FR
Back to the help centre

AI and automation

Setting AI agent sources and guardrails

Give the AI agent a clear framework for answering, abstaining, and handing off.

Updated October 12, 2026

Separate facts from decisions

A source may confirm a timeline or policy. It should not implicitly authorize a commercial exception or sensitive decision.

Establish priority

Identify approved sources, their owners, and review dates. When sources conflict, define which one prevails or require a human handoff.

Define the limits

  • Missing or conflicting information.
  • Sensitive or exceptional requests.
  • Actions not available in doue.
  • Tone or language that needs team review.

Test abstention

Review cases where the agent must stop as carefully as cases where it may answer. A clear handoff is a successful outcome when it protects the customer and team.

Steps in the application

  1. Open the assistant and check its sources, languages and instructions.
  2. Define handoff cases and check authorized tools.
  3. Add tests for expected answers, insufficient evidence and sensitive requests.
  4. Recheck those tests after changes before expanding assignments.

Visible options depend on role, enabled features and plan. If an option is missing, ask an administrator to check account entitlements.