Struct GuardrailsConfig 

pub struct GuardrailsConfig {
    pub injection: Option<InjectionConfig>,
    pub pii: Option<PiiConfig>,
    pub tool_policy: Option<ToolPolicyConfig>,
    pub llm_judge: Option<LlmJudgeConfig>,
    pub secret_scan: Option<SecretScanConfig>,
    pub behavioral: Option<BehavioralConfig>,
    pub action_budget: Option<ActionBudgetConfig>,
}

Top-level guardrails configuration.

Enables declarative guardrail setup via TOML. Each sub-section creates the corresponding guardrail and adds it to the agent’s guardrail chain.

Use GuardrailsConfig::build to convert this config into runtime guardrail instances.

Fields

injection: Option<InjectionConfig>

Prompt injection classifier configuration.

pii: Option<PiiConfig>

PII detection and redaction configuration.

tool_policy: Option<ToolPolicyConfig>

Declarative tool access control rules.

llm_judge: Option<LlmJudgeConfig>

LLM-as-judge safety evaluation.

secret_scan: Option<SecretScanConfig>

Secret scanning configuration.

behavioral: Option<BehavioralConfig>

Behavioral monitoring configuration.

action_budget: Option<ActionBudgetConfig>

Action budget guardrail configuration.

Implementations

impl GuardrailsConfig

pub fn is_empty(&self) -> bool

Returns true if no guardrails are configured.

pub fn build(&self) -> Result<Vec<Arc<dyn Guardrail>>, Error>

Build runtime guardrail instances from this configuration.

Returns a Vec<Arc<dyn Guardrail>> ready to be passed to AgentRunnerBuilder::guardrails() or OrchestratorBuilder::guardrails().

Order: injection → PII → tool policy → LLM judge. Each section that is Some creates the corresponding guardrail instance.

Note: If [guardrails.llm_judge] is configured, you must use build_with_judge instead, passing the judge provider. This method ignores the llm_judge section.

pub fn build_with_judge( &self, judge_provider: Option<Arc<BoxedProvider>>, ) -> Result<Vec<Arc<dyn Guardrail>>, Error>

Build runtime guardrail instances, optionally including the LLM judge.

Pass Some(provider) to enable the LLM-as-judge guardrail when [guardrails.llm_judge] is configured. The provider should be a cheap model (e.g., Haiku, Gemini Flash) separate from the main agent.

Trait Implementations

impl Clone for GuardrailsConfig

fn clone(&self) -> GuardrailsConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GuardrailsConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for GuardrailsConfig

fn default() -> GuardrailsConfig

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for GuardrailsConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for GuardrailsConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.