Navigation

Mobile Docs

AI Guardrails

Apply safety guardrails to AI inputs and outputs

Validate and filter AI model inputs and outputs with configurable guardrails. Supports content filtering, PII protection, topic restrictions, output format enforcement, token limits, and custom regex rules.

Core Blocksai_guardrails

Usage

01Add the block to your workflow and connect it to the upstream step.
02Configure any required credentials or tokens in the inputs.
03Fill in required inputs and optional parameters for the run.
04Run a test execution, inspect outputs, and iterate before deploying.
05Deploy the ai_guardrails block with monitoring enabled in production.

Inputs (UI)

Text to Validate

long-input

Placeholder: Enter text or reference a variable to validate against guardrails...

Layout: full

Direction

dropdown

Layout: half

Options: Input (before AI), Output (after AI)

Default Severity

dropdown

Layout: half

Options: Block, Warn, Log Only

Guardrail Types

checkbox-list

Layout: full

Options: Content Filter, PII Protection, Topic Restriction, Output Format

Max Tokens (optional)

short-input

Placeholder: e.g., 4096

Layout: half

Blocked Topics (comma-separated)

short-input

Placeholder: e.g., politics, religion

Layout: half

Custom Regex Patterns (one per line)

long-input

Placeholder: e.g., \bforbidden-word\b

Layout: full

Inputs (API)

input

string

Required

direction

string

Optional

severity

string

Optional

guardrailTypes

json

Optional

maxTokens

string

Optional

blockedTopics

string

Optional

customPatterns

string

Optional

Outputs

Primary response type:

{
  "passed": "boolean",
  "filteredText": "string",
  "violations": "json",
  "warnings": "json",
  "metadata": "json"
}

Tool Access

ai_guardrails_validate