Gamepad

Gamepad is Lovable for onchain mini-games. In the age of vibe coding, it’s time for people to monetize their wildest ideas. On Gamepad, all you do is type an idea - no matter how weird, silly, or ambitious and it spins up into a fully onchain mini-game you can play and share instantly.

Because these games run on crypto rails, creators can finally earn from them in ways that make sense: in-game purchases, ad-free upgrades, pay-to-play, tournaments, or premium modes - all without losing 30% to Apple or Google.

Gamepad lives as a mini app inside the World App, which means creators don’t have to fight for distribution or worry about spammy bot traffic. World App gives us real human players (via World ID) and a massive global audience right out of the box.

It’s as easy as making a TikTok, but instead of videos, you’re spinning up games you actually own, share, and monetize.

Gamepad is built as a Next.js mini app, with MiniKit powering World App and Worldcoin integrations. MiniKit made it seamless to plug into the World ecosystem for identity and distribution.

For AI, we use Together AI as our LLM provider, managing multiple models but primarily Qwen for code generation. This is what lets us turn raw user prompts into working game code.

Every mini-game is tied to an onchain token on Worldchain, using a factory smart contract written in Solidity. This means every game launches with its own token economy.

We render games in Sandpack code sandboxes to isolate them from the rest of the app. If one game crashes it does not affect the main application. One of the hardest problems was that scores live inside the sandbox but our token rewards API runs outside. To solve this we built a message passing architecture that allows the sandbox to talk securely to the app. This way scores can trigger onchain token rewards without breaking isolation.

From an architecture perspective, we also designed a registry of 10–12 core game templates such as tap games, endless runners, and 3D runners. Each template is broken into chunks for physics, collisions, and scoring logic. When a user enters a prompt, we map it to the closest template in the registry and then feed that into Qwen for customization. This hybrid approach combines the stability of pre-built mechanics with the flexibility of AI generation, making game creation both faster and more reliable.