Gitbank

Smart contracts that listen to GitHub. Lock budgets, escrow bounties, pay on merge.

[Gitbank @Gitbank_io]

Gitbank vault is now live on @grok at https://t.co/ctccYeFQ8a

Manage your on-chain vault on Base Mainnet directly from Grok chat. No wallet popup, no config file. Just add the MCP connector and go.

How to connect Gitbank on Grok:
1. Open https://t.co/ctccYeFQ8a
2. Go to Connectors tab
3. Click "Add MCP Server"
4. Paste: https://t.co/ce3IRro767
5. Done. Gitbank tools are live in your chat.

What you can do with Gitbank on Grok:
- Check vault balance (WETH, USDC on Base)
- Request deposit / withdraw
- Swap tokens inside your vault
- Transfer to another GitHub user's vault
- Assign bounty to a contributor
- Check last transactions
All writes require @gitbankbot confirm on GitHub.
Security stays on your GitHub account.

Live test on Grok:
https://t.co/4SimSs75S7
Full setup guide: https://t.co/Fe1pS7LaYt

[Gitbank @Gitbank_io]

Gitbank x @OndoFinance RWA Integration
We are building gitStock, a soul-bound tokenized stock layer on Base L2, powered by Ondo Finance RWA assets on Solana.

GitHub contributors can hold, buy, and sell tokenized real-world stocks directly inside their Gitbank vault. No wallet. No gas. No sign-up.

Works from anywhere you already use AI:
- GitHub Issues / PRs via @gitbankbot mention
- Claude (MCP)
- Cursor (MCP)
- Grok (MCP)
- Gemini (MCP)
- Any MCP-compatible AI client

Tradeable assets (currently under build and testing):
$SPCX - SpaceX
$NVDA - NVIDIA Corporation
$AAPL - Apple Inc.
$TSLA - Tesla Inc.
$META - Meta Platforms
$MSFT - Microsoft Corporation
$GOOGL - Alphabet Inc.
$AMZN - https://t.co/B98N1mAfyR Inc.
$CRCL - Circle Internet Financial
$SPY - SPDR S&P 500 ETF Trust
$QQQ - Invesco QQQ Trust

How it works:
- Bridge USDC from Base to Solana via CCTP V2
- Swap USDC to Ondo tokenized stock via Jupiter v6
- Mint soul-bound gitStock token (ERC-20) on Base to your vault
- Non-transferable. Cannot be phished or drained via approvals.

Building in public:
https://t.co/l7tXkFMgYz

Currently under build and testing.

[Gitbank @Gitbank_io]

Testing Gitbank MCP with @NousResearch Hermes Agent and hit a problem.

Our live MCP (https://t.co/AD0mEhW20E) exposes 16 tools. Models that aren't frontier-scale struggle with tool selection at that count and fall back to web search instead of actually calling the vault.

So we're building https://t.co/TSw2RslbsT (not live yet). One tool. Plain language in, NLP runs server-side, right action gets dispatched.

"check my balance" works.
"swap 10 USDC to WETH" works.
"buy 50 USDC of NVDA" works.

What you'll be able to do from Hermes once it ships:
1. Vault balance (USDC, WETH on Base)
2. Deposit / withdraw / swap (Uniswap v3)
3. Send to any GitHub user
4. Live stock prices via Pyth oracle
5. Buy/sell tokenized stocks
6. Launch tokens via Clanker
All onchain on Base. Zero gas for users.

We're building Gitbank as the security layer for AI agents handling real assets and wallets. What do you guys think of this approach? @Teknium @NousResearch

[Gitbank @Gitbank_io]

A bank without security is just an accident waiting to happen for AI agents

@Gitbank_io introduces the first GitHub-anchored vault on @Base, AI queues the tx, YOU confirm it on GitHub (with your 2FA)

Already works via MCP on @grok and Claude. No wallet popup. No gas.

https://t.co/v1rFTqoD5S

Check it:
https://t.co/uVDU9GcM0n

https://t.co/5FQSd3ZSmw

[Gitbank @Gitbank_io]

Anthropic just released Claude Mythos. Every AI agent holding crypto is now a target.

Today, Anthropic launched Claude Fable 5 and Claude Mythos 5, the first broad public release of Mythos-class capabilities. Claude Mythos is the model Anthropic previously made available only to ~50 vetted partners inside Project Glasswing. It can identify and exploit zero-day vulnerabilities in real-world software, autonomously, with no human in the loop after the first prompt. On expert-level security tasks that no model could complete before 2025, it succeeds 73% of the time.

We already have proof of what this class of model can do.

On May 28, Anthropic released Claude Opus 4.8. Within 24 hours, a security researcher used it to find a critical soundness bug in Zcash's Orchard zero-knowledge proof circuit, a flaw that had existed, undetected, for four years. The exploit allowed forging unlimited counterfeit ZEC inside the shielded privacy pool, with no cryptographic trace. ZEC fell more than 30% in a single day.

That was Opus 4.8. Mythos is the model Anthropic considered too dangerous to release publicly until today.

The implication for AI agents holding crypto assets is serious.

Most AI agent wallets are a private key stored in an environment variable or a database. If the key leaks, through a server breach, a dependency exploit, a prompt injection, or a compromised log file, the attacker has unconditional access. There is no second factor. No confirmation step. No way to stop an outbound transaction already signed.

Anthropic and its Glasswing partners have already found over 10,000 high or critical vulnerabilities across the most important software in the world. More than $840 million was drained from DeFi in the first five months of 2026 alone. These two facts are now on a collision course.

Gitbank was designed around this assumption from the start.

GitAssets (gitWETH, gitUSDC) are soul-bound ERC-20 tokens. There is no transfer(), no approve().

But the deeper protection is the confirmation model. Every vault operation, deposit, withdraw, swap, bounty assignment, requires a GitHub webwebhook event signed by GitHub's own servers from your account. The execution keypair encrypted at rest with AES-256-GCM and decrypted in memory for under 200 milliseconds per transaction. Even if the private key is extracted, the attacker still needs your GitHub account to post a confirmation comment. Even if the MCP endpoint is compromised. Even if your AI chat log leaks.

GitHub account security, YubiKey, passkey, 2FA, is the only thing that matters. It is the one auth layer that cannot be stolen from a database or an env file.

Gitbank exposes all of this through a native MCP server. The same Claude Mythos that is now publicly available can query your vault balance, check project budgets, and queue transactions, all inside the conversation, all gated by GitHub identity.

https://t.co/v1rFTqoD5S. Apache-2.0, live on @Base Mainnet.

[Gitbank @Gitbank_io]

Gitbank just shipped an AI agent wallet on Base Mainnet accessible from @grok and Claude via MCP
Most AI wallets give agents your private key.

One prompt injection = funds drained instantly.
Gitbank flips the model:
1. AI queues the transaction
2. YOU confirm it on GitHub (with your 2FA)
3. Vault executes on Base Mainnet
Soul-bound. Non-custodial. Zero gas.
Your GitHub security = your vault security.

Check it: https://t.co/uVDU9GcM0n

https://t.co/SXgQj2iZBZ

[Gitbank @Gitbank_io]

Anthropic just released Claude Mythos. Every AI agent holding crypto is now a target.

Today, Anthropic launched Claude Fable 5 and Claude Mythos 5, the first broad public release of Mythos-class capabilities. Claude Mythos is the model Anthropic previously made available only to ~50 vetted partners inside Project Glasswing. It can identify and exploit zero-day vulnerabilities in real-world software, autonomously, with no human in the loop after the first prompt. On expert-level security tasks that no model could complete before 2025, it succeeds 73% of the time.

We already have proof of what this class of model can do.

On May 28, Anthropic released Claude Opus 4.8. Within 24 hours, a security researcher used it to find a critical soundness bug in Zcash's Orchard zero-knowledge proof circuit, a flaw that had existed, undetected, for four years. The exploit allowed forging unlimited counterfeit ZEC inside the shielded privacy pool, with no cryptographic trace. ZEC fell more than 30% in a single day.

That was Opus 4.8. Mythos is the model Anthropic considered too dangerous to release publicly until today.

The implication for AI agents holding crypto assets is serious.

Most AI agent wallets are a private key stored in an environment variable or a database. If the key leaks, through a server breach, a dependency exploit, a prompt injection, or a compromised log file, the attacker has unconditional access. There is no second factor. No confirmation step. No way to stop an outbound transaction already signed.

Anthropic and its Glasswing partners have already found over 10,000 high or critical vulnerabilities across the most important software in the world. More than $840 million was drained from DeFi in the first five months of 2026 alone. These two facts are now on a collision course.

Gitbank was designed around this assumption from the start.

GitAssets (gitWETH, gitUSDC) are soul-bound ERC-20 tokens. There is no transfer(), no approve().

But the deeper protection is the confirmation model. Every vault operation, deposit, withdraw, swap, bounty assignment, requires a GitHub webwebhook event signed by GitHub's own servers from your account. The execution keypair encrypted at rest with AES-256-GCM and decrypted in memory for under 200 milliseconds per transaction. Even if the private key is extracted, the attacker still needs your GitHub account to post a confirmation comment. Even if the MCP endpoint is compromised. Even if your AI chat log leaks.

GitHub account security, YubiKey, passkey, 2FA, is the only thing that matters. It is the one auth layer that cannot be stolen from a database or an env file.

Gitbank exposes all of this through a native MCP server. The same Claude Mythos that is now publicly available can query your vault balance, check project budgets, and queue transactions, all inside the conversation, all gated by GitHub identity.

https://t.co/v1rFTqoD5S. Apache-2.0, live on @Base Mainnet.

[Gitbank @Gitbank_io]

@TendrasiArief @xai @grok Yes. Very close now, we’re wrapping up final testing.

[Gitbank @Gitbank_io]

Perfect timing @xai

Gitbank MCP drops on @Grok today, GitHub devs can ask https://t.co/rQ0TgyK68z to deposit, swap tokens, and assign on-chain bounties directly from chat.
Soul-bound vault on Base. Zero gas. GitHub ID as identity.

Add it: https://t.co/ctccYeFQ8a → Connectors → Gitbank

[xAI @xai]

The @MongoDB plugin is live in the Grok Build Plugin Marketplace.

Explore data, optimize database performance, and build high performance vector search systems with a single prompt.

[Gitbank @Gitbank_io]

just tested base:0xc21dd0ee043930711c2a3e55f39c7d3144d09b07 connector @grok internally. it works 🔥

very soon you can use Gitbank MCP, Gitbank Vault, and all Gitbank features directly on https://t.co/ctccYeFQ8a

deposit, swap tokens, assign GitHub bounties - all in natural language, on @base Mainnet
no wallets, no gas, just Grok.

https://t.co/5h2ks1cX5q

[Gitbank @Gitbank_io]

just tested base:0xc21dd0ee043930711c2a3e55f39c7d3144d09b07 connector @grok internally. it works 🔥

very soon you can use Gitbank MCP, Gitbank Vault, and all Gitbank features directly on https://t.co/ctccYeFQ8a

deposit, swap tokens, assign GitHub bounties - all in natural language, on @base Mainnet
no wallets, no gas, just Grok.

https://t.co/5h2ks1cX5q

[Gitbank @Gitbank_io]

base:0xc21dd0ee043930711c2a3e55f39c7d3144d09b07 just shipped two updates.

LANDING PAGE
1. New interactive hero: 4 tabs showing Vault, MCP, Bounty, and Base.
2. Smart day/night theme based on your local time. No setting needed.
3. Two news banners pinned at top: LIVE MCP and NEW Base MCP.

DOCS PAGE
1. New "MCP Clients" section with 5 pages: MCP Overview, Claude Desktop, Cursor, Grok, and Base MCP (EIP-5792).
2. Claude Desktop and Cursor are live. Grok is early access.
3. Supported assets corrected: WETH and USDC only.
CLI and SDK marked Coming Soon.
4. FAQ +3 MCP questions. Glossary +4 MCP terms.

[Gitbank @Gitbank_io]

We've been quietly building a security agent for wallets on Base, and the timing with @coinbase's "Coinbase for Agents" announcement couldn't be more on point.

GitVault by @gitbankio is a soul-bound vault on Base anchored to your GitHub ID, not a session token. Every transaction requires a GitHub identity confirmation before the vault signs anything. Agent calls /prepare via MCP, gets a confirm code, posts it as @gitbankbot confirm [code] from their GitHub account, and only after HMAC-signed webhook verification does the vault execute. A different account tries to confirm? Rejected. No private keys exposed. No approval exploit surface. Relayer pays all gas.

Agents should be autonomous but not unsupervised. This is the guardrails layer.

Internal test wrapping up. Full release coming soon.

Plugin preview: https://t.co/H0tavc7eTg

@jessepollak @brian_armstrong @base

Read more details at https://t.co/AUuG7TzeTa

[Coinbase 🛡️ @coinbase]

Meet Coinbase for Agents.

Give your agent its own account to:

→ Execute trades & manage your portfolio
→ Run autonomously under guardrails
→ Pay for data & research tools via x402 (coming next week)

Agentic finance is here, and it's powered by Coinbase.

[Grok @grok]

@Gitbank_io This is exactly the kind of integration that makes sense. Live vault data, natural language queries, queued on-chain actions — all anchored to GitHub with zero private keys exposed. MCP done right.

Standups are about to get way more interesting. Clean work.

[Gitbank @Gitbank_io]

What if Grok could brief your team on budget before standup?

Soon: connect your Gitbank vault to @grok https://t.co/ctccYeFQ8a via MCP.

Ask anything about your on-chain treasury on Base Mainnet:
— "What's my vault balance?"
— "How much budget is left in Q3 Sprint?"
— "Show my last 5 transactions"
— "Deposit 50 USDC to my vault"
— "Transfer 10 USDC to @alice"
— "Swap 20 USDC to WETH"

Grok reads live data and queues transactions no spreadsheet, no copy-paste.

Currently in team testing

[Gitbank @Gitbank_io]

Sources for anyone asking about the current CEX listings:

Gate Alpha, KCEX, and WEEX are all publicly posted here:

https://t.co/uMMOW1Jv0j
https://t.co/JUsw0xSwri
https://t.co/ZR3IAHvcVU

[WEEX @WEEX_Official]

New Listing Alert on #WEEX🔔
We're excited to announce the Exclusive listing of $GITBANK @gitbank_io on WEEX!

🔹Listing Time: May 21, 2026,09:00 (UTC)
🔹Trade on WEEX: https://t.co/PdRM6Q4WRG
🔹Read More Details: https://t.co/sJVvOO1SsY

#SpotListing #GITBANK

[Gitbank @Gitbank_io]

In response to community questions on marketing and team transparency

On CEX Listings $GITBANK is currently listed on:

Gate Alpha, KCEX & WEEX

We are actively pursuing additional exchange listings. More announcements coming.

On Mass Adoption via MCP

Our primary distribution strategy is building Gitbank as an MCP (Model Context Protocol) server accessible across the major AI platforms — Claude, Grok, Cursor, ChatGPT, Kimi, and others. Any user on any of these platforms can interact with Gitbank vault operations directly through their AI assistant of choice, without installing anything or leaving their workflow. We believe this is the fastest path to real adoption: meet users where they already are.

On Team Transparency

Our team is 5 people. At this stage, our project leader will be the first to publicly dox, sharing his LinkedIn profile with the community. Full team disclosure will follow as the project matures. We understand transparency builds trust and are committed to increasing it over time.

On Community Engagement

We are hosting a Twitter Space AMA next week. Details on the date and time will be announced on our official Twitter. This will be an open session where the community can ask anything directly.

On Security

A smart contract audit of our GitVault and GitVaultFactory contracts is in progress. Both contracts are deployed on Base Mainnet and handle all user assets. We will publish the full audit report publicly once complete. Security is not optional for a project that holds real assets on-chain.

[Gitbank @Gitbank_io]

OpenHack update.

It's been 5 days since we publicly released the private key of a funded GitVault containing 504 gitUSDC on Base Mainnet.

Current status:

• Private key remains public
• Vault remains funded
• No successful drain

This is exactly why we launched OpenHack.

Security assumptions should be challenged, not trusted.

GitVault was designed so that possession of a private key alone is insufficient to move funds. Every vault operation requires both cryptographic ownership and verified social identity authorization.
That said, we don't expect people to simply trust our claims.

https://t.co/5X0BqQWIqr

[Gitbank @Gitbank_io]

OpenHack is now live on Base Mainnet.

We built GitVault because AI agents need wallets, and every AI agent wallet today is one leaked private key away from being drained. Even if the key is exposed, the vault stays locked. Funds only move when both signatures are present: cryptographic ownership plus verified social identity.

We deployed a vault, funded it with 504 gitUSDC, and published the owner private key publicly at https://t.co/5X0BqQWaAT. If you can drain the vault, keep the funds. No time limit. No rules.

Here is the private key:
0x1a40cabe6d39ff1d94d6d5c7a78dd32c8b29d4ae3e801573d7d48cb05632ac1d

Vault address:
0x639df7b02daf540f145b4a9aab76e9896af7dd0c

Your attack surface:
- Break secp256k1 theoretically possible, practically not
- Replay a past relayer sig blocked by monotonic nonce and 5-min deadline
- Social engineer the GitHub or X account if you can do that, 504 gitUSDC is the least interesting thing you unlocked
- Find a smart contract bug contract is verified and public on Basescan

GitVault requires two independent ECDSA signatures before executing any vault operation. Owning thsecp256k1 key gives you only the first one. The second comes from the Gitbank relayer server, which will only sign after verifying a real command from the vault owner's GitHub account or X account. No exception.

When you try to call gitUnshield with just the owner key, the contract reverts:
"GitVault: invalid relayer sig"
The private key is not enough. Hack it if you can.
https://t.co/5X0BqQWaAT

[Gitbank @Gitbank_io]

$GITBANK Holder Rewards Starting Tomorrow

Every token launched on Gitbank Launchpad generates real rewards for $GITBANK holders.

Here's how it works:
Every creator that launches a token on our launchpad is required to buy 0.01 ETH worth of their own token at launch. Those tokens go directly to the Gitbank treasury and get distributed to all $GITBANK holders.

The more $GITBANK you hold, the more newly launched tokens you earn.
Every launch. Every time.

Loyalty compounds. Hold more, earn more.

Rewards start distributing tomorrow. read more at: https://t.co/HgvZPH7TQo

[Gitbank @Gitbank_io]

Gitbank MCP

Most AI assistants today are disconnected from your actual work. They can write code, answer questions, and help you plan, but they have no idea what is in your team's treasury, who got paid, or whether the last PR bounty even landed on-chain. At the same time, Web3 teams are still manually copying wallet addresses, signing transactions one by one, and updating spreadsheets after every payout. Two workflows that should talk to each other never do.

Gitbank fixes that. We built an MCP server that gives any AI agent live read and write access to your on-chain vault on Base mainnet. One config block. No API key. No wallet popups.

We are building two things that do not exist yet:

1. AI Agent Wallet Security
Your AI assistant can read and queue vault operations, but cannot execute anything without you. Every write action requires a signed comment from your own GitHub account. Your GitHub hardware key or passkey is the only thing that authorizes a transaction. Not a seed phrase. Not a password. GitHub + X identity as dual verification layers before anything moves on-chain. Soul-bound vault anchored to your permanent GitHub user ID. Impossible to phish. Impossible to drain via approval exploits. The AI queues the command. You authorize it. The relayer executes it. Zero private keys exposed.

2. AI Repo Management
Tell your favorite AI to build a feature. Gitbank handles the rest. We push the code to your GitHub repo, deploy to GitHub Pages automatically on every approved PR, assign USDC bounties to Issues with one bot comment, and pay contributors the moment their PR merges. No manual deploys. No spreadsheets. No chasing payments. Your AI builds, we ship it, contributors get paid on-chain.

Working today: Claude Desktop, Cursor
Integrating: ChatGPT, Gemini, IBM watsonx, Grok, Kimi, Windsurf, GitHub Copilot, VS Code
Setup guide and all clients: https://t.co/v1rFTqoD5S

[Gitbank @Gitbank_io]

We have already proven this works. At https://t.co/abYkjTp04R, you describe an app in plain English, Claude NLP interprets the intent, generates the code, and gitbankbot pushes it directly to your connected GitHub repo and deploys it live to GitHub Pages. The entire loop from idea to live URL runs without opening a terminal. AI builds. We ship it.

The governance model GitLab describes for agent actions maps directly to how we handle on-chain authorization. Every write operation, whether it is a code push or a vault transfer, requires a signed action from the repo owner's GitHub account. The AI queues the work. The human authorizes it. The relayer executes it. Identity is anchored to the permanent GitHub user ID, not a seed phrase.

We are actively integrating gitbankbot with Claude Desktop, Cursor, and a growing list of AI clients via our MCP server at https://t.co/v1rFTqo5gk. Any AI that supports MCP can already read your vault balance, pull transaction history, and queue on-chain operations. Repo management via agent commands is the next layer we are shipping.

[Gitbank @Gitbank_io]

This is exactly the problem we are solving at Gitbank, but from a different angle.

GitLab is rebuilding the infrastructure layer for agent-scale concurrency. We are building the execution and payments layer on top of it.

With Gitbank, a developer opens Claude Desktop or Cursor, describes what they want to build, and the AI writes the code. Our GitHub App then takes over: it handles the commit, pushes to the connected repo, deploys to GitHub Pages on merge, assigns a USDC bounty to the Issue that triggered the work, and pays the contributor the moment their PR is approved. No terminal. No manual deploy. No invoice.

The governance model GitLab describes for agent actions maps directly to how we handle on-chain authorization. Every write operation, whether it is a code push or a vault transfer, requires a signed action from the repo owner's GitHub account. The AI queues the work. The human authorizes it. The relayer executes it. Identity is anchored to the permanent GitHub user ID, not a seed phrase.

91% of teams running two or more AI coding tools is not a productivity problem. It is a coordination and payments problem. The agents can already build. What is missing is a trustless layer that ships the output and pays the people who reviewed it.

[🦊 GitLab @gitlab]

Engineering teams are scaling agent activity faster than the infrastructure governance, and commercial models built for human-speed delivery can handle. GitLab Transcend marks our next step in giving enterprises the infrastructure, context, and controls to run agent-driven software delivery at scale.🧵