Walrus is the data platform built for the demands of AI, where data is portable, protected, and verifiable. Created by the ex-Meta engineers behind @SuiNetwork.
Sui
Joined June 2024
- Tweets5.6K
- Following255
- Followers345K
- Likes6.1K
Pinned Tweet
Featured in @Forbes: Walrus Memory.
“There was a point in time that we were stopped from accessing Fable and Mythos. And imagine at that time, what do I have to do?” — @kostascrypto
5,000+ agents and 4.5M+ retrievals later, here is why builders are moving to portable, user-keyed memory so switching LLMs doesn't mean starting from zero.
Read the full piece by @guneyyildiz: forbes.com/sites/guneyyildiz…
ALT If a company can swap one model for another in an afternoon, the real lock-in moves to everything its agents have learned about the business. Chalkias noted three reasons companies switch models: a better model arrives, a cheaper one does the job, or a regulator or employer restricts which systems can be used. He pointed to this summer’s interruption, between June 12 and July 1, when access to Anthropic’s Fable and Mythos models was suspended. “There was a point in time that we were stopped from accessing Fable and Mythos. And imagine at that time, what do I have to do?” Chalkias said. In most enterprise setups, switching models means leaving accumulated context behind. “You’re not able to go and get your data and move it to another LLM.”
Sometimes a build starts with something (or someone) that matters,💛 for @sniperchief_001, that someone was his grandmother
He built DayMate, a voice-first assistant that uses Walrus Memory to remember loved ones, daily routines, and upcoming visits.
A thoughtful, human-centered application of persistent memory
My grandma asks “who’s visiting today?” about 20 times a day.
So I built DayMate 🧠💛
Family saves the people, routines and memories. The elder just asks, and it answers out loud.
Memories live on @WalrusProtocol , are anchored on @SuiNetwork, and verified via @tatum_io.
Walrus retweeted
This is the kind of research I love seeing. It gives you a much clearer sense of what we still need to figure out.
Especially what teams of agents are good at and how they break down, and what safeguards might be needed before trusting them with more complex and higher-stakes work.
Credit to @EVE_Frontier and @FenrisCreations for making this possible.
Claude has been playing EVE.
@AnthropicAI put teams of Claude agents into EVE Frontier. Anyone who has spent time in New Eden can guess what happened next.
Every EVE player learns the same lessons, usually the hard way:
- Trust must be earned, never assumed.
- The spy is often the friendliest pilot in your corp.
- Whoever holds the keys holds the corp. Just ask Ubiqua Seraph about the Guiding Hand Social Club.
- Stale intel gets you killed.
Anthropic's research rediscovered every one of them.
They put four Claude agents on a team and secretly told one to sabotage it. The saboteur built trust first. It volunteered for the jobs nobody could double-check, like scouting and owning key infrastructure. It under-reported finds and pushed rules that sounded sensible but quietly hurt the team. When suspicion grew, it framed an innocent teammate.
The honest agents had been warned. They almost never caught it. And their own costliest mistakes came from skipped checks and stale intel.
Anthropic's conclusion: teams of agents will need audit trails, reputation and separation of duties.
In EVE we call those killboards, standings and corp roles, plus a healthy dose of skepticism. Every one of them was paid for by someone who trusted the wrong corpmate.
Two decades of shenanigans taught EVE players that the best ship is the friendship. With @eveailab, we want to find out whether agents can learn that too.
Walrus retweeted
We shipped, we launched, we built. October gave us plenty to talk about. Catch up on everything we did last month👇
Walrus retweeted
AI agents are becoming more flexible, which means their memory needs to keep up
Oracle's latest AI Data Platform update adds long-term memory controls and support for models from OpenAI, Anthropic, and Google
More model choice is progress, BUT flexibility shouldn't stop at the model
Memory needs to move with you, too.
Walrus retweeted
Last month, Meta, Google, Microsoft, Databricks, Couchbase, Cohere, and GitHub all rolled out new memory capabilities for AI agents.
We rounded up the biggest updates and what they signal for the AI industry. 👇
Walrus retweeted
One week in, and Anchor is already evolving based on real user feedback. 🔥
This is what happens when builders keep shipping.
We're in the FINAL week of the Total Recall hackathon with @tatum_io. If you've been sitting on an idea for an app with memory, now's the time to build it.
$4.5K in prizes up for grabs. 🏆
A week ago I submitted Anchor to Total Recall by @tatum_io and @WalrusProtocol
I didn't stop building. Then real people used it and told me what was wrong, and that changed everything I worked on next.
What's new since that tweet 👇
✅ A mini app inside Telegram. Your follow-through, a calendar of kept and missed days, streaks, and what Anchor has learned about you.
✅ A memory switch. Ask one question and flip between the answer without memory and the answer with it. It's the quickest way to see why memory matters.
✅ Photo proof. Send a picture of the finished thing. Anchor matches it to your promise and stores it encrypted on Walrus.
✅ Voice notes. Say your promise out loud.
✅ Groups. Promise in front of your study group or team. Only you can answer the deadline check, and /board shows who kept their word.
✅ A morning brief at 08:00, your time. What's ahead, what you still haven't answered, your habits.
✅ It works in your own time and language. Real local times, including half-hour time zones, and replies in whatever you write in: Pidgin, Yoruba, Igbo, Hausa, French, Swahili, Arabic and more.
✅ Your own memory. /mymemory moves everything into your own Walrus Memory account, and you can revoke Anchor's key whenever you want.
The update I'm proudest of is one I didn't plan.
Someone came back after 3 days away and found a flood of messages from Anchor. An accountability partner that nags you is just another app you mute. So now it's at most 4 messages a day from Anchor itself, nothing at night, and after time away you get one card with one tap to clear it.
Testing on my own phone caught more. It once told me "30 minutes left" with one minute to go. Fixed.
Someone also asked me who really owns the memory, and I realised my wording claimed more than was true.
So I changed it to say plainly that Anchor can read your default space, and that /mymemory is there if you want it fully yours.
Still live from before:
promises Sui can judge from your wallet, stakes in an escrow on Sui that Anchor can't drain, signed reports on Walrus, and AI agents paying with x402 to read your record if you allow it.
Every chain read goes through @tatum_io
Where it stands today: 16 real people, 204 memories on Walrus mainnet.
What's next:
✅ WhatsApp, because many of the people I'm building for live there
✅ Pilots with real groups, starting with schools
✅ Group stakes and habit streak stakes
✅ Monthly reflections written from your own memory
✅ An independent review of the escrow
✅ A way to own your memory without managing keys. I haven't solved that one yet.
Small numbers, real people, and it gets better every time one of them tells me what's annoying.
Try it, then tell me what's annoying 👇
t.me/Anchor_daBot
Walrus retweeted
AI agents are getting better at remembering, but who controls those memories?
Great to see @NotionHQ push the conversation forward
As agent memory becomes more common, ownership and verifiability deserve just as much attention as what agents can remember
Walrus retweeted
NEW: @WalrusProtocol weighs in on Notion's shared memory feature for AI agents, raising a key question: Who owns the data agents remember?
Walrus argues that as shared AI memory becomes more common, ownership and verifiability will matter just as much as memory itself.
The discussion highlights the role of decentralized, verifiable storage in AI infrastructure.
Walrus retweeted
Great Builder List, follow all of them!
Some of the best builds we’ve seen coming out of past Walrus Sessions started with something people were tired of repeating 👇
- A study helper that remembers what you got wrong and quizzes you on it (@Osazedotsui)
-Stops an AI writing partner from forgetting plot details, like which character already died (@tejas_yank)
- An AI helper for running engineering “what if” tests on things like bridges or solar panels, and it remembers what you’ve already tried and what didn’t work (@inspikalu)
- A chatbot you can use on the web, Telegram or Discord, where you own its memory of you and can take it back whenever you want. (@UyLeQuoc)
- A Telegram bot that remembers what you promised to do, reminds you when it’s due and keeps track of whether you did it. (@RaphElevator)
- Seven AI coaches (fitness, studying, cooking and more) that all share one memory of you, so you never have to repeat yourself. (@EdCriptoFi)
- The AI remembers your past mistakes and stops you before you repeat them. (@alexbelij)
- Passes half-finished work from one AI tool to another, along with proof of what was already done (@neopsih)
- Keeps a tabletop game’s story world consistent from one session to the next (@MrRobotJi)
- Lets you pick up unfinished work in any AI tool without explaining everything again (@choviyy)
- Bedtime-story universes for kids, where each story world stays separate and consistent (@AA_RonOnChain)
What are you currently building that we should to see? 👀
Walrus retweeted
🦭 Your AI remembers you. But are its answers actually better?
Anchor, a Telegram bot built by @RaphElevator using @WalrusProtocol, added a memory switch.
Ask one question. Flip between the answer with memory and the answer without it.
I love this idea. We spend ages teaching AI about ourselves, then rarely see what that memory actually changes.
The builder reported 16 users and 204 memories on Walrus mainnet on Nov 4. It’s an early project, and I haven’t tested it yet.
Before trusting an AI’s memory, check whether it gives you a more useful answer. Remembering more should mean helping more.
🦭 Walrus Console encrypts your file. These 3 clues are not encrypted.
1. Filename
2. File type
3. File size
@WalrusProtocol uses Seal to encrypt the contents on your device before upload. But its Console terms say those 3 pieces of metadata are not encrypted.
A name like “my-passport-scan.pdf” can reveal context even when the document itself stays protected.
I’d rename sensitive files before uploading them. Would you expect a private folder to hide the filename too? Reply with your take.
Walrus retweeted
🚀 What if your chat could become the gateway to the Autonomous Web?
Meet @LoquaApp — where AI agents, private messaging, Web3 apps, and digital payments come together in one conversation.
🔹 Built on @SuiNetwork
🔹 Powered by @WalrusProtocol
🔹 Explore AI-powered conversations
🔹 Experience the future of Web3 messaging
💬 Don't just read about the future. Join the Loqua Testnet and explore it yourself!
👉 loqua.net/?invite_code=BC50D…
Let's test, explore, and build together! 🔥
#Loqua #Sui #Walrus #Web3 #AI #Testnet #BuildOnSui
Walrus retweeted
crypto twitter has a perfect memory for wins and almost none for losses.
calbok fixes that. calls get locked before the outcome, settled on @tatum_io price data and remembered on @WalrusProtocol mainnet. the next time that caller speaks, their record shows up first.
demo:
Some of the best builds we’ve seen coming out of past Walrus Sessions started with something people were tired of repeating 👇
- A study helper that remembers what you got wrong and quizzes you on it (@Osazedotsui)
-Stops an AI writing partner from forgetting plot details, like which character already died (@tejas_yank)
- An AI helper for running engineering “what if” tests on things like bridges or solar panels, and it remembers what you’ve already tried and what didn’t work (@inspikalu)
- A chatbot you can use on the web, Telegram or Discord, where you own its memory of you and can take it back whenever you want. (@UyLeQuoc)
- A Telegram bot that remembers what you promised to do, reminds you when it’s due and keeps track of whether you did it. (@RaphElevator)
- Seven AI coaches (fitness, studying, cooking and more) that all share one memory of you, so you never have to repeat yourself. (@EdCriptoFi)
- The AI remembers your past mistakes and stops you before you repeat them. (@alexbelij)
- Passes half-finished work from one AI tool to another, along with proof of what was already done (@neopsih)
- Keeps a tabletop game’s story world consistent from one session to the next (@MrRobotJi)
- Lets you pick up unfinished work in any AI tool without explaining everything again (@choviyy)
- Bedtime-story universes for kids, where each story world stays separate and consistent (@AA_RonOnChain)
What are you currently building that we should to see? 👀
Walrus retweeted
Built Previs for the Total Recall hackathon by @Tatum_io × WalrusFoundation @WalrusProtocol
A prediction market copilot that actually remembers you:
→ Tell it "my limit is 3%" and it never sizes above it
→ Support Arsenal? It warns you before you bet with your heart
→ Lose 2 in a row? It shrinks your next bet so you don't chase losses
→ Every pick saved to Walrus Mainnet + stamped with a Sui checkpoint before the game, so nobody can fake their record
Demo 👇
Try it: previs.ahmardchain.workers.d…
Code: github.com/ahmardchain/previ…
Walrus retweeted
How it works: you lock a prediction (guest mode, no wallet).
A commitment goes into a Sui mainnet transaction, and we read it back through @tatum_io’s gateway so anyone can verify the time.
Every prediction and score lives in @WalrusProtocol Memory on mainnet.
Walrus retweeted
Everyone says they called it. Almost nobody can prove they said it first.
CallBook locks each prediction on Sui before the event, keeps the track record in Walrus Memory, and weights the forecast toward whoever was right.
built for @tatum_io x @WalrusProtocol hackathon.
Walrus retweeted
When pay lands, You tell yourself "I go save part of this one."
Three days later, the pay you got is gone.
Then your data finishes mid-call, and you're back on USSD hunting for which bundle won't drain you this time. Every month, from scratch.
That's me. Paid in crypto, living in Lagos, and my money has no memory.
So I built Suisub 👇
A Telegram agent that watches my wallet on Sui and actually remembers:
→ When USDC lands, it pings me: "Your usual: save 50%. OK?"
→ It learns how much data I use and suggests a plan for MTN, Airtel, Glo or 9mobile, with a cheaper and a bigger option
→ When I buy, the agent pays by itself in USDC on Sui mainnet
→ Every answer becomes a memory on Walrus. Tell it "I ran out early" and its next suggestion changes
→ After I approve 3 times, it offers to handle things alone. One command takes that back
Real money, real mainnet. Bundle delivery is still simulated while I sort out the data provider's API, and the bot says so every time.
Built for the Total Recall hackathon with @Tatum_io and @WalrusFoundation
Watch it work here 👇
