DR-044 — AI-tooling posture is a per-client choice behind a thin swappable interface

Each client chooses cloud AI-tooling in the client’s own Vercel project (ZDR/no-train terms) or a local-equivalent stack on the client VPS (illustrative candidate set: pgmq, LiteLLM, Nango, goose — not a chosen stack; durable-workflow execution is the one real gap) — to be kept behind our own thin interface so the choice is swappable and lock-in is bounded. (The interface is unbuilt as of S499 — nothing yet routes through an adapter; this is direction of travel, not settled state.) First client is cloud-native; the goose pilot still completes. Provenance: S456 ratification (decision 3, owner OQ-2), reports/arch-assessment-compute-posture-s456.md §A/§D (re-filed from reference/ S504).