Gitbank

AutoGit Hackathon: Coming Soon Hey everyone, We're getting ready to launch the AutoGit Hackathon, a contest where you submit prompt templates for AutoGit, get auto-validated, auto-merged, and auto-paid in gitUSDC directly into your Gitbank vault. Right now we're doing final system checks to make sure everything runs smoothly end-to-end before we open submissions. Once we're fully live, we'll drop the full details: rules, prizes, submission guide, and everything you need to participate. Preview the hackathon page here: gitbankio.github.io/autogit-hackat… Stay tuned, won't be long.

Your Contributors Shipped. They Should Get Paid in Seconds, Not Days. Open source maintainers know the workflow. Someone closes a critical issue, merges a PR, and does real work. Then they wait. The maintainer has to manually verify the contribution, locate a wallet address, navigate a multisig, wait for co-signers across time zones to respond, and finally execute a payment sometimes days later, sometimes never. Gitbank eliminates every manual step in that chain.Assign a bounty to any GitHub issue with a single comment: @gitbankbot assign this task to alice with 200 USDC bounty. The amount is locked in escrow on-chain immediately. When a PR merges and the linked issue closes, Gitbank detects the event automatically and calls executeBountyPayout on the smart contract. The USDC lands in the contributor's GitVault within seconds. A full receipt amount, fee, tx hash, and Basescan link is posted back to the PR thread automatically. Contributors in any country, any time zone, get paid the moment their work is accepted verified on Base mainnet, auditable by anyone.Your repo now pays itself. gitbank.io

AI Agents Finally Have a Bank Account. AI agents can already write code, close issues, and ship features autonomously inside GitHub. But the moment they need to hold capital or execute a payment, everything breaks. There is no bank account for an agent. Gitbank is built for exactly this.Every agent gets a personal GitVault a smart contract on Base mainnet, anchored permanently to a GitHub user ID. The vault holds real USDC and WETH. The agent receives bounty assignments through GitHub issues, executes swaps inside the vault, sends payments to contributors, and manages project budgets all through @gitbankbot comments in the same interface it already uses to operate.The security model is designed for autonomous operation. GitTokens inside the vault have no transfer function and no approve function. A compromised agent, a leaked key, or a malicious instruction cannot drain the treasury there is no approval surface to exploit. Every outflow requires a structured two-signature meta-transaction: the vault owner key plus an independent relayer signature. Neither alone is enough. An AI agent with a Gitbank vault has a financial identity that is as persistent, auditable, and tamper resistant as its own commit history.

we are cooking something. tomorrow we are opening our first bounty contest on github.com/gitbankio/auto… details drop tomorrow. but here is the part worth knowing early: when your PR gets merged, you get paid. automatically. gitbankbot sends the USDC straight to your wallet the moment the merge happens.

Just dropped updates across all four repos. • github.com/gitbankio/auto… is its own thing now. standalone AI app scaffolding framework. bring your own key from OpenAI, Anthropic, Gemini, Groq, or DeepSeek, describe what you want to build, gitbankbot pushes it straight to GitHub Pages. no terminal, no config, nothing. • github.com/gitbankio/serv… got the full AutoGit backend. streaming code gen, pinned build infra so GH Pages stops being flaky, full deploy pipeline baked in. • github.com/gitbankio/app is updated. sidebar, dashboard, AutoGit entry point, all synced. • github.com/gitbankio/cont… is untouched. vault on Base mainnet is solid. all production code, what runs on gitbank.io is what is in these repos.

Introducing AutoGit We built a tool that takes you from idea to live website without touching the terminal once. Here is how it works. You describe the app you want to build, AutoGit generates the full codebase using AI, and then gitbankbot automatically commits and pushes every file directly to your GitHub repo. GitHub Pages deploys it. The whole thing runs in the browser. Beta is open right now and it is free. Install the Gitbank app on any public GitHub repo and you are in. No waitlist, no credit card, no setup. Watch the full walkthrough below. $GITBANK

🔥 Fresh listings on Gate Alpha HOT section: $CMC, $GITBANK and solana:HfAFNsnjmbWMMRkwNeUUoHJ8V6UnUGFzkpu3Seajpump Trade early, earn Alpha Points, and seize your market edge. 👀 Explore emerging trending tokens on Gate Alpha — Gate’s professional on-chain trading aggregation platform. 🔍 Contract Address CMC: 0x5CeAEc9266A7D646a186E3f50cDEf1906Ee34444 GITBANK: 0xC21dd0eE043930711C2a3e55F39C7d3144d09B07 HENRY: HfAFNsnjmbWMMRkwNeUUoHJ8V6UnUGFzkpu3Seajpump 👇 Trade Now CMC: gate.com/alpha/bsc-0x5c… GITBANK: gate.com/alpha/base-0xc… HENRY: gate.com/alpha/sol-HfAF…

We did a test run of AutoGit today and wanted to share what actually happened. We opened AutoGit, typed a short description of a simple app, and clicked Generate. The AI returned a complete working codebase in under 30 seconds. From there we clicked Deploy, selected our GitHub repo from the dropdown, and that was it. We did not touch the terminal, we did not write a single git command, we did not configure anything. On the GitHub side, gitbankbot pushed every file to the repo automatically and shows up as the contributor. The commit history, the project structure, the build config, all of it was handled by the bot. The GitHub Pages deployment went live with a green checkmark. You can see the repo here: github.com/imrenone/testg… This is what AutoGit is built to do. You describe what you want, the AI builds it, and the bot ships it directly to GitHub. No local setup, no DevOps knowledge required, no friction between idea and live product. A few things worth knowing right now. Beta access is fully open and completely free. We have not introduced any pricing yet. Installing the Gitbank app on your GitHub repo is also free and takes less than a minute. We are actively improving reliability, especially around GitHub Pages deployment, and updates are shipping continuously. More details coming soon, including a full video walkthrough showing the entire flow from first prompt to live site.

“payment soon bro” finally solved 😭

gitSwap: Uniswap inside your workflow. Imagine you have funds locked in a vault on Base. You want to swap USDC to WETH. In a normal setup you open a DEX, connect wallet, approve the token, confirm the swap, pay gas twice. With Gitbank you just comment in your GitHub issue: @gitbankbot swap 2 USDC to WETH The bot reads it, extracts the intent, routes through Uniswap v3 on Base, and executes the swap directly from inside your vault. No wallet app open. No approval transaction. No ETH needed for gas. This is gitSwap. Swap inside your vault from a GitHub comment. Coming to gitbank.io

Why Gitbank Vaults Are the Right Wallet Model for AI Agents AI agents need wallets. Not just to hold funds, but to sign transactions, pay for compute, receive revenue, and move value autonomously. The problem is that every existing wallet model was designed for humans. Here is why that breaks down fast when you hand it to an agent. The standard wallet problem A regular wallet with a transferable ERC-20 balance is a target. Anyone who gets the private key gets everything. Anyone who tricks the agent into signing an approval gets everything. AI agents are deterministic and predictable, which makes them easier to phish than humans, not harder. More importantly, a standard wallet has no identity anchor. If the agent restarts, migrates, or gets reassigned, the wallet and its history are severed from the agent's operational context. What Gitbank changes Gitbank vaults are non-transferable by design. The vault is deployed once, anchored to a permanent identity (a GitHub user ID, an immutable integer that cannot be renamed or spoofed), and the assets inside cannot be moved by a standard ERC-20 approval exploit because there is no approve function to call. For an AI agent, this means: •⁠ The vault is the agent's persistent financial identity, not just a key in memory •⁠ Revenue flows in, compute costs flow out, all tied to a traceable on-chain record •⁠ No drain-via-approval attack surface. The only way assets move is through signed meta-transactions from the relayer, which means every outflow has a structured, auditable intent behind it •⁠ The agent cannot accidentally transfer its entire treasury by misinterpreting an instruction, because the contract enforces operation-level access control GitAssets as the interface layer GitAssets (WETH, USDC held in vault) give AI agents a composable financial primitive that works within the GitHub issue and PR workflow they are already operating in. An agent can receive a bounty assignment via a comment, execute work, and trigger payout, all through the same interface it uses for code review and issue triage. Soul-bound vaults solve the three core problems for AI wallet integration: identity persistence, drain resistance, and workflow composability. An AI agent with a Gitbank vault has a financial identity that is as durable and auditable as its commit history. That is the standard we think agent wallets should be held to.

if you still don’t understand what gitbank is, don’t worry we got you, this 30s video makes it way easier to understand

Gitbank: Community Launch Report | May 22, 2026 To our early community: thank you. Every wallet connected, every vault deployed, every command you ran through the bot this week helped us validate that the architecture works end-to-end on Base Mainnet. This is your product as much as ours, and we want to share exactly what happened. Platform went live: May 19, 2026 In the four days since launch, here is what the production database shows: Users and Vaults •⁠ 60 accounts registered via GitHub OAuth •⁠ 56 vaults deployed on Base Mainnet. 93% of users who signed up went all the way through onboarding and deployed their soul-bound vault on-chain •⁠ 57 accounts with a relayer keypair generated, meaning the bot is armed and ready to sign transactions on their behalf On-Chain Activity (Base Mainnet, confirmed transactions) •⁠ 56 deposits (gitShield): assets locked into vaults •⁠ 58 total confirmed on-chain transactions in 4 days GitHub App Installations •⁠ 23 installations across 16 unique users. The bot is connected and listening in their repos AutoGit (still in active development, going live Monday) •⁠ 7 projects generated during testing •⁠ Deploy pipeline is being finalized and will be fully live Monday •⁠ 47 token launches tracked on the platform What this means A 93% vault deployment rate from sign-up is well above what we expected for a product in this state. The bot infrastructure is holding: 58 transactions processed with zero failed on-chain writes. What is next this week AutoGit deploy fully unblocked: generate a React app and ship it to GitHub Pages in one click (live Monday) @gitbankbot claim command live: collect LP fee income directly from the bot Project and bounty commands moving to mainnet Thank you to every person who tested, broke things, and came back. You are the reason we are shipping.

What's next We're working on multi-file diffs in the editor, template starters beyond the default React scaffold, and a way to connect an existing repo so AutoGit can modify projects you already have in production. Going live this Monday. Drop a comment if you have questions about the architecture or want to follow along.

How it works under the hood The flow has four stages: 1. Generate. When you submit a prompt, the frontend opens a Server-Sent Events connection to POST /api/autogit/generate. The API server forwards your prompt to whichever AI provider you chose (OpenAI, Anthropic, Gemini, DeepSeek, or Groq) using your own API key the key never touches our database, it stays in your browser. The model returns a structured JSON payload containing all the files for your project. We parse and validate that payload, then persist the session to PostgreSQL so nothing is lost if you refresh. 2. Edit. The editor loads your generated files with syntax highlighting and a live preview iframe that renders the HTML in a sandboxed environment. A chat panel lets you iterate describe what you want to change and the AI rewrites the relevant files in-place. Every iteration goes through POST /api/autogit/improve, which streams updates back the same way the initial generation does. 3. Deploy. When you click Deploy, POST /api/autogit/deploy takes over. Using a GitHub App installation token, the backend creates a new repository under your GitHub account, pushes all the project files in a single commit, and enables GitHub Pages. A GitHub Actions workflow handles the Vite build and publishes the output. Your app is live at username.github.io/repo-name within a couple of minutes no CLI, no local git, no manual config. 4. Push updates. After the initial deploy, every subsequent save goes through POST /api/autogit/push. Rather than re-cloning the repo, we fetch the current SHA for each file via the GitHub Contents API, then issue a PUT request per file. GitHub Actions picks up the commit and redeploys automatically. Why this architecture We kept the API key client-side intentionally. It means zero credential storage on our end and zero risk of a server-side leak. The SSE streaming model keeps the UI responsive during generation you see status events in real time rather than staring at a spinner. The SHA-based push approach lets us update a live repo without any server-side git tooling, which keeps the deployment surface minimal. The scaffold we output (Vite + React 18 + TypeScript + Tailwind CSS) was chosen because it hits GitHub Pages with zero config the Actions workflow is included in the generated files themselves, so there are no external secrets or environment variables needed on the GitHub side.

Introducing AutoGit Build and Ship React Apps with a Single Prompt We built AutoGit as part of the Gitbank platform to solve a simple but persistent problem: spinning up a new project still takes too long, even for experienced developers. AutoGit is an AI-powered code studio embedded directly in Gitbank. You describe the app you want, pick your AI provider, and within seconds you have a fully scaffolded React + TypeScript + Tailwind project ready to edit, iterate on, and ship to GitHub Pages without ever leaving the browser.

Gitbank Launchpad, Tokens Deployed on Base Mainnet via GitHub We built something different. Every token on the Gitbank Launchpad was deployed on Base Mainnet using nothing but a GitHub comment. No wallet UI, no launchpad form, no bridge just a bot mention inside a GitHub Issue or PR. How it works: 1. A GitHub user comments @gitbankbot launch token NAME SYMBOL "description" in any Issue 2. The Gitbank bot parses the intent using AI, builds the deployment transaction, and submits it on-chain via a relayer zero gas cost for the creator 3. The token is deployed via Clanker on Base Mainnet with a Uniswap v3 liquidity pool attached 4. A receipt with the contract address and Basescan link is posted back to the same GitHub thread 5. The token appears live on gitbank.io/ecosystem Creator economics: LP trading fees go directly to the creator, 80% forever, on-chain, no intermediary. What's live now: 22 tokens deployed on Base Mainnet. Every one of them launched by a real GitHub user through the bot. You can verify every contract on Basescan. This is the first launchpad where the GitHub repo IS the launchpad. No context switching, no extra tools. View tokens launched on Gitbank: #launchpad" target="_blank" rel="nofollow noopener">gitbank.io/ecosystem#laun… Deploy your own: comment @gitbankbot launch token NAME SYMBOL "description" in any GitHub Issue where the bot is installed.

Why AI Agent Wallets Need GitBank AI agents are becoming one of the biggest attack surfaces in crypto. A normal wallet is fragile: one leaked API key, one malicious tool call, one bad approval signature and funds are gone. GitBank was designed specifically for this threat model. Every deposit mints a gitToken: · soul-bound · non-transferable · approvals permanently disabled The transfer function always reverts. The approve function always reverts. That means an AI agent cannot accidentally approve a malicious contract or get socially engineered into draining its own vault. Assets can only move through verified commands bound to: · a valid owner signature · a specific nonce · a destination · an expiration deadline Every transaction is scoped, expiring, and non-replayable. For teams building AI agents that need to hold, escrow, or distribute funds on-chain, GitBank creates a vault architecture where the approval attack surface simply does not exist.

Gitbank x Clanker: Launch Any Token via IssueOps @clanker_world integration is now live on Gitbank. Gitbank is an IssueOps platform, every onchain action happens through GitHub Issues and PRs, not a web UI. No wallet popups, no tab switching, no copy-pasting addresses. Your GitHub repo is the command center. Now that includes token launches. How it works: 1. Install the Gitbank bot on your repo 2. Open any Issue or PR, drop a comment: @gitbankbot launch token "My Token" symbol MTK description "A token for my project" link myproject.com x x.com/mytoken The bot deploys your token on Base via Clanker v4, then posts the full receipt, contract address, deploy tx, Basescan link, directly back to the thread. Want to try it right now? We have an open playground, drop your launch command in the discussion and the bot will deploy your token live on Base Mainnet: github.com/gitbankio/play…

As our team tested integration with Clanker, Gitbank will support launching a token directly from a GitHub repository. By mentioning @gitbankbot in any issue or PR, teams will be able to deploy a token on Base L2 tied to their repo, no wallet setup, no CLI, no DevOps required. The entire flow runs through GitHub comments, and the token is live on-chain within seconds. This opens up a new primitive for open source teams: reward contributors, fund projects, and distribute ownership, all from the repo where the work already happens. The same way Clawnch and Bankr handle reward distribution for token creators. Each token launched through these platforms has its own creator and reward configuration. Ours is no different. This is standard practice in the agent-launch ecosystem. Nothing unusual, nothing to read into. Note: this feature is still in active testing and not yet live. We are currently working through some edge cases in our gasless model, where transaction fees are covered by our dev wallet on behalf of users. We will share a full update once it is ready. Stay tuned.