If the model is just a predictor, where does the intelligence come from?

A model supplies fluency and general capability, available to every competitor on identical terms. What it does not supply is any view about your organization: which records may be retrieved, under whose entitlement, governed by which rules, with what retained afterwards. Those decisions are what make fluent output correct here rather than generically.

Centralpoint holds that layer in your environment — the governed corpus, the skill graph encoding how your organization actually works, the prompts that encode how questions should be handled. The model executes against it and can be changed at runtime across providers without re-indexing or rewriting a rule. The capability is rented; the layer that makes it useful is owned.