How can I incorporate safety checks or thresholds to the pipeline?

Runtime
,
1

I’m building an app for kids, I need to ensure that certain topics/responses have guardrails. How does this work?

2

Hey Danya,

This is done through our Safety node, it analyzes text input for potentially harmful content across multiple topic categories and returns safety assessment results.:

const safetyNode = runtime.nodes.safety({
filters: [‘toxicity’, ‘pii’, ‘prompt_injection’],
threshold: 0.8
});

This checks both user input and model output. Flagged content is blocked before reaching the LLM or user.

For more information you can look into our Safety Checker Node documentation