Crate switchyard_llm_client 

HTTP LLM client that speaks Switchyard’s neutral IR directly.

TranslatingLlmClient maps a model name (and the wire format resolved from the request) to a Backend, encodes a switchyard_protocol::Request to that backend’s wire format via switchyard-translation, applies auth and forwards caller headers, makes the HTTP call with a shared [reqwest::Client], and decodes the wire response back to a switchyard_protocol::Response — supporting both buffered and streamed responses.

SystemOneClient serves typed decision requests. Register decision targets with ClientRouter::new_with_decision_clients or ClientRouter::single_with_decision_clients to serve them alongside LLM calls.

run() pairs the client with a libsy algorithm: it drives switchyard_libsy::Algorithm::run_stream, serves routing-time calls, and makes the terminal answer call from the routing outcome when needed. A host that just wants the answer does not have to drive the step stream itself.

Re-exports

pub use backend::Backend;

pub use backend::DEFAULT_MAX_RETRIES;

pub use backend::HttpBackendConfig;

pub use client::AuxiliaryOperation;

pub use client::ModelConfig;

pub use client::TranslatingLlmClient;

pub use error::Result;

pub use raw::RawResponse;

pub use run::ClientRouter;

pub use run::decide;

pub use run::run;

Modules

backend

Per-provider backend configuration: wire format, upstream URL, and auth.

client

TranslatingLlmClient: encode a neutral request, call the configured backend over HTTP, decode the neutral response.

error

Canonical client error re-export and shared context-window-overflow detection.

metrics

Metric labelling inherited from Python

raw

The raw wire result of a call.

run

Drive a libsy algorithm to completion, making the model calls it offloads.

Structs

ModelCallObservation

One completed model call observed while serving an algorithm run.

OpenAiDecisionClient

Serves typed questions through OpenAI’s Decisions API.

SystemOneClient

Serves decision requests through a System One endpoint, such as TypeSafe’s Jev API.

Enums

LlmClientError

Failures a routed LLM client can surface to its caller.

RunObservation

Events emitted inline while crate::run() serves a routing request.

Functions

initialize_metrics

Registers process-wide compatibility gauges with the global meter provider.

Type Aliases

RawEventStream

A stream of wire-format event objects in one format — the unframed body of an SSE response. The serving layer frames each Value (e.g. as an SSE data:/event: block).

RunObserver

Request-scoped callback for algorithm-run observations.