Skip to main content

Guardrails

Guardrails are validators that check the safety, quality, and compliance of every input or output passing through your applications. They are the platform's safety layer — used to filter, flag, or fix content based on rules you configure.

This section covers:

  • Guardrail types — every validator available in SimplAI: PII detection, Competitor Check, Toxic Language, Gibberish, Bias Check, Prompt Injection, Logic Check, Reading Level, Reading Time.
  • Project-level policies — define named Guardrails configurations at the project level and reuse them across agents.
  • Guardrails as Tool steps — drop any guardrail into a Tool's step graph as a regular node.

What a guardrail does​

Every guardrail follows the same pattern:

  1. The validator is configured to apply to a field — either the Input or the Output of a step or agent.
  2. The validator runs against that field at execution time.
  3. If the validator triggers (e.g., it detects PII, toxic language, or a competitor mention), a corrective action is applied. Standard corrective actions:
  • FIX — modify the content (e.g., redact PII, replace competitor name with a placeholder).
  • NOOP — flag the content but don't modify it. The downstream step still receives the original.
  • Exception — throw an exception to halt processing and alert the caller.

Where guardrails attach​

Guardrails attach at three different places:

  • On an Agent — Guardrails section of the agent configuration. Applied to the agent's input messages or final responses.
  • On a Tool step — as a per-node setting in the Tool builder, applied to that node's input or output. See Tool steps — Advanced.
  • As a Tool step — using each validator as a standalone step in the Tool graph. See Guardrails as Tool steps.

Voice Agents and guardrails​

Guardrails are not yet configurable on Voice Agents. The voice agent configuration screen intentionally omits the Guardrails section while voice-specific guardrail behavior is being worked through.

Where to start​

  • New to guardrails? Read Guardrail types for what each validator does.
  • Operating multiple agents in a project? Set up Project-level policies so you don't reconfigure the same guardrails on every agent.
  • Building a custom Tool? Guardrails as Tool steps lets you put validation exactly where it belongs in your step graph.