erc721-ai: on-chain provenance for model weights

If models are assets, they need provenance and transfer rules. erc721-ai explores a token standard and reference contracts for weight bundles tied to training metadata.

erc721-ai
Open-source research artefact · MIT

The question

Fine-tuned models sit at the intersection of IP, data rights, and compute. Blockchains are good at who owns what and what changed when — less good at storing weights on L1. erc721-ai focuses on anchors: commitments, metadata, and composable marketplace interfaces.

What we ship

  • Solidity reference implementation (MIT)
  • Patterns for linking off-chain storage to on-chain ownership
  • Hooks for royalty and licensing logic

Research agenda

We are tracking verifiable training claims, ZK/TEE attestations, and how DEX-style discovery could work for model artefacts.

Docs & code

Full reference documentation lives on docs.kcolbchain.com/erc721-ai. Source and issues: GitHub.