Different businesses, the same problems underneath.

Primes run AI work that cannot leave a controlled perimeter. Tier-N suppliers face cybersecurity requirements that exceed what most can build in-house. Citrate handles both, on your hardware, behind your perimeter, with your keys.

Supplier qualification, part provenance, MOQ variance, classification gating, cross-organization transfer with cryptographic provenance, and exportable audit bundles run on the same primitives.

Federated learning lets multiple programs train together without sharing raw data. Every result is checked, not trusted, and leaves an auditable receipt ready for third-party review.

Manufacturers face a hard trade-off: pool data across plants to train better models, or keep process IP behind the wall. Citrate aggregates model updates across plants without exposing raw parameters, and proves the aggregation was done correctly.

Results are deterministic. A model trained in one plant produces the same output in another.

For supply-chain integrity, the provenance primitives that serve defense primes adapt to automotive recalls, semiconductor export control, and FDA-regulated production.

Healthcare AI carries one hard constraint: pool insight across institutions without moving patient data, while producing evidence that satisfies HIPAA, HITRUST, FDA SaMD, and IRB review.

Citrate is built for that shape. Genomics consortia, rare-disease networks, and multi-institution clinical pilots train models locally, prove the aggregation, and keep a complete audit lineage on the record.

Your data stays yours, and stays protected. The network runs the work and leaves the receipt.

Federal Reserve SR 26-2 (April 2026, superseding SR 11-7) raised the bar for AI model risk management. The EU AI Act Annex III raised it again for institutions with European exposure. Both ask the same question: how do you prove your AI is governed when the vendor controls the substrate?

On Citrate, the network is yours, the model weights are yours, and the audit trail is yours. Every privileged action by every AI agent, human-mediated or automated, leaves a record you can tell apart and replay.

Federated fraud detection across an industry consortium produces the same provable provenance, without anyone surrendering their data.

The data-center load problem is real. NERC's Level 3 Alert on hyperscaler-driven voltage events made it explicit. Citrate runs the inverse: sub-MW behind-the-meter compute distributed across facilities, with provable load attestation suitable for regulator filings.

For federated learning across utilities, including grid resilience, demand-response optimization, and equipment failure prediction, the same network that serves defense and pharma serves the grid, with NERC CIP-aligned audit on the record.