The problem: Defense contractors need air-gapped LLMs. Hospitals want AI but fear HIPAA violations. Banks need audit trails for every AI decision. Current solutions either lock you into cloud providers or leave you on your own for security.
OmnAI provides: - 16 isolated vaults using gVisor sandboxing (zero data leakage between tenants) - Trust-based governance - mandatory human review when confidence drops below 0.90 - Three deployment tiers: SUPERFLY (air-gap/DoD IL6), SOVEREIGN (on-prem/finance), EXCEED (hybrid/R&D) - Compliance-ready architecture (FedRAMP, HIPAA, SOC, ITAR)
Technical stack: vLLM/Triton for inference, offline PKI with AES-GCM-256, per-vault fine-tuning pipelines.
It's in pilot state - I'm looking for feedback on whether this architecture resonates with production AI challenges you're seeing, especially around data sovereignty and regulatory compliance.
https://github.com/TadTanyaTalaTadenTadhgTaya/OmnAI-v3.5
Happy to answer questions about the design decisions, particularly around the trust system and vault isolation model.