I. The forgetting
The dominant pattern in AI is forgetting. A foundation model is trained, published, then forked a thousand times. Each fork becomes a product. Each product earns. None of the earnings flow back. The lineage is hidden in config.json at best, lost at worst.
We treat models like software, but they are more like soil. They are cultivated, not engineered. They carry the residue of every dataset and every contributor. To pretend a fine-tune is a fresh creation is to lie about where intelligence comes from.
II. The proposal
Mekar makes lineage structural. Every agent is an INFT (ERC-7857) whose mint requires declaring its parents. The protocol refuses orphans. The lineage is public, immutable, and queryable.
When an agent serves an inference, the payment is split — recursively — up the family tree. Your fork's fork's fork still pays you. Forever. Without permission. Without trust.
III. Why botany
Trees, not graphs. Blooms, not nodes. We chose botanical metaphors because they're honest about what's happening: a model is a bloom whose roots run deep, whose petals are the surface you touch, and whose seeds become the next generation. Cut a bloom and the lineage dies with it. Plant it in the right soil — Mekar's lineage registry — and it pollinates forever.
IV. Built on 0G
Mekar lives on the 0G modular stack. 0G Storage pins the model weights and training cards. 0G Compute serves inferences with verifiable execution inside TEE. 0G Chain settles royalty splits atomically. Alignment Nodes audit the lineage for drift. We didn't build a chain — we cultivated on someone else's soil. That's the point.
V. Royalty as oxygen
The royalty cascade isn't a tip jar. It's oxygen. It changes who can afford to publish a model. A grad student who fine-tunes a base model and gets it adopted earns from every descendant, not just their own queries. An open-source data curator who licenses their corpus into the protocol earns from every model trained on it, in perpetuity.
We believe this fixes the perverse incentive at the heart of AI: the race to obscure your sources, because attribution costs money. On Mekar, attribution is money.
VI. What we will not do
- We will not gate-keep. Anyone can plant a bloom.
- We will not custody weights. Mekar holds hashes, not models.
- We will not invent a separate token. Royalties pay in $0G.
- We will not police output. The protocol is a registry, not a regulator.
VII. The garden grows
This is v0.4, deployed on 0G Aristotle mainnet for the APAC hackathon. The garden grows as the seeds we've planted prove their roots.
— Mekar Labs · Bandung & Singapore · 2026