Hanami is an AI bouncer for NFT whitelists.

NFT projects have a screening problem. When 4,400 people apply for 200 spots, there is no way to interview them all, and the usual filters do not help. Forms get gamed in an afternoon, captchas get solved by click farms, and follow-checks reward whoever scripts the fastest. The list fills up with bots and flippers.

Hanami replaces the application form with a bouncer: an AI character the project designs and owns as an ERC-7857 iNFT. Each applicant connects a wallet and is interviewed in character for three to six turns. The bouncer asks for the specific, lived-in detail a genuine participant would know, weighs the whole conversation against criteria the project keeps private, and returns an approve or reject.

The screening is verifiable. Every conversation runs inside a Trusted Execution Environment on 0G Compute, which keeps the project's criteria sealed so applicants cannot reverse-engineer them. Each decision is recorded on 0G Chain with a TEE attestation hash that anyone can check on Chainscan. The bouncer's persona and its AI-generated portrait are stored on 0G Storage. When a campaign ends, the project exports a Merkle root of the approved wallets and uses it in a mint contract on any EVM chain.

Hanami is live on 0G mainnet, with two bouncers minted and every contract verified.

Hanami was built from an empty repo during the hackathon. Everything below shipped in that window.

I started by de-risking the parts I wasn't sure 0G could do. Before writing any product code, I confirmed the 0G Storage upload path, the 0G Compute inference and image endpoints, and the one that mattered most: how to pull a TEE attestation hash out of a Compute response and prove it on chain. Locking that down early shaped the rest of the build. Then the contracts. An ERC-7857 iNFT registry for the bouncers and a campaign contract for each whitelist, with ten passing tests, deployed and verified on the Galileo testnet. Next was the core loop. The backend uploads each bouncer's persona and project context to 0G Storage, runs the multi-turn screening conversation through 0G Compute inside a TEE, and commits every decision to chain with its attestation hash. I built the three frontend pages (mint, applicant chat, and admin) and gave the bouncers real personality instead of placeholder copy.

After that came the admin dashboard: live screening counters and a Merkle root export with copy-paste Solidity, so a project can take its approved list to a mint contract on any EVM chain. I minted a second bouncer, Kenji, as a real iNFT, and wrote an adversarial test set for jailbreak and prompt-injection attempts to confirm the bouncer holds character under pressure. The final push moved everything to 0G mainnet and tightened the trust model. Minting is now fully user-signed: the project owner signs all three transactions, so Hanami never custodies the iNFT or the campaign contract. I regenerated the AI portraits to fit the oracle-character look, moved the index database to Turso so it survives host restarts, and deployed the app on Vercel with the backend on Render. At submission, Hanami is live on 0G mainnet with two bouncers minted, every contract verified on Chainscan, and the whole flow working end to end: mint, screen, decide, export.

Hanami is not raising and has taken no outside funding. It was built solo during the hackathon and is fully bootstrapped.