Trait InlineCompletionBackend 

pub trait InlineCompletionBackend: Send + Sync {
    // Required method
    fn stream(
        &self,
        req: &BackendRequest,
        sink: &mut dyn FnMut(StreamChunk) -> StreamControl,
    ) -> Result<(), BackendError>;

    // Provided method
    fn complete(
        &self,
        req: &BackendRequest,
    ) -> Result<Vec<String>, BackendError> { ... }
}

Trait for AI inline completion backends.

Implementations provide streaming token generation. The default complete() method buffers the stream into a one-shot result, so backends only need to implement stream().

The trait is sync and callback-based to keep this crate dependency-light and runtime-agnostic. Network I/O happens in the provider crate.

Required Methods

fn stream( &self, req: &BackendRequest, sink: &mut dyn FnMut(StreamChunk) -> StreamControl, ) -> Result<(), BackendError>

Stream completion chunks to a callback sink.

Each StreamChunk.text is cumulative — the full candidate so far, not a delta. The sink returns StreamControl::Stop to cancel early.

Provided Methods

fn complete(&self, req: &BackendRequest) -> Result<Vec<String>, BackendError>

One-shot completion: returns the final candidate texts.

Default implementation buffers the stream.

Implementors