. Grouping files together significantly reduces overhead, bringing storage costs down by over 100× for ~100KB files and up to 400× for very small files. This resulted in a savings of more than 3+ million WAL for Walrus partners.
This is an approach the team applies to everything we do: pay close attention to how builders use Walrus in practice, and respond quickly with solutions that remove friction.
Year 2: Scaling more use cases
If the first year of Walrus showed what was possible, the next phase is about scaling what works with the tools that are making building faster and easier than ever.
“If you are looking to actually build something, I would start with a real use case and move quickly into something people can actually use,” says Rebecca. “Vibe code something up and get it out there. Walrus is built for speed, scale and cost efficiency-I recommend leaning into that.”
But speed alone isn’t enough. The bar is higher.
Adeniyi emphasizes building things that make a big impact: “Always focus on products where it’s really key for the value chain…where if your product was to be removed from the flow, that flow would completely not work anymore.”
That opportunity is starting to take shape. Over the next year, we expect to see real categories emerge, especially where mission critical data and AI intersect. And as more of the internet shifts toward agent-driven systems, the role of infrastructure becomes even more important.
We’re excited to see how Walrus-from our developer community, partners, and beyond-help make the agentic future a reality.
Follow us on X for the latest →
FAQs
What is Walrus Memory?
Walrus Memory is a portable, verifiable memory layer for AI agents. It lets agents store context, recall past decisions, and share memory across apps and workflows – persistent, verifiable, and under builders' control – instead of stitching memory together from databases and vector stores.
How does Walrus Memory relate to Walrus?
Walrus Memory is built on Walrus, the Verifiable Data Platform. It applies the same foundation the platform spent 2025 building – verifiable data, ownership, and access control – to the specific problem of agent memory, so the data agents store and recall carries the same guarantees as anything else on Walrus.
Why do AI agents need verifiable memory?
As agents move from demos into autonomous, long-running workflows, the decisions they make are only as trustworthy as the memory behind them. Verifiable memory lets builders trace which data and reasoning produced an action, and confirm that context wasn't altered – essential as agents take on roles in financial systems and coordinate across applications.
About Walrus Memory
Walrus Memory is a portable memory layer for AI agents. It enables agents to operate reliably across apps and sessions without losing context — portable, verifiable, and fully under builders' control. Walrus Memory lets agents handle complex, long-running workflows and coordinate using data they can trust.
