AI privacy and control
Scoped workflow agents, approved processing and customer-controlled budgets.
Updated
AI runs inside an authorised workflow. The agent receives only permitted context for that run, with versioned knowledge, permissions, provider policy and usage limits.
Data flows
A local provider can keep processing inside your environment, subject to its configuration. An external provider receives the allowed context under its contract and the deployment's approved policy. Self-hosting does not prove that a third-party cloud retains no data or performs no training. Review endpoint, retention, training, cache and processing-location terms.
Decisions and evidence
A model response is not business authority. Required human decisions, deterministic verification and transactional business checks remain in force. Missing or stale required knowledge prevents the affected run or requires human handling.
Usage, cancellation and support
Budgets bound new runs and external usage. Cancellation is a durable process command; closing a chat does not itself settle in-flight work. Support uses sanitised material or explicitly authorised scoped access, not installation-wide visibility into private prompts and records.
FAQ
Is cloud AI required?
No. Ordinary deterministic workflows operate without AI. Local or cloud AI needs a compatible, approved Product and provider configuration.
Does connecting a provider activate every agent?
No. Product release, evaluated model and knowledge evidence, company grants and budgets must also permit the run.
Does a recommendation change a business record?
No. The declared action must pass permissions, policy, verification and the workflow's commit checks.