Struct OpenAiCompatVisionVerifier 

pub struct OpenAiCompatVisionVerifier { /* private fields */ }

A verifier over the OpenAI-compatible chat-completions wire format — how graphics-focused open models (Qwen-VL, GLM-4V, …) are typically served, both hosted and self-hosted (vLLM). Same discipline as the Anthropic verifier: verify-only, chain tail only, every failure mode folds to unknown with a reason (principles 4/7), and the same look-then-judge answer protocol so verdicts stay comparable across providers.

Implementations

impl OpenAiCompatVisionVerifier

pub fn new( api_key: Option<String>, model: impl Into<String>, base_url: impl Into<String>, ) -> Self

Builds a verifier from explicit configuration.

pub fn from_env() -> Option<Self>

Builds from the environment: POINTLOCK_VISION_BASE_URL and POINTLOCK_VISION_MODEL are both required (None without them — there is no canonical public endpoint or model to default to); POINTLOCK_VISION_API_KEY is optional (self-hosted endpoints commonly need none).

Trait Implementations

impl VisionVerifier for OpenAiCompatVisionVerifier

fn verify<'life0, 'life1, 'async_trait>( &'life0 self, request: VisionRequest<'life1>, ) -> Pin<Box<dyn Future<Output = VisionVerdict> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Answers request.prompt against the screenshot. Infallible by construction: anything that prevents an answer is an unknown verdict with a reason, never an error (principle 4).