Definition: A mechanism that limits and controls an Agent’s inputs, outputs, or actions to prevent unsafe or undesirable behavior.

Simply put: A Guardrail tells the Agent: “You can go this far, but no further.”

Examples:

  1. Preventing execution of rm -rf.
  2. Preventing the transmission of confidential information.
  3. Rejecting responses that violate security rules.

AI My-Journey-In-Codeless Agent