Running AI inside your own cloud, not somebody else's API

Where a model runs is not a footnote

Most AI features get built by calling an API somewhere on the public internet. That is fine for a demo. It is a harder story to tell a compliance officer, because it means your operational data, trip details, driver records, dispatch history, has to leave your boundary to get an answer back.

OneOps took the harder path instead. Every model the platform uses, Claude Code for agent work and Gemini for the rest, runs on Google Cloud's Gemini Enterprise Agent Platform (formerly Vertex AI), inside the same HIPAA-eligible perimeter and the same Business Associate Agreement that covers the rest of your data. There is no separate, unaccounted for model endpoint in the path.

What that buys an operator

Practically, it means the AI answering a dispatcher's question or drafting a report is not a third party. It runs where your data already lives, under the same agreement, and every agent session can carry its own model choice without opening a new vendor relationship to do it.

AI Operations is the runtime that puts real controls around that AI once it is doing routine work: logging every tool call, metering the cost of every session before it shows up on a bill, and giving an operator an emergency stop and an undo for the moment a workflow does not behave the way it should. Binding changes wait for a named human to approve them.

Frontier models, governed like infrastructure

The promise is not just that the AI is capable. It is that running it looks like running infrastructure rather than trusting a black box. Sessions are visible. Costs are visible. Approvals are logged. And the model itself, whichever one an agent happens to use, never leaves the boundary that the rest of your operation already lives inside.

That is the argument for keeping AI inside your own cloud: not a slower feature, but the same feature with a paper trail you can actually show someone.