Corpus

Dataset and model provenance on Filecoin.

Corpus is a version-controlled, cryptographically anchored data registry for ML teams. Upload datasets to Filecoin, track lineage across versions, share with collaborators, and anchor model provenance hashes on-chain — so every trained model can be traced back to the exact data it was built on.

Features

• Decentralized storage — datasets uploaded via Synapse to Filecoin, content-addressed by piece CID. Every version gets its own immutable CID.
• On-chain custody — every upload calls recordAndDeduct on StorageTreasury.sol atomically, creating a permanent on-chain record.
• Named datasets + versioning — give datasets a name, upload new versions, list history, set default. All versions live on Filecoin.
• Access control (ACL) — share datasets with other wallet addresses; revoke access at any time.
• Model provenance — register a model run (dataset CID + artifact CID + config hash + code hash). Backend computes keccak256 provenance hash and anchors it on-chain via a self-send transaction.
• Treasury billing — USDFC-denominated storage fees via StorageTreasury.sol on Filecoin Calibration. Monthly debit worker sweeps active datasets.
• API key management — create, rotate, and revoke named API keys per wallet. Each key returns 401 immediately on revoke.
• TypeScript SDK — corpus-sdk on npm wraps all endpoints with typed responses.

Future scope

• Encryption at rest — AES-256-GCM dataset encryption with per-user key management.
• Cross-chain treasury — accept deposits from other EVM chains via bridge, not just Filecoin Calibration.
• Dataset marketplace — list datasets publicly, allow other wallets to purchase access via on-chain payment.
• Provenance verification API — public endpoint to verify any provenance hash against the on-chain anchor without an API key.
• Fine-grained ACL — time-limited access grants, read vs. write permissions, team-level sharing.
• IPFS gateway fallback — serve CIDs via public IPFS gateway when Synapse retrieval is unavailable.
• CLI tool — corpus upload, corpus get, corpus register as first-class terminal commands.

Quick start

cp backend/.env.example backend/.env
# fill in PRIVATE_KEY, SYNAPSE_*, TREASURY_ADDRESS

cd backend && npm install && npm run dev
cd ../frontend && npm install && npm run dev

• Frontend: http://localhost:3000
• Backend API: http://localhost:3001

SDK

npm install corpus-sdk

import Corpus from "corpus-sdk"

const corpus = new Corpus({ apiKey: "your-api-key" })

const { cid } = await corpus.datasets.upload(file, { name: "my-dataset" })
const bytes   = await corpus.datasets.get(cid)

await corpus.models.register({
  datasetCID: cid,
  modelArtifactCID: "baga6ea4...",
  trainingConfigHash: "0x...",
  trainingCodeHash: "0x...",
})

Full SDK docs: npmjs.com/package/corpus-sdk