Skip to main content

InferenceEngine

roder_api::inference

Trait InferenceEngine 

Source 
pub trait InferenceEngine:
    Send
    + Sync
    + 'static {
    // Required methods
    fn id(&self) -> InferenceEngineId;
    fn capabilities(&self) -> InferenceCapabilities;
    fn list_models<'life0, 'life1, 'async_trait>(
        &'life0 self,
        ctx: InferenceProviderContext<'life1>,
    ) -> Pin<Box<dyn Future<Output = Result<Vec<ModelDescriptor>>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait;
    fn stream_turn<'life0, 'life1, 'async_trait>(
        &'life0 self,
        ctx: InferenceTurnContext<'life1>,
        request: AgentInferenceRequest,
    ) -> Pin<Box<dyn Future<Output = Result<InferenceEventStream>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait;

    // Provided methods
    fn metadata(&self) -> InferenceProviderMetadata { ... }
    fn tool_result_image_input(&self, _model: &str) -> bool { ... }
    fn requires_native_compaction(&self) -> bool { ... }
    fn compact_turn<'life0, 'life1, 'async_trait>(
        &'life0 self,
        _ctx: InferenceTurnContext<'life1>,
        _request: AgentInferenceRequest,
    ) -> Pin<Box<dyn Future<Output = Result<Option<InferenceEventStream>>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait { ... }
}

Required Methods§

Source

fn id(&self) -> InferenceEngineId

Source

fn capabilities(&self) -> InferenceCapabilities

Source

fn list_models<'life0, 'life1, 'async_trait>( &'life0 self, ctx: InferenceProviderContext<'life1>, ) -> Pin<Box<dyn Future<Output = Result<Vec<ModelDescriptor>>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Source

fn stream_turn<'life0, 'life1, 'async_trait>( &'life0 self, ctx: InferenceTurnContext<'life1>, request: AgentInferenceRequest, ) -> Pin<Box<dyn Future<Output = Result<InferenceEventStream>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Provided Methods§

Source

fn metadata(&self) -> InferenceProviderMetadata

Source

fn tool_result_image_input(&self, _model: &str) -> bool

Whether this engine puts the image a tool result carries under crate::transcript::VIEW_IMAGE_DISPLAY_KEY in front of model.

This is separate from InferenceCapabilities::image_input, which is about images the user attaches. An engine returns true only when its request mapping forwards tool-result images. Everywhere else the runtime sends the request without the image, appends a one-line notice to that tool result, hides screenshot-only tools and does not charge image tokens for it. The stored transcript is never changed, so switching to an engine that returns true shows it the same images.

Source

fn requires_native_compaction(&self) -> bool

Whether compaction must remain provider-owned, without local pruning or text-summary fallback. Native failures propagate to the caller.

Source

fn compact_turn<'life0, 'life1, 'async_trait>( &'life0 self, _ctx: InferenceTurnContext<'life1>, _request: AgentInferenceRequest, ) -> Pin<Box<dyn Future<Output = Result<Option<InferenceEventStream>>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Native, opaque provider compaction. Unsupported engines return None. The stream must publish a completed boundary and one terminal completion.

Dyn Compatibility§

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors§