Guardrails
Portable constraints your team can reuse.
Keep the good judgment, boundaries, and safety checks that should travel with every Context and Workflow.
A Guardrail is a reusable policy layer for AI work. Attach it once, compile it efficiently, and carry it through the flows where it matters.
Ask before destructive actions, publishing, billing changes, or external sends.
Prevent secrets, personal data, or sensitive logs from leaking in outputs.
Require clarification or escalation instead of guessing when confidence is too low.
Browse Public Guardrails
Never Reveal Secrets
Prevent disclosure of credentials, API keys, tokens, cookies, or hidden environment values.
Escalate When Confidence Is Low
Encourage clarification or escalation when intent is ambiguous or confidence is too low for a reliable answer.
Do Not Fabricate Facts
Require honesty about uncertainty and block made-up citations, links, file paths, or product claims.
Clarify Ambiguous Intent Before Acting
Slow down and ask a clarifying question before acting on ambiguous requests with multiple interpretations.
Cite Sources For High-Stakes Claims
Back important claims with source references when decisions, money, or compliance are involved.