Security and governance
Agents act only within rights, and only after approval.
Governance is not a feature added on top; it is how the platform is built.
Principles
Approvals
Every action with effect — publish, pay, send — goes through a human validation. The request, the approver and the outcome are traced.
Zero-value secrets
Agents use accesses by reference. Credential values never enter a prompt, a transcript or a log.
Rights per agent and per verb
Read, draft, propose, act: each agent is granted verbs on given scopes, and nothing else.
Journal
Who asked, which agent, which access, which action, who approved, what happened: kept and searchable.
Hosted where you need it
By default in the European Union (Ireland), on cloud infrastructure operated by our team. A customer can require another region or jurisdiction, a sovereign cloud, or dedicated hosting for a large group or an institution. The agents, the memory and the rules are the same wherever they run.
No training on your data
Customer data is never used to train a model. Models are called through APIs under terms that exclude training on those exchanges; we keep only what the customer's journal and memory require.
Questions about governance?
We are happy to walk through the approval flow and the journal with your team.