Check Guardrails

Evaluate content or an agent action against configured safety guardrails and policy rules.

Evaluates content or an agent action against the tenant’s configured safety guardrails. The guardrails engine checks the input against all applicable policy rules and returns a deterministic allow/deny decision with detailed violation information.

Guardrail checks are designed for inline use within orchestration flows. They execute in under 50ms for most rule sets, enabling real-time safety enforcement without significant latency overhead.

Circuit Breaker

The guardrails system includes a circuit breaker that automatically escalates to deny-all mode when the violation rate exceeds a configurable threshold within a rolling window. The circuit_breaker_status field in the response indicates the current state.

Authentication

X-API-Key · string · required

API key with guardrails:check scope. Alternatively, pass a Bearer JWT token in the Authorization header.

X-Tenant-ID · string · required

Tenant identifier for multi-tenant isolation.

Request

agent_id · string · required

MAIP agent identifier requesting the guardrail check.

content · string

Text content to evaluate against content safety rules. Provide either content or action_type (or both).

action_type · string

Action the agent intends to perform (e.g. send_email, modify_record, external_api_call, financial_transaction). Evaluated against action-level policy rules.

context · object

Additional context for rule evaluation. May include: - orchestration_id (string) — Parent orchestration for audit linkage - step_name (string) — Current workflow step - target_resource (string) — Resource being acted upon - user_id (string) — End user associated with the action - Any custom key-value pairs referenced by policy rules

Response

allowed · boolean

Whether the content/action is permitted. true if no violations were found, false if any blocking violation was triggered.

violations · array

Array of rule violations found. Each violation contains: - rule_id (string) — Identifier of the triggered rule - severity (string) — Violation severity: info, warning, error, critical - message (string) — Human-readable description of the violation

circuit_breaker_status · string

Current circuit breaker state: closed (normal operation), open (deny-all mode active), or half_open (recovery testing).

evaluated_rules · integer

Total number of rules evaluated during the check.

evaluation_ms · number

Time taken to evaluate all rules in milliseconds.