Governing open-weight and frontier models together

Why enterprises use both

A self-hosted open-weight model keeps prompts and responses in the enterprise's own environment, and it can be tuned on the enterprise's own data. Frontier models are used where their capability is needed. The choice is often made per task, so one agent estate ends up calling both.

The governance problem

The two kinds of model arrive through different paths, with different keys, endpoints and logs. A policy written against one provider does not cover a self-hosted endpoint, and an agent can fall back to a model nobody has approved.

One policy across both

Governance at the agent level treats every model the same way: only approved models are callable, the same controls apply to each call, and every decision lands in the same record, wherever the model runs.