A persistent epistemic substrate for AI. Nous treats language models as larynx, not mind, and benchmarks epistemic structure instead of output fluency.