orchestration integration (#623)

* orchestration integration

* Convert compact json to spaced json

crates/Cargo.lock

@@ -2624,6 +2624,7 @@ version = "1.0.140"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "20068b6e96dc6c9bd23e01df8827e6c7e1f2fddd43c21810382803c136b99373"
 dependencies = [
+ "indexmap 2.9.0",
  "itoa",
  "memchr",
  "ryu",

crates/brightstaff/src/router/mod.rs

@@ -1,3 +1,5 @@
 pub mod llm_router;
+pub mod orchestrator_model;
+pub mod orchestrator_model_v1;
 pub mod router_model;
 pub mod router_model_v1;

crates/brightstaff/src/router/orchestrator_model.rs

@@ -0,0 +1,30 @@
+use common::configuration::AgentUsagePreference;
+use hermesllm::apis::openai::{ChatCompletionsRequest, Message};
+use thiserror::Error;
+
+#[derive(Debug, Error)]
+pub enum OrchestratorModelError {
+    #[error("Failed to parse JSON: {0}")]
+    JsonError(#[from] serde_json::Error),
+}
+
+pub type Result<T> = std::result::Result<T, OrchestratorModelError>;
+
+/// OrchestratorModel trait for handling orchestration requests.
+/// Unlike RouterModel which returns a single route, OrchestratorModel
+/// can return multiple routes as the model output format is:
+/// {"route": ["route_name_1", "route_name_2", ...]}
+pub trait OrchestratorModel: Send + Sync {
+    fn generate_request(
+        &self,
+        messages: &[Message],
+        usage_preferences: &Option<Vec<AgentUsagePreference>>,
+    ) -> ChatCompletionsRequest;
+    /// Returns a vector of (route_name, model_name) tuples for all matched routes.
+    fn parse_response(
+        &self,
+        content: &str,
+        usage_preferences: &Option<Vec<AgentUsagePreference>>,
+    ) -> Result<Option<Vec<(String, String)>>>;
+    fn get_model_name(&self) -> String;
+}

crates/common/Cargo.toml

@@ -14,7 +14,7 @@ derivative = "2.2.0"
 thiserror = "1.0.64"
 tiktoken-rs = "0.5.9"
 rand = "0.8.5"
-serde_json = "1.0"
+serde_json = { version = "1.0", features = ["preserve_order"] }
 hex = "0.4.3"
 urlencoding = "2.1.3"
 url = "2.5.4"

crates/common/src/configuration.rs

@@ -267,6 +267,39 @@ pub struct RoutingPreference {
     pub description: String,
 }
 
+#[derive(Serialize, Deserialize, Debug)]
+pub struct AgentUsagePreference {
+    pub model: String,
+    pub orchestration_preferences: Vec<OrchestrationPreference>,
+}
+
+/// OrchestrationPreference with custom serialization to always include default parameters.
+/// The parameters field is always serialized as:
+/// {"type": "object", "properties": {}, "required": []}
+#[derive(Debug, Clone, Deserialize)]
+pub struct OrchestrationPreference {
+    pub name: String,
+    pub description: String,
+}
+
+impl serde::Serialize for OrchestrationPreference {
+    fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
+    where
+        S: serde::Serializer,
+    {
+        use serde::ser::SerializeStruct;
+        let mut state = serializer.serialize_struct("OrchestrationPreference", 3)?;
+        state.serialize_field("name", &self.name)?;
+        state.serialize_field("description", &self.description)?;
+        state.serialize_field("parameters", &serde_json::json!({
+            "type": "object",
+            "properties": {},
+            "required": []
+        }))?;
+        state.end()
+    }
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize)]
 //TODO: use enum for model, but if there is a new model, we need to update the code
 pub struct LlmProvider {