Loading
Citrate is an open marketplace for verifiable compute. Independent machines run your AI work, every result is checked rather than trusted, and each settled job leaves a receipt you can export and audit.
Three sides meet in one place: machines for hire, models with provenance built in, and data you can train on without sharing it. Every side settles against the same receipts.
Independent hardware runs inference and training and earns from real demand, not token emissions. You pay only for what you use. Idle machines become a line of revenue. 773K transactions per second, independently benchmarked on parallel execution.
Every model carries where it came from and what it was trained on. Q16.16 deterministic ML produces bit-identical results on any hardware, so a result computed on one machine verifies on another. Drop-in compatible with the OpenAI and Anthropic APIs, so existing tools work unchanged.
Federated learning is live at federated.citrate.ai. Participants train locally and contribute updates that aggregate without exposing raw data, with cryptographic dispute paths for any suspected-bad contribution. Your data stays yours, and stays protected.
Every job leaves evidence. You decide how strong that evidence has to be. Three tiers, priced by the cost of proving the work, from a simple commitment to secure hardware.
Whichever tier you pick, every settled job produces an exportable receipt. Dispute a result and the protocol slashes the party at fault.
The worker commits to the result and stakes against it. Fastest and cheapest, with a receipt and the same dispute-and-slash settlement as every other tier.
The work carries a cryptographic proof of correct execution that anyone can verify independently, without re-running it. Pick this when a receipt alone is not enough.
Execution runs inside a secure hardware enclave with attestation, so the result is proven correct and the data stays sealed even from the operator. The strongest level of proof.
The substrate inherits its primitives from RFCs and NIST guidance. No invented primitives, no homegrown curves, no single-vendor dependency. Core protocol logic is formally verified.
Point your existing clients at Citrate and they keep working. The API surface matches the OpenAI and Anthropic APIs, so adopting verifiable compute does not mean rewriting your stack.
Anything Solidity-compatible runs unchanged on chain 40204, extended with deterministic precompiles for inference. 80+ smart contracts are live across compute, models, data, and governance.
First-class SDKs for TypeScript and Python wrap settlement, verification, and federated training, so building on verifiable compute feels like building on any modern API.
The verification model is backed by engineering discipline most networks never attempt. The core protocol logic is formally verified, and every claim here is checked rather than asserted.
Thirty minutes is enough for a technical walk-through. We answer the questions that matter and tell you plainly what we have not yet built. Status and date, never a bare certified.