feat(client): serve System One decision calls

crates/libsy-llm-client/Cargo.toml

@@ -29,6 +29,7 @@ parking_lot.workspace = true
 http.workspace = true
 httpdate.workspace = true
 serde_json.workspace = true
+serde.workspace = true
 tokio.workspace = true
 tracing.workspace = true
 tracing-opentelemetry.workspace = true

crates/libsy-llm-client/src/client.rs

@@ -1,7 +1,7 @@
 // SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 // SPDX-License-Identifier: Apache-2.0
 
-//! [`TranslatingLlmClient`] — the crate's single public entry point: encode a neutral
+//! [`TranslatingLlmClient`]: encode a neutral
 //! request, call the configured backend over HTTP, decode the neutral response.
 
 use std::collections::{BTreeMap, BTreeSet, HashMap};
@@ -896,7 +896,7 @@ fn record_gen_ai_request(url: &str, model: &str, streaming: bool) {
     }
 }
 
-fn convert_reqwest_error(error: reqwest::Error) -> LlmClientError {
+pub(crate) fn convert_reqwest_error(error: reqwest::Error) -> LlmClientError {
     // Reqwest labels truncated or otherwise unreadable response bodies as decode
     // errors, so distinguish them from serde JSON failures at the call site.
     let error = error.without_url();

crates/libsy-llm-client/src/lib.rs

@@ -11,6 +11,9 @@
 //! back to a [`switchyard_protocol::Response`] — supporting both buffered and
 //! streamed responses.
 //!
+//! [`SystemOneClient`] serves typed decision requests. Register decision targets
+//! with [`ClientRouter::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
@@ -24,14 +27,16 @@ mod observability;
 mod observation;
 pub mod raw;
 pub mod run;
+mod system_one;
 
 pub use backend::{Backend, DEFAULT_MAX_RETRIES, HttpBackendConfig};
 pub use client::{AuxiliaryOperation, ModelConfig, TranslatingLlmClient};
 pub use error::{LlmClientError, Result};
-pub use observation::{LlmCallObservation, RunObservation, RunObserver};
+pub use observation::{ModelCallObservation, RunObservation, RunObserver};
 pub use raw::RawResponse;
 pub use run::{ClientRouter, decide, run};
 pub use switchyard_translation::RawEventStream;
+pub use system_one::SystemOneClient;
 
 /// Registers process-wide compatibility gauges with the global meter provider.
 pub fn initialize_metrics() {

crates/libsy-llm-client/src/observation.rs

@@ -11,7 +11,7 @@ use switchyard_protocol::{ModelId, Usage};
 
 /// One completed model call observed while serving an algorithm run.
 #[derive(Clone, Debug)]
-pub struct LlmCallObservation {
+pub struct ModelCallObservation {
     /// Model selected for the completed call.
     pub selected_model: ModelId,
     /// Whether the call completed successfully.
@@ -22,15 +22,17 @@ pub struct LlmCallObservation {
     pub usage: Option<Usage>,
 }
 
-/// Events emitted inline while [`crate::run`] serves a routing request.
+/// Events emitted inline while [`crate::run()`] serves a routing request.
 #[derive(Clone, Debug)]
 pub enum RunObservation {
     /// Metadata attached to the completed routing outcome.
     Outcome(OutcomeMetadata),
     /// A completed model call requested by the algorithm for routing work.
-    LlmCall(LlmCallObservation),
+    LlmCall(ModelCallObservation),
+    /// A completed decision call requested by the algorithm for routing work.
+    DecisionCall(ModelCallObservation),
     /// A completed terminal model call made from the routing outcome.
-    AnswerCall(LlmCallObservation),
+    AnswerCall(ModelCallObservation),
     /// Routing time recorded by the `switchyard.routing_overhead_ms` metric.
     RoutingOverhead(Duration),
 }