Every agent run touches three places: the tool the work arrives in, the model that reasons about it, and the log that proves what happened. We draw all three before we write a line of code.

01 — Regions first

Data stays in the region you already use. EU customers run on EU infrastructure end-to-end, including the model endpoint and the audit log.

02 — Nothing trains on your data

Model providers are contracted for zero retention. Prompts and outputs are stored only in your own log, for as long as your policy says.

“If we can’t show you the full path on one sheet, the agent doesn’t ship.”

03 — Permissions follow people

Agents inherit the permissions of the systems they read. If a person can’t open a document, neither can the agent answering them.