Skip to content

The Primitive

What zkde.fi Is

zkde.fi is a reference implementation of proof-gated private finance built on top of the Obsqra Labs proving infrastructure. It is not a DeFi app. The DeFi surfaces — Capital OS, Trade Desk, privacy vaults — are demonstrations of the proof system working in production.

The parent infrastructure layer is StarkForge. zkde.fi is one application built on top of it.

The Gap This Closes

On-chain execution is trustless — smart contracts enforce rules deterministically. Off-chain computation is not. When an AI model says "this pool is safe" or "allocate 40% here," there is no cryptographic guarantee that the model ran correctly, that the inputs were real, or that the output wasn't manipulated.

The primitive closes that gap: off-chain computation produces on-chain-verifiable proof before capital moves.

The Architecture: SNARK-in-STARK Dual Lane

zkde.fi uses a dual-lane proving architecture:

  • SNARK lane (Garaga/Groth16): EZKL converts ML model inference into Groth16 proofs. These are verified on-chain via Garaga's BN254 pairing check implemented natively in Cairo. This lane handles zkML verification — proving that a specific model produced a specific output from specific inputs.

  • STARK lane (Stone): The same STARK prover infrastructure that generates validity proofs for Starknet blocks. This lane handles execution integrity — proving that the full computation (policy checks, gate evaluation, settlement routing) executed correctly.

When both lanes are active (FULL_DUAL_PROVER mode), a single action produces two independent cryptographic guarantees: the AI output is correct (SNARK), and the execution that consumed it was faithful (STARK).

flowchart LR
  SIG[Signal / intent] --> EZKL[EZKL inference]
  EZKL --> G16[Groth16 proof]
  G16 --> GAR[Garaga KZG check in Cairo]
  GAR --> STONE[Stone STARK envelope]
  STONE --> GATE[Policy gate]
  GATE --> SETTLE[Settlement + receipt]
  SETTLE --> L3[Madara L3 fact registry]

What The System Can Prove

  1. Input provenance — the data fed to a model came from a verified source (oracle, on-chain state, committed private input)
  2. Model integrity — the ML model that produced a recommendation is the same model whose weights were committed on-chain via KZG
  3. Execution constraints — the transaction that resulted from the recommendation respected slippage bounds, allocation limits, and policy gates
  4. Policy compliance — the user's trust posture (reputation tier, collateral ratio, proof history) met the minimum threshold for the action
  5. Settlement finality — the proof was registered on Madara L3 (or Starknet L2 fallback) and is independently verifiable via block explorer

What's Deployed Right Now

NetworkContractsExamples
Starknet SepoliaZkmlVerifier, ModelBridgeVerifier, GaragaVerifier, ReputationRegistry, VaultController, ObsqraFactRegistry, ReceiptRegistry, DAOConstraintManager, 5 FICO-pack verifiers, and moreZkmlVerifier on Voyager
Ethereum SepoliaHalo2Verifier (EZKL KZG), L1 Bridge SenderHalo2Verifier on Etherscan
Madara L3VerifiedFactRegistryDedicated appchain for proof fact registration
  • 31 Circom circuits compiled to WASM + zkey
  • 2 EZKL ML models (yield forecast, anomaly detector) with Halo2/KZG proofs
  • 136+ on-chain trust receipts
  • 5 verified proof lanes: Groth16 (Garaga), STARK (Stone), Noir HONK, Native KZG, hash-only fallback

Full live verification: zkde.fi/test

Next

Built by Obsqra Labs