Walrus 🦭/acc Re-poster

Next week at Sui Basecamp, our team is excited to demonstrate to creators the possibilities that emerge when artificial intelligence memory is paired with a clean user interface.

We recognize that although expanding context windows offers an effective solution for temporary scaling needs, achieving genuine agent autonomy relies on maintaining a persistent state that survives whenever an application restarts.

We would love to hear about your own experiences with this challenge. At what specific point do context restrictions begin to cause failures in your current technology stack?

The team at Walrus had an incredibly productive September, and we are excited to share our latest releases with you.

Starting things off, @matterhornso is officially adopting Walrus Memory as its standard backend. Because of this integration, team members can seamlessly share agent conversations, audit results, deployment history, and project context, ensuring all of this information persists between sessions. Additionally, more than 5,000 users of @ASI_Alliance will now gain access to decentralized AI models supported by Walrus.

In other major news, we are seeing $7.7B in Real-World Assets move to our ecosystem, as @Blockticity prepares to introduce over 1M authenticated trade records to Walrus. We also want to congratulate the winners of Sui Overflow. The victorious projects include Carry, which provides verifiable memory receipts, Suize, offering one-call verified sites, and Kraterion, an application for S3 compatible sealed files developed by @StatescuRazvan.

For those looking to get involved, Walrus Session 8 is currently LIVE. You have until October 9 to participate in building chatbots that possess memory capabilities.

Furthermore, we are proud to act as a presenting sponsor for season four of @chain_ofthought. Be sure to tune into the newest episodes, where you can hear @chainguard_dev CEO @dlorenc discuss the reasons why time to exploit has gone negative, alongside @sang_wen from @genspark_ai detailing how to build a second brain for agents.

Finally, if you happen to be in Singapore for Sui Basecamp on Oct 7 and 8, we would love to connect. Feel free to visit the Walrus booth in person or watch our Head of Product, @kimblgn, speak on stage.

We have seen incredible progress throughout our ecosystem this past month, with developments ranging from integrated agent memory systems on Matterhorn to the introduction of $7.7B in real-world assets. Here is a comprehensive look at our most recent launches and updates.

In an exciting collaboration, @matterhornso has chosen Walrus Memory to serve as its primary backend solution. This integration ensures that team members can seamlessly share and retain agent conversations, audit outcomes, deployment logs, and overall project context from one session to the next. Additionally, we are thrilled to share that more than 5,000 individuals using @ASI_Alliance will now have access to decentralized artificial intelligence models supported directly by Walrus.

Expanding into massive real-world value, @Blockticity is migrating over 1M authenticated trade records to the Walrus platform, representing a staggering $7.7B in real-world assets. We also want to congratulate the champions of the Sui Overflow competition. The winning projects include Kraterion developed by @StatescuRazvan which features S3-compatible sealed files, Suize which offers one-call verified sites, and @CarrySui which provides verifiable memory receipts.

For developers looking to build, Walrus Session 8 is currently active until October 9, inviting participants to create memory-enabled chatbots. In media news, Walrus Memory is proud to act as the presenting sponsor for the fourth season of the @chain_ofthought podcast. Be sure to tune into the newest episodes where @sang_wen from @genspark_ai discusses creating a second brain for autonomous agents, and @dlorenc, the CEO of Chainguard, explains the phenomenon of negative time to exploit.

If you are attending Sui Basecamp in Singapore next week, please come find us. You can visit the Walrus exhibition booth or watch our Head of Product, @KimberlyLogan, speak on the main stage.

To receive all of our comprehensive monthly updates directly, please subscribe to our newsletter by visiting https://walrus.xyz/newsletter/

It can be frustrating when your artificial intelligence sessions reset, requiring you to clarify your project context from the very beginning. To solve this problem, Walrus and @matterhornso are introducing persistent memory to blockchain workflows that utilize AI assistance. With this new integration, your progress is securely maintained across different teams and future sessions, saving you from ever having to start over again. You can view the complete announcement and learn more at the following link: https://blog.walrus.xyz/matterhorn-walrus-persistent-memory-ai-workflows/

Although artificial intelligence agents are capable of executing tasks, it is your data that truly unlocks their potential. Every piece of context, historical record, file, and detail gathered by your applications and agents serves to build your unique competitive edge. As we progress through the #AI era, it is essential for this data to remain fully portable and travel alongside you. Be ready for October 7th, when we will introduce an innovative approach for linking your data seamlessly between various applications and artificial intelligence tools.

Please join our Discord community tomorrow for an engaging AMA session featuring Satyam Bansal from @redsentinel_ai.

During this event, we will explore the ways Red Sentinel develops AI security on Walrus and @SuiNetwork. We will also unpack the mechanics of their architecture and discuss the critical importance of verifiable context. As an added bonus, attendees will get an exclusive preview of what to expect in the upcoming Walrus Session.

Mark your calendar for September 30, 2026, at 12:00 PM UTC. You can access the conversation by visiting the Walrus Discord at http://discord.gg/walrus. We hope to see you there.

In preparation for Sui Basecamp, our team is outlining the future trajectory of stateful AI infrastructure. While artificial intelligence systems are constantly improving, they still lose all contextual information as soon as a conversation closes. We want to hear your perspective on this challenge. If the issue of persistent memory remains unresolved by 2027, what do you believe will be the most critical workflow breakdown caused by AI agents? Please share your thoughts with us in the responses.

Right now in Amsterdam, @kostascrypto and the broader Walrus team are present at @HumanXCo. Our current focus is integrating Walrus Memory into the very heart of the AI stack, which directly resolves the problem of context loss for multi-agent systems.

#HumanX2026 #AI #WalrusMemory

During a recent discussion in the @HumanXCo room, an audience member asked if the technology is designed for everyday consumers or enterprise clients. @kostascrypto replied by highlighting that businesses will likely become the most prominent users. He explained that corporations have a distinct need to selectively manage data access based on their geographic region, their specific applications for LLMs, and several other factors.

Moving to the personal side of the equation, he observed that individuals frequently share deeper secrets with their AI agents and LLMs than they would with their own partners or best friends. For these everyday users, Walrus Memory acts as a shield to secure that private information while giving them full control over when to relocate or limit access to their data.

Many people hesitate to transition between different LLMs simply because they worry about leaving behind the valuable context and conversation history they have already built up. Addressing this common concern during a live presentation at @HumanXCo Amsterdam, @kostascrypto showcased a seamless solution. He provided a demonstration of Walrus Memory, illustrating exactly how the platform simplifies the process of changing models by allowing you to effortlessly bring all your data along with you.

Imagine the implications if the permanent memory of your AI agent inadvertently recorded the actual name of a client. During Walrus Session 7, @UyLeQuoc tackled this exact scenario by conducting an audit on the cross-tool memory prompt known as Continuum. This thorough review led to the successful resolution of two crucial vulnerabilities regarding privacy and ranking.

To address these issues, the first solution introduced local client codenaming, which guarantees that personal identities are kept securely off-chain. The second improvement established monthly namespaces, a change designed to prioritize time-based sorting rather than relying on semantic rank.

You can explore the comprehensive security audit and access the prompt for yourself at this link: https://inkray.xyz/article?id=i-gave-my-ai-a-memory-it-can-never-delete-then-i-read-what-it-was-about-to-write-701fd1d0f587c800

When an agent pulls up contextual information, is it possible to demonstrate exactly where that data originated or confirm that the system had permission to view it? Right now, the memory systems of most agents function entirely as a black box.

In Build Notes #3, we explore how this challenge is being addressed by Carry, a winning project from Sui Overflow. By introducing pre-retrieval access control alongside verifiable answer receipts, Carry brings these important capabilities directly to Walrus Memory.

You can dive into the complete technical breakdown by visiting our blog at https://blog.walrus.xyz/carry-verifiable-agent-answers-built-on-walrus-memory/

The inner workings of most agent memory systems currently resemble an opaque black box. Whenever an agent retrieves specific context, proving the exact origin of that data or confirming that the access was properly authorized presents a significant challenge.

To address this, Build Notes #3 explores Carry, a Sui Overflow winner that is equipping Walrus Memory with pre-retrieval access control alongside verifiable answer receipts.

You can dive into the complete technical breakdown on the blog by visiting https://blog.walrus.xyz/carry-verifiable-agent-answers-built-on-walrus-memory/

Have you ever considered how the memory of most agents operates completely out of sight? When an agent retrieves specific context, it is often difficult to demonstrate the exact origin of that data or confirm whether the access was actually authorized. We address this challenge in Build Notes #3 by examining Carry. As a recognized Sui Overflow winner, Carry integrates pre-retrieval access control and verifiable answer receipts directly into Walrus Memory. You can read the full technical breakdown on our blog at https://blog.walrus.xyz/carry-verifiable-agent-answers-built-on-walrus-memory/

We are incredibly proud to sponsor @chain_ofthought for the current season. In case you happened to miss it, the COO of @genspark_ai, @sang_wen, recently joined @chain_ofthought for an insightful discussion regarding persistent context. The conversation delved into the ways their second brain manages memory decay, the mechanics of updating state dynamically, and the process of capturing real-world room audio for agents. You can watch the complete episode through this link: https://youtu.be/ED3m6A7ScKA?si=qM-_kQe7okEcgmQr

Have you ever dedicated a complete D&D session to defeating a dark wizard, only for your AI Dungeon Master to resurrect the villain three sessions later due to a simple memory lapse? One of the winners from our Walrus Session 7, @MrRobotJi, successfully tackled this exact frustration by developing the Aetheris World Engine utilizing Walrus Memory.

Rather than leaning on unpredictable fuzzy text similarity, this innovative tool securely logs your ongoing game state directly onto an immutable ledger hosted on Walrus Mainnet. This approach guarantees that essential game elements are tracked flawlessly. For example, it strictly monitors NPC status so that deceased characters permanently stay dead. Additionally, the system accurately logs world-tick timelines to completely prevent any timeline bugs from occurring. The engine even incorporates SEAL-encrypted secret notes, which securely conceal your upcoming plot twists from the players.

You can explore the full details of this project by reading the article over on @Medium at https://medium.com/@mrrobotji/the-dead-boss-who-wouldnt-stay-dead-how-we-built-an-immutable-tabletop-world-engine-on-walrus-a7651697df9c?postPublishedType=initial

We are excited to share that Walrus Session 8 is officially underway, and the focus this time is on Chatbots That Remember. Between Sep 18 and Oct 9, you are invited to compete for your share of a $2,000 WAL prize pool. Taking part is straightforward. First, introduce a persistent state to your existing bot. Next, ensure it is deployed live. Finally, submit a write-up that showcases the before and after progress of your creation. For comprehensive guidelines and submission details, please head over to https://www.deepsurge.xyz/hackathons/c0141a4a-21be-4009-bc63-7c168608c849

What occurs when a new AI session simply trusts the information left behind from a previous execution? Often, it ends up resuming operations on a broken state.

To fix this issue of unverified states during agent handoffs, @neopsih utilized Walrus Memory to upgrade Markov, an accomplishment that made them one of our Walrus Session 7 winners.

With this new integration, handoffs have been transformed into self-proving objects. This means that receiving agents are now obligated to verify terminal receipts and check data integrity before they are allowed to accept a new state.

Find out all the details over on @Medium: https://medium.com/@anna.stolbovskaja/i-gave-a-fresh-agent-a-perfect-handoff-and-it-was-still-wrong-6020fa9501b5

While a prompt history informs an agent about the events of a single session, memory provides an enduring state that continues functioning through system restarts, different tools, and multiple runtimes. We invite you to examine persistent state execution alongside us at @HumanXCo on Sept. 23 at 3:00 p.m. CEST. Make sure to put the @KostasKryptos Masterclass on your calendar.