Guardrails

Guardrails in Kompass serve as the safety and validation layer for your AI workflows, ensuring that inputs and outputs adhere to specific standards before proceeding to the next node. They act as "quality gates" that prevent hallucinations, enforce brand guidelines, and block inappropriate content in real-time.

Key Features

• Safety & Compliance: Automatically filter out restricted topics or toxic language to maintain a secure user experience.

• Input/Output Validation: Verify that the data entering or leaving an LLM meets your expected format and quality criteria.

• Custom Rule Sets: Create specific logic to handle "fail" states, such as rerouting a workflow if a guardrail check is triggered.

Integrating with Workflows

While the Guardrail configurations are managed in the Secure section of the platform, they are implemented directly within the canvas as functional nodes.

• Usage in Canvas: To see how to add and configure these within your logic, refer to the Guardrail Component Guide.

-Agent-Level Protection: Guardrails can also be attached directly to Agents to monitor their specific inputs and outputs for brand safety. To learn how to secure your agents, refer to the Agent Configuration Guide.

• Navigation: You can access the global Guardrail management dashboard by navigating to the Guardrails link in the primary sidebar.

Guardrails