security & guardrails

Detect and block AI threats

Every message that enters or leaves your agents passes through a shield: it detects prompt injection, jailbreak attempts, sensitive-data leaks and malicious content — before they reach the model or your systems.

  • Prompt injection

    Hidden instructions that try to derail the agent are detected and blocked automatically before the model sees them.

  • PII & secrets

    Personal data, credentials and keys are masked or blocked according to your organization's policy.

  • Per-member audit

    Every input and output is recorded per organization member: who asked what, what was blocked and why.

how it's configured

Detectors with policy-driven actions

Each detector is configured independently: block, transform or report only. Your team decides how strict the shield is per channel, per agent or per topic — without writing code.

  • Detectors enabled per channel and per agent.
  • Gradual actions: block, transform or report.
  • Banned-topic lists managed by your team.