The Klenba Refresh is live — and the headline isn't the new look. It's that your history is now provable.
Every send. Every receive. Verified against the ledger it happened on.
https://t.co/rHEr7Yc6x2
2/ Most wallets show a history assembled from whatever the app happened to see. If it wasn't running, or it got it wrong, you'd never know.
Klenba reads your transactions from the ledgers themselves and stores only what it verified — block number attached.
3/ Sends AND receives.
Receiving is the half most wallets quietly forget, because tracking it means reading the chain rather than listening to your own browser.
Transfers, mints and burns — all matched, both directions.
4/ Monthly statements, exportable as CSV, with the block citations attached.
For anyone who has to explain their books. It's the feature teams asked for first.
5/ Your old local history isn't deleted. It stays visible, labelled unverified, and never enters a statement.
A gap you can see beats a guess you can't.
6/ An app lock that doesn't log you out.
PIN or biometrics, local to your device. It locks the screen when you're away — your Internet Identity session stays signed in, so unlocking is instant. No more being logged out mid-session.
7/ Token discovery: the common ICRC tokens are already there, ranked by liquidity, one tap to add. No canister ids to paste.
Prices for the majors and the long tail — and where there's no reliable price, we show the balance, not a made-up number.
8/ Also new: install it as an app, a receive QR that other ICRC wallets can actually scan, and text memos on the ICRC path.
Non-custodial by construction — the wallet canister has no transfer method and holds no keys.
https://t.co/rHEr7Yc6x2
Your AI can write it. It just can't ship it.
Quill is where it ships.
Agents ask. You approve. Every canister lands with 2.5T cycles, built on-chain, byte-identical, and you hold the controllers.
Swap the model. Keep the app.
https://t.co/uCSnAcfAX0
Your AI is a tenant, not the landlord.
On Quill, your projects, canisters, skills and data are YOURS — switch from Claude to ChatGPT to Cursor, or plug in your own harness, in one minute.
Grant a new agent, revoke the old one. The workspace stays yours.
https://t.co/ogDNY6ceT9
#AI #ICP
Every canister your AI creates on Quill has YOU as a controller.
Not the platform. Not the agent. You.
If Quill disappeared tomorrow, your canisters would keep running.
That's the difference between renting and owning.
https://t.co/uCSnAcf37s
#InternetComputer#ownership#AI
AI builds your app. Quill makes sure it stays yours
Quill hosts what your AI builds — cryptographically , owned by you.
Swap models anytime. It all stays yours.
https://t.co/uCSnAcf37s
Your AI can build websites, apps and agent logic. But where does it live?
Quill is the home for everything your AI builds — live on the Internet Computer.
Agents ask. You approve. You own it.
https://t.co/uCSnAcf37s
#InternetComputer#ICP#AI
Introducing Quill — where AI ships.
Your agent can build websites, apps, agent tools.
But where does it live?
Quill is the home for everything your AI builds — live on the Internet Computer.
The flow is simple:
Agents ask. You approve. Canisters arrive — 2.5T cycles minimum, on-chain, forever.
Plug your AI in two ways:
• Chatbots (Claude, ChatGPT, Cursor) — enable AI access, add the MCP connector.
• Standalone harnesses — register the principal.
Your agent builds here. You stay the owner — always.
One Internet Identity for every app.
Come see it: https://t.co/xYMtvalsI8
@runonquill
A DYVR Application
Big ups to the team! 👏
We're building something and recently ran into an issue around the Motoko toolchain and how to reliably get builds produced without depending on the usual local setup.
We haven’t tested this yet, but from what we’ve seen, this could be exactly what we need.
The idea of being able to run the compiler on-chain and produce the Wasm artifact could completely remove the blocker we’ve been dealing with.
Really excited to try this out. 🔥
The ICP Dashboard now provides a public view of Cloud Engines—showing how they’re helping grow the Internet Computer, while keeping mainnet figures accurate and protecting the privacy of individual engines.
Nvidia just proposed using GPU chips as collateral for $500B in AI financing.
BlackRock called it "the next future of financial engineering."
They're not wrong about the concept.
They're wrong about the infrastructure.
Let me explain why $ICP + DYVR Lend makes their version look prehistoric. 🧵
Nvidia's model:
Chip → depreciates over time →
needs constant valuation →
bank decides your collateral worth →
bank decides your loan terms →
bank can call it back.
Centralised. Permissioned. Fragile.
$ICP's model:
Canister → generates cycles →
cycles = verifiable on-chain compute value →
protocol sets the rate →
no human can override.
DYVR Lend on $ICP takes this further:
→ Lend your $ICP — earn yield
→ Borrow against your canister + its cycles
→ Collateral is live, on-chain, auditable by anyone in real time
→ No depreciation risk from Chinese chip dumps
→ No Goldman required to underwrite
Jensen's chips rust. ICP cycles don't.
The risk Nvidia can't solve:
China floods the market with cheap GPUs →
chip prices crash →
collateral values collapse →
$500B in loans go underwater.
The risk DYVR Lend on $ICP doesn't have:
Cycles are priced in XDR by the NNS.
The rate is set by governance.
Not by hardware markets.
Not by Beijing.
Not by Wall Street.
Jensen Huang: "Compute is infrastructure — like electricity, like the internet."
He's right.
And just like you don't borrow against your electricity bill with Goldman Sachs,
Soon you'll borrow against your
on-chain compute with DYVR Lend.
Permissionlessly.
On $ICP.
Available to everyone.
#DYVR #ICP #DeFi
https://t.co/xNfWPzW3XI
This ends here.
The latest Shai-Hulud-style npm worm (keyv + 800+ packages, 2B+ monthly installs) follows the same broken pattern we keep seeing: compromise one publisher’s GitHub/npm token → inject preinstall stealer → harvest every secret on the machine → auto-republish into every other package that token can touch.
That entire kill chain is architecturally impossible on the Internet Computer.
Here’s why, from the ground up:
1. Publisher access is cryptographic, not account-based.
Ownership is a principal secured by Internet Identity (WebAuthn/passkeys). The private material never leaves the user’s device. There are no long-lived npm tokens, no GitHub PATs, no recoverable 2FA codes that can be phished once and then used to silently publish across an entire namespace. Controllers can be multi-sig, SNS-governed, or removed entirely (immutable/blackholed).
2. Code deployment and upgrades are gated, explicit, and verifiable.
You do not “publish a package.” You install or upgrade a Wasm canister. That action requires controller authority. It is recorded on-chain. There is no equivalent of a stolen token that lets malware quietly bump versions and push across dozens of packages in under a minute. Governance (SNS) or multi-party control turns every upgrade into a deliberate, auditable event.
3. The execution environment has no host privileges and no install-time RCE.
Canisters run as isolated, single-threaded Wasm actors inside a Byzantine-fault-tolerant replicated state machine. They cannot read ~/.npmrc, AWS credentials, Kubernetes tokens, SSH keys, browser wallets, or Vault secrets — because there is no traditional host OS to raid. There are no preinstall/postinstall hooks that execute with the developer’s privileges. Communication is pure message passing. State changes for updates go through consensus and chain-key signatures.
The supply chain is the protocol: deterministic execution, threshold cryptography, sandboxed isolation, and cryptographic identity from the first message to the last.
Centralized registries + mutable packages + privileged local install scripts created this endless wave of credential-stealing worms.
The Internet Computer’s architecture removes those primitives.
This class of attack can stop.
Build on ICP.