Skip to main content

AI & Orin

Guardrails

Guardrails are the rules that keep Orin acting safely. They apply at every autonomy level — the AI can never step outside them.

What guardrails are

Guardrails are limits you place on what Orin is allowed to do. They let you grant autonomy confidently, because even an independent action stays inside boundaries you defined.

Built-in protections

Some guardrails are always on, regardless of settings:

  • Permission limits — Orin can never do something the requesting user couldn't do themselves.
  • Separation of duties — Orin can't both create and approve the same sensitive action. See Separation of Duties.
  • Approval routing — actions that require sign-off still go through approvals.
  • Full audit — every AI action is recorded in the audit trail.

Configurable guardrails

Administrators can tighten Orin's behavior with rules such as:

  • Value thresholds — require approval above a certain amount (e.g. invoices over ₹50,000).
  • Action caps — limit how many actions Orin can take in a single request.
  • Scope limits — restrict which modules or record types Orin can touch.
  • Autonomy per module — set a different autonomy level for each module.

Cost and performance protection

BOS caps the number of actions in a single AI request. If a request would exceed the cap, Orin completes part of it safely and tells you what's left, rather than running unbounded work.

Guardrails grow with autonomy

The higher you set autonomy, the more you should rely on guardrails. Together they let you move fast on routine work while keeping risk firmly contained.