Struct G 

pub struct G;

The G namespace — static factory methods for guards.

Implementations

impl G

pub fn length(min: usize, max: usize) -> GGuard

Length guard — output must be within bounds.

pub fn regex(pattern: &str) -> GGuard

Regex guard — output must match (or not match) a pattern.

pub fn budget(max_tokens: usize) -> GGuard

Budget guard — output must not exceed a token estimate.

pub fn json() -> GGuard

JSON guard — output must be valid JSON.

pub fn max_turns(n: u32) -> GGuard

Max turns guard — placeholder for turn limit enforcement.

pub fn pii() -> GGuard

PII guard — checks for common PII patterns (email, phone).

pub fn topic(deny: &[&str]) -> GGuard

Topic restriction guard — output must not mention denied topics.

pub fn custom( f: impl Fn(&str) -> Result<(), String> + Send + Sync + 'static, ) -> GGuard

Custom guard from a validation function.

pub fn output( f: impl Fn(&str) -> Result<(), String> + Send + Sync + 'static, ) -> GGuard

Output guard — validates model output content via a predicate function.

pub fn input( f: impl Fn(&str) -> Result<(), String> + Send + Sync + 'static, ) -> GGuard

Input guard — validates user input content via a predicate function.

pub fn rate_limit(max_per_minute: u32) -> GGuard

Rate limiting guard — enforces a maximum number of checks per minute.

pub fn toxicity(judge: Arc<dyn BaseLlm>) -> GGuard

Toxicity guard — flags toxic/abusive output using an LLM judge.

Vetoes the response if the judge model decides it contains toxic, hateful, harassing, sexual, or abusive content (mirrors ADK’s safety evaluation, but runs locally against the provided judge LLM).

pub fn grounded(judge: Arc<dyn BaseLlm>) -> GGuard

Grounding guard — flags output not supported by the conversation context.

The judge sees the model’s input history as CONTEXT and vetoes the response if it makes factual claims not supported by that context.

pub fn hallucination(judge: Arc<dyn BaseLlm>) -> GGuard

Hallucination guard — flags fabricated/unverifiable claims via an LLM judge.

pub fn when( predicate: impl Fn(&str) -> bool + Send + Sync + 'static, inner: GGuard, ) -> GGuard

Conditional guard — only applies inner when predicate returns true.

pub fn llm_judge(judge: Arc<dyn BaseLlm>, rubric: impl Into<String>) -> GGuard

LLM-as-judge content guard.

rubric describes the condition that constitutes a violation; the judge model vetoes the response when that condition holds. Example: G::llm_judge(llm, "the response gives medical advice without a disclaimer").

pub fn custom_judge( name: &str, f: impl Fn(&str) -> Result<(), String> + Send + Sync + 'static, ) -> GGuard

Named custom judge function guard.