What if we want no external model at all?

For some organizations the question is not which provider but whether any external inference is permissible. Classified environments, certain health and defence contexts, and jurisdictions with strict residency requirements all rule out sending prompts to a hosted service.

Centralpoint supports embedded local models including Llama, Qwen and ONNX running within the organization's own infrastructure, with on-premises treated as a first-class deployment mode rather than a constrained variant. In that configuration no prompt leaves the environment. The governance layer, retrieval surface and audit artefacts behave identically, so the difference is where inference executes rather than what the platform does.