n8n Guardrails

Native n8n nodes (v1.119+) that enforce rules on text flowing into or out of AI agents. Two operations: ‘check text for violations’ (AI-powered, uses a model via OpenRouter) and ‘sanitize text’ (non-AI, deterministic cleanup). Supports eight guardrail types – keywords, jailbreak detection, NSFW, PII/personal data, secret keys, topical alignment, URLs, and custom regex/prompt – stackable in a single node with pass/fail branching that controls downstream flow.

Key points

• Check-text-for-violations uses AI (slower, richer); sanitize-text is deterministic (faster, safer for pre-LLM cleanup)
• PII guardrail detects credit cards, SSNs, emails, IPs, locations, dates – either block all or select specific entities
• Jailbreak guardrail returns a 0-1 confidence score – tune the threshold per use case
• Topical alignment locks conversations to a business scope (example: ‘n8n workflow automation’)
• URL guardrail can allow only specific domains, restrict schemas (HTTPS-only), and block user info or subdomains
• Custom guardrails accept either a natural-language prompt or a regex pattern
• Stackable – one node can enforce keywords + jailbreak + PII simultaneously

Related entities

• n8n

Related concepts

• Agent Budget Controls
• Agentic AI

Source references

• ^[src-005] Nate Herk cluster — Nate Herk — n8n cluster (18 videos)

– Videos referenced: oWdJMJp2HgM