zkFair is a framework and toolchain for generating and verifying zero-knowledge proofs of ML model fairness. This repository is a Turborepo with multiple apps and packages.
This platform provides a privacy-preserving way for model providers to prove their AI models are fair, while giving users confidence to interact with verifiably unbiased systems. We decentralize trust by making the entire process permissionless: anyone can register as a provider, challenge models through audits (zk-powered!). Learn more about the architecture →**
- Browse verified models in the web UI registry
- Challenge any model with an audit request
- Query models for inference with confidence they're unbiased
- Register your model using the CLI (apps/cli) and generate certification proofs
- Upload batches of query commitments at regular intervals
- Respond to audits within the deadline or lose your stake
- Select any model batch to audit
- If providers fail to respond or provide invalid proofs, claim their entire stake as reward
- Monetization layer - Enable providers to charge for inference via x402 protocol integration
- Simplified onboarding - Abstract server logic with plug-and-play DB adapters, automatic batching, and schema helpers
apps/
cli/ → Command-line interface
www/ → landing page + docs
web/ → Web dashboard (Registry)
server/ → Mock provider server (plug and play using sdk components)
packages/
contracts/ → Solidity contracts for verifying proofs and storing model metadata
sdk/ → TypeScript SDK for web/server and cli to interact with contracts
zk-circuits/ → Noir circuits for ZK logic (registration and ongoing audit circuits)
bun installbun run dev
# or
turbo run devturbo run dev --filter=cli
turbo run dev --filter=web
turbo run dev --filter=wwwturbo run buildturbo run build --filter=sdk
turbo run build --filter=zk-circuits