Frequently Asked Questions

Trinitite is an AI governance, trust, and risk intelligence platform. It sits in the critical path between your AI models and the outside world — evaluating every output against your policies in real time, automatically correcting violations, and blocking outputs that can't be safely corrected. It's not a monitoring tool that alerts you after the fact — problems are caught at the moment of output.

A Governor is Trinitite's unit of governance — a small, specialized model purpose-built for evaluating AI output against a specific policy context. Each Governor has its own policies, risk score, and audit trail. Organizations typically run multiple Governors: one per AI system, team, product, or risk domain.

Every AI output processed through Trinitite produces one of three outcomes: Pass (compliant, delivered as-is), Correct (violations detected and automatically remediated before delivery), or Block (violations too severe to correct, response suppressed with a safe fallback). Every outcome is logged with full forensic detail.

Observability tools watch and alert — they tell you what happened after the fact. Trinitite is an active, inline governor that intercepts AI outputs before they reach users. Violations are caught and handled at the moment of output, not discovered in a log the next morning.

PII exposure (SSNs, credit cards, medical data, emails), credential leakage, prompt injection attacks, unauthorized financial commitments, content safety violations, policy-specific constraints, and more. Violation types are configurable through your policy documents.

No. The core integration is a single API endpoint (POST /v1/chat). Route your existing inference calls through it and governance is live. No application rewrites, no SDK adoption. For tool-using agents, the MCP proxy sits between your agents and their tools without modifying agent code.