Runtime protection for AI agents
Guardian sits between your users and your model, inspects each message against your rules, and blocks, redacts, alerts on or logs what breaks them.
In this release
You can use this in your workspace today.
- GuardianBYOK gateway inspection for traffic explicitly routed through Guardian, with log-only and pre-delivery block modes.Live
- Agent sessionsPersisted gateway session timelines with redacted content and recorded detection flags.Live
- Runtime logsPersisted gateway decisions, enforcement state and reason codes from the last seven days.Live
Injection checks
Inspect incoming messages for injection patterns such as instruction overrides, role-play framing, prompt extraction and encoding tricks.
Personal data detection
Find personal data such as email addresses, phone numbers and payment card or IBAN numbers in what the agent says, and redact it before it leaves.
Your own rules
Add blocked topics, custom pattern rules and a per-agent request rate limit, each with the action it takes.
Sessions and logs
Follow each runtime conversation Guardian inspected and every event it raised, with the reason for each.
Block or redact
Stop a message that breaks a rule, or remove the matching personal data and let the rest through.
Alert
Let the message through and raise an event you can route to your alert channels.
Log
Let the message through and record the match, to tune a rule before you enforce it.