Most LLM guardrails ask: "Is this toxic?" UMEQAM asks: "How much can you trust this answer?" Runtime epistemic risk engine for AI in medicine, law, finance, and mental health. Per-response, per-domain verdict: PASS / REVIEW / FAIL. 87.5% accuracy on standard benchmark. 100% on mental health. GPT-4o + DeepSeek ensemble. EU AI Act aligned. Built-in PII redaction. Live API. No setup. $200-500/mo pilot.