2026-08-18
LoopCraft adds Guardrail nodes with PII detection and output validation (2026-08-18)
A new Guardrail node can be attached after any step of a loop to inspect every payload before it moves downstream — PII detection, content filtering, and structural validation — with every check recorded in the trace view.
← Back to NewsStarting 2026-08-18, LoopCraft ships Guardrail nodes for the visual loop builder. Drop the node after any agent or tool step, and every payload that passes through it is inspected before it moves downstream — so a loop can catch sensitive data, policy violations, and malformed output at the step where they happen, not in a log review after the run.
Three checks are built in. PII detection flags email addresses, phone numbers, physical addresses, and national ID patterns before they reach a tool or a final answer. Content filtering screens output against configurable policy categories. Output validation checks structure — required fields, types, and value ranges — so a downstream node never receives a payload it cannot process. The checks are deterministic and run locally, which means a Guardrail node adds no model call to your loop's cost.
When a check fails, you choose the consequence: block the step and fail the run, redact the offending value and continue, or route the iteration down a remediation branch — for example, back to the generator with a note about what was rejected. Every inspection is written to the trace view with the check that fired, the value it saw, and the action it took, so a guardrail decision is never a mystery.
Guardrail nodes pair naturally with the Human Approval node — validate first, then pause for a reviewer — and with the cost budget guard we shipped earlier this month. Guardrail nodes are available on all plans. Try them at https://getloopcraft.com. Feedback: sapsap@qq.com.