seal
sealing model checkpoint
seal
0 sealedseal a model ↗
decentralized provenance registry

every model,
sealed on-chain.

Seal fingerprints every AI model and dataset, then records its lineage on-chain — so anyone can audit provenance before they deploy.

0+
models sealed
0
datasets indexed
0
validators
registry · live● mainnet
dataset · common_voice model · whisper-base license · mit fine-tune · v2 deployed · prod audit · passed
sha256: 9f3acb17…b3d81
sealed 7c41f0…b2e8 · llama3-8b-instructsealed e9a2c4…7d1f · whisper-seal-v2sealed a3b9d1…c0e4 · sd-xlaudit passed 0 findings · fineweb-edu-v3sealed 5d8e2f…a1b7 · gpt-neox-20bsealed 7c41f0…b2e8 · llama3-8b-instructsealed e9a2c4…7d1f · whisper-seal-v2sealed a3b9d1…c0e4 · sd-xlaudit passed 0 findings · fineweb-edu-v3sealed 5d8e2f…a1b7 · gpt-neox-20b
the registry

a public ledger
for model fingerprints.

every entry is hashed, timestamped, and written to the chain. no more trusting a model card — verify it.

model / datasetsealfingerprintlicensestatussealed at
whisper-seal-v2speech-to-texte9a2c4…7d1fmitverified2 min ago
llama3-8b-instructchat7c41f0…b2e8llama2auditing9 min ago
stable-diffusion-xlimagea3b9d1…c0e4openrailverified17 min ago
gpt-neox-20btext5d8e2f…a1b7apache-2.0verified31 min ago
common-crawl-cleandatasetb7f3a9…0c2dcc-bynew1 hr ago
fineweb-edu-v3dataset2c8a7e…f4b6odc-byverified3 hr ago
provenance

trace it back
to the first byte.

one lineage, end to end.

every checkpoint links back to the data it was trained on — and forward to where it is deployed.

datasetscommon-voice · fineweb base modelwhisper-base · llama3 fine-tuneinstruction · rlhf checkpointhash 7c41f0…b2e8 auditpassed · 0 findings deployprod · endpoint sealedon-chain ✓
how it works

sealing takes
four steps.

i

fingerprint

your model or dataset is hashed with sha256 — a fixed fingerprint that changes if anything changes.

ii

lineage

we link the fingerprint to its parents: training data, base model, license, contributor.

iii

attest

independent validators check the fingerprint and sign it — cryptographically.

iv

seal

the attestation is written on-chain. it is now public, permanent, and auditable.

inside seal

everything that
keeps a model honest.

verification badge

one seal, three states.

verified (audited + signed) · auditing (in review) · new (awaiting first attestation). deploy teams can filter by state.

● verified● auditing● new
license matrix

know what you can build on.

mit, apache-2.0, llama2, openrail, cc-by, odc-by — every seal carries its license, machine-readable.

mit ✓ · apache-2.0 ✓ · openrail ✓ · cc-by ✓
contributor leaderboard

top sealers this week.

verified datasets and clean fingerprints earn $SEAL.

01 · novabench · 1,204 $SEAL
02 · datadiver · 982 $SEAL
03 · modelminer · 811 $SEAL
audit reports

findings, on the record.

every audit is published with its signature — the full history, not a summary.

0
open findings this week
staking

put skin in the game.

validators stake $SEAL to attest fingerprints — slashed if they sign a bad one.

12.4%
current apy
network

distributed, not datacenter.

9,412 validators across 74 regions sign every seal. no single party controls provenance.

uptime 99.99% · finality 2.1s
roadmap

where the
ledger is heading.

phase 01 · now

the registry ships.

  • fingerprint + seal any model or dataset
  • public registry with license matrix
  • waitlist open, first 10k badges
phase 02 · next

token + staking.

  • $SEAL launch on robinhood chain
  • validator staking + slashing
  • contributor rewards go live
phase 03 · later

attestation everywhere.

  • inference-time attestation (zk)
  • ci/cd plugin: seal on every release
  • cross-chain provenance queries