Struct GuardrailEngine 

pub struct GuardrailEngine { /* private fields */ }

The per-generation compiled guardrail engine (DW-082): the compiled rules from the ai.guardrails config block. Built at dataplane refresh; immutable once built. Stored on the dataplane behind an ArcSwap and swapped on reload, so a guardrail change applies to the next request with no restart.

Implementations

impl GuardrailEngine

pub fn compile(cfg: Option<&AiConfig>) -> Self

Compile from the ai: config block’s guardrails section. Absent or empty guardrails yields an empty engine (no rules -> every prompt and response passes through uninspected).

pub fn is_empty(&self) -> bool

Whether the engine carries any rules (cheap dataplane skip — an empty engine allows everything and records nothing).

pub fn check_prompt( &self, consumer_policies: &[String], route_policies: &[String], service_policies: &[String], listener_policies: &[String], global_policies: &[String], chat_req: &ChatRequest, ) -> GuardrailResult

Check the prompt (before the provider call). Runs all prompt-phase rules (phase = Prompt or Both) that apply to the consumer’s policy chain. The first block or redact result wins (declaration order); log actions record but continue.

pub fn check_response( &self, consumer_policies: &[String], route_policies: &[String], service_policies: &[String], listener_policies: &[String], global_policies: &[String], chat_resp: &ChatResponse, ) -> GuardrailResult

Check the response (after the provider call, before returning to the client). Runs all response-phase rules that apply to the consumer’s policy chain. The first block result wins; log actions record but continue. Non-streaming only (streaming responses cannot be schema-validated on partial content).

pub fn check_stream_chunk(&self, chunk_text: &str) -> Option<String>

Check a streaming chunk’s text against banned-content rules (response phase). Returns true if the chunk should be suppressed (a banned pattern matched). Used by AiStreamBody’s poll_frame to cut off banned content mid-stream.

pub fn redact_stream_chunk(&self, chunk_text: &str) -> String

AI-08 (#197): Redact PII from a streaming chunk’s text. Runs all response-phase PII rules’ redactors against the chunk text and returns the redacted string. If no PII rules are configured or no matches are found, the original text is returned unchanged. This enables mid-stream PII scrubbing for SSE responses without buffering the entire response.

pub fn check_text(&self, text: &str) -> GuardrailResult

AI-08 (#197): Check an arbitrary text string (e.g., an MCP tool-call output) against response-phase guardrail rules. Returns Block if a banned pattern matches with a block action, Redact if a PII rule matches with a redact action (carrying the redacted text), or Allow otherwise. Used to inspect MCP tool outputs before they are returned to the client.

pub fn apply_redaction( &self, chat_req: &ChatRequest, redacted_prompt: &str, ) -> ChatRequest

Apply prompt-phase redaction to a ChatRequest’s messages, returning a new ChatRequest with the redacted content. Used when a Redact result is returned: the caller replaces the original chat_req with the redacted one before the provider call.

Trait Implementations

impl Clone for GuardrailEngine

fn clone(&self) -> GuardrailEngine

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GuardrailEngine

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

Formats the value using the given formatter.

impl Default for GuardrailEngine

fn default() -> GuardrailEngine

Returns the “default value” for a type.