Guardrails
Stop dangerous actions before they happen.
Pre-built guardrails stop dangerous actions on every call, including tool poisoning, prompt injection, credential exposure, and rug pulls. Add your own for the protections that matter to your business.
See how it worksHow it works
Built-in & Custom
Start with defaults, add your own
Built-in guardrails are ready when you turn them on, covering risks such as tool poisoning, prompt injection, credential exposure, and rug pulls. Custom rules sit beside them for the protections that matter to your business.
Modes
Block, alert, monitor, or redact
Not every guardrail needs to block on day one. Start in monitor or alert to see what it catches, then switch to block when you're confident. You can also redact sensitive values and let the action go through.
Scope
Set once for the org, tune it per workspace and device
Apply guardrails across the whole org, then adjust them for a workspace or a single device.
Anomalies
Catch what guardrails can't
Guardrails stop known dangerous patterns. Anomalies alert on unusual access, data movement, or tool use that still gets through. They don't block the action.
Learn about AnomaliesUltra product
Ultra puts enforceable boundaries around autonomous systems without slowing down the work they're built to do.
Get started
Secure every action your agents take.
Free to start. Pay per action. No per-seat, no per-host, no SSO tax.