removed model_server from envoy.template and forwarding traffic to bright_staff

arch/envoy.template.yaml

@@ -262,19 +262,16 @@ static_resources:
                       domains:
                         - "*"
                       routes:
-                        {% for internal_cluster in ["arch_fc", "model_server"] %}
                         - match:
                             prefix: "/"
                             headers:
                               - name: "x-arch-upstream"
                                 string_match:
-                                  exact: {{ internal_cluster }}
+                                  exact: bright_staff
                           route:
                             auto_host_rewrite: true
-                            cluster: {{ internal_cluster }}
+                            cluster: bright_staff
                             timeout: 300s
-                        {% endfor %}
-
                         {% for cluster_name, cluster in arch_clusters.items() %}
                         - match:
                             prefix: "/"
@@ -868,24 +865,6 @@ static_resources:
             tls_params:
               tls_minimum_protocol_version: TLSv1_2
               tls_maximum_protocol_version: TLSv1_3
-
-    {% for internal_cluster in ["arch_fc", "model_server"] %}
-    - name: {{ internal_cluster }}
-      connect_timeout: 0.5s
-      type: STRICT_DNS
-      dns_lookup_family: V4_ONLY
-      lb_policy: ROUND_ROBIN
-      load_assignment:
-        cluster_name: {{ internal_cluster }}
-        endpoints:
-          - lb_endpoints:
-              - endpoint:
-                  address:
-                    socket_address:
-                      address: host.docker.internal
-                      port_value: 51000
-                  hostname: {{ internal_cluster }}
-    {% endfor %}
     - name: mistral_7b_instruct
       connect_timeout: 0.5s
       type: STRICT_DNS

crates/common/src/api/open_ai.rs

@@ -4,7 +4,6 @@ use crate::{
 };
 use core::{panic, str};
 use serde::{ser::SerializeMap, Deserialize, Serialize};
-use serde_yaml::Value;
 use std::{
     collections::{HashMap, VecDeque},
     fmt::Display,
@@ -265,7 +264,7 @@ pub struct ToolCall {
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct FunctionCallDetail {
     pub name: String,
-    pub arguments: Option<HashMap<String, Value>>,
+    pub arguments: String,
 }
 
 #[derive(Debug, Deserialize, Serialize)]

crates/common/src/consts.rs

@@ -7,7 +7,7 @@ pub const ARCH_FC_REQUEST_TIMEOUT_MS: u64 = 30000; // 30 seconds
 pub const DEFAULT_TARGET_REQUEST_TIMEOUT_MS: u64 = 30000; // 30 seconds
 pub const API_REQUEST_TIMEOUT_MS: u64 = 30000; // 30 seconds
 pub const MODEL_SERVER_REQUEST_TIMEOUT_MS: u64 = 30000; // 30 seconds
-pub const MODEL_SERVER_NAME: &str = "model_server";
+pub const MODEL_SERVER_NAME: &str = "bright_staff";
 pub const ARCH_ROUTING_HEADER: &str = "x-arch-llm-provider";
 pub const MESSAGES_KEY: &str = "messages";
 pub const ARCH_PROVIDER_HINT_HEADER: &str = "x-arch-llm-provider-hint";

crates/prompt_gateway/src/stream_context.rs

@@ -371,7 +371,26 @@ impl StreamContext {
 
         let tools_call_name = self.tool_calls.as_ref().unwrap()[0].function.name.clone();
         let prompt_target = self.prompt_targets.get(&tools_call_name).unwrap().clone();
-        let tool_params = &self.tool_calls.as_ref().unwrap()[0].function.arguments;
+        let tool_params_str = &self.tool_calls.as_ref().unwrap()[0].function.arguments;
+
+        // Parse arguments JSON string into HashMap
+        // Note: convert from serde_json::Value to serde_yaml::Value for compatibility
+        let tool_params: Option<HashMap<String, serde_yaml::Value>> = match serde_json::from_str::<HashMap<String, serde_json::Value>>(tool_params_str) {
+            Ok(json_params) => {
+                let yaml_params: HashMap<String, serde_yaml::Value> = json_params
+                    .into_iter()
+                    .filter_map(|(k, v)| {
+                        serde_yaml::to_value(&v).ok().map(|yaml_v| (k, yaml_v))
+                    })
+                    .collect();
+                Some(yaml_params)
+            },
+            Err(e) => {
+                warn!("Failed to parse tool call arguments: {}", e);
+                None
+            }
+        };
+
         let endpoint_details = prompt_target.endpoint.as_ref().unwrap();
         let endpoint_path: String = endpoint_details
             .path
@@ -384,7 +403,7 @@ impl StreamContext {
 
         let (path, api_call_body) = match compute_request_path_body(
             &endpoint_path,
-            tool_params,
+            &tool_params,
             &prompt_target_params,
             &http_method,
         ) {
@@ -870,7 +889,7 @@ mod test {
                         id: "1".to_string(),
                         function: common::api::open_ai::FunctionCallDetail {
                             name: "test".to_string(),
-                            arguments: None,
+                            arguments: "{}".to_string(),
                         },
                         tool_type: common::api::open_ai::ToolType::Function,
                     }]),