Trait ModelClient 

pub trait ModelClient: Send + Sync {
    // Required method
    fn stream<'life0, 'async_trait>(
        &'life0 self,
        input: ModelTurnInput,
    ) -> Pin<Box<dyn Future<Output = Result<BoxStream<'static, Result<ModelChunk, ModelClientError>>, ModelClientError>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;

    // Provided method
    fn next<'life0, 'async_trait>(
        &'life0 self,
        input: ModelTurnInput,
    ) -> Pin<Box<dyn Future<Output = Result<ModelResponse, ModelClientError>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait { ... }
}

Required Methods

fn stream<'life0, 'async_trait>( &'life0 self, input: ModelTurnInput, ) -> Pin<Box<dyn Future<Output = Result<BoxStream<'static, Result<ModelChunk, ModelClientError>>, ModelClientError>> + Send + 'async_trait>>

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

Stream the model’s response as a sequence of ModelChunks. Production ScriptedModelClient / OpenAiCompatibleModelClient implement this; next() is provided as a folding convenience for callers that don’t need token-level events.

Provided Methods

fn next<'life0, 'async_trait>( &'life0 self, input: ModelTurnInput, ) -> Pin<Box<dyn Future<Output = Result<ModelResponse, ModelClientError>> + Send + 'async_trait>>

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

Buffer the stream into a single ModelResponse. Default impl awaits every chunk and then runs collect_model_response. Override only if a provider has a cheaper non-streaming path (none today — even OpenAI’s non-stream call still goes through the same SSE-or-not branch on our side).

Dyn Compatibility

This trait is dyn compatible.

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

Implementors

impl ModelClient for AnthropicModelClient

impl ModelClient for OpenAiCompatibleModelClient

impl ModelClient for ScriptedModelClient