iNFTs, registry, splitter.
ERC-7857 with a lineage layer on top. Sub-second finality. The graph of who-trained-on-what lives on-chain, immutable and append-only.
Lineage is the provenance & royalty protocol for AI agents on 0G. Every dataset, model, and skill is an iNFT. Every inference produces a signed attribution receipt. Royalties stream — automatically — to every contributor in the lineage.
scanning registry…
scanning registry…
connect to mint
settled to a contributor, verifiable on explorer
AI in 2026 sits on top of a stack with no provenance. Contributors don't get paid. Models are trained on uncompensated data, fine-tuned with uncredited labels, and executed inside agents that compose tools and memories from many authors. Every dollar of agent revenue is built on uncredited work.
Agent memory is under attack: $45M+ was lost in 2026 to memory-poisoning and tool-injection. iNFTs trade as opaque blobs — the moment one is used, attribution evaporates. These are three faces of the same gap. AI has no audit trail.
Every inference produces a signed attribution receipt — a JSON object listing every iNFT touched and its weight, signed by the TEE that ran the compute and the agent host that called it. Receipts are written to 0G Storage. A periodic settler reads them, computes the weighted payout, and posts a Merkle root on-chain.
Contributors withdraw with a Merkle proof. Pulls, not pushes. The receipt is the record.
One inference uses Skill #37 (a web-search skill). It composes Model #36 (a news summariser), which was trained on Alice's DataINFT #33 (70%) and Bob's DataINFT #34 (30%). Revenue: 0.001 OG. Royalties cascade upstream:
| Contributor | Calculation | Payout |
|---|---|---|
| Skill owner | 50% × 3% × 0.001 | 0.000015 OG |
| Model owner | (50%↑) × 60% × 5% × 0.001 | 0.000009 OG |
| Alice (data #33) | (40%↑) × 70% × 2% × 0.001 | 0.0000024 OG |
| Bob (data #34) | (40%↑) × 30% × 2% × 0.001 | 0.0000010 OG |
ERC-7857 with a lineage layer on top. Sub-second finality. The graph of who-trained-on-what lives on-chain, immutable and append-only.
Datasets, model weights, tool source, memory checkpoints. Content-addressed roots; the iNFT carries the storage root, not the bytes.
Per-inference receipts persisted to 0G DA. Cheap, infinitely scalable, verifiable. The settler reads from here to compute payouts.
The TEE is the trust root for 'this inference really used these iNFTs.' Every receipt is dual-signed: TEE + host operator.
Lineage adds the graph layer — what each iNFT was built from and how royalties cascade. The standard for agentic provenance.
Merkle root posted on-chain. Contributors withdraw with a proof. Gas-efficient, audit-friendly, and recoverable across batches.
—
—
—
—
—
—
Upload an artifact — dataset, model weights, a skill manifest. Declare upstream parents. Set your royalty %. Your iNFT is now an on-chain contributor.
Open the mintPick a model, type a prompt. Your wallet sends 0.001 OG as the inference fee. The agent host returns a TEE-signed receipt and a Merkle batch posts on-chain.
Run the demoIf your iNFT was in the lineage, a proof is waiting under your address. Claim it. OG lands in your wallet.
View earnings