AXIOMOS is a governed AI runtime designed to place deterministic policy, structural analysis and audit controls around model execution. I designed and developed the architecture around governance-before-generation, fail-closed decision making, isolated case tracking and cryptographically verifiable evidence. The technical-preview release includes an installable Python runtime, API, automated regression and self-tests, concurrency testing, audit records and Ed25519 evidence signing. A 500-request packaged benchmark completed successfully with unique case and transaction IDs and verified evidence for every tested request.