diff --git a/src/cache/environment.rs b/src/cache/environment.rs
index a2cbed1..5f025ad 100644
--- a/src/cache/environment.rs
+++ b/src/cache/environment.rs
@@ -14,8 +14,10 @@ pub trait EnvironmentsCache: Send + Sync {
     /// Returns Arc to avoid cloning large JSON on every request
     async fn get_environment(&self, environment_key: &str) -> Option<Arc<Value>>;
 
-    /// Get the pre-computed evaluation context (for flag evaluation)
-    async fn get_context(&self, environment_key: &str) -> Option<EngineEvaluationContext>;
+    /// Get the pre-computed evaluation context (for flag evaluation).
+    /// Returns an Arc so the caller doesn't pay a deep clone per request —
+    /// the context is read-heavy and only replaced on polling-interval refresh.
+    async fn get_context(&self, environment_key: &str) -> Option<Arc<EngineEvaluationContext>>;
 
     /// Store environment document and compute context. Returns true if changed.
     async fn put_environment(&self, environment_key: &str, document: Value) -> bool;
@@ -29,8 +31,10 @@ pub struct LocalMemEnvironmentsCache {
     /// Raw environment documents (for /environment-document endpoint)
     /// Stored as Arc<Value> to avoid cloning large JSON on every request
     environments: Arc<RwLock<HashMap<String, Arc<Value>>>>,
-    /// Pre-computed evaluation contexts (for flag evaluation)
-    contexts: Arc<RwLock<HashMap<String, EngineEvaluationContext>>>,
+    /// Pre-computed evaluation contexts (for flag evaluation).
+    /// Stored as Arc so readers share the same allocation — avoids a deep
+    /// clone of the full context (features + segments) per request.
+    contexts: Arc<RwLock<HashMap<String, Arc<EngineEvaluationContext>>>>,
     /// Identity overrides extracted from environments
     identity_overrides: Arc<RwLock<HashMap<String, HashMap<String, Value>>>>,
 }
@@ -52,9 +56,9 @@ impl EnvironmentsCache for LocalMemEnvironmentsCache {
         environments.get(environment_key).cloned() // Clones Arc (cheap), not Value
     }
 
-    async fn get_context(&self, environment_key: &str) -> Option<EngineEvaluationContext> {
+    async fn get_context(&self, environment_key: &str) -> Option<Arc<EngineEvaluationContext>> {
         let contexts = self.contexts.read().await;
-        contexts.get(environment_key).cloned()
+        contexts.get(environment_key).cloned() // Arc clone, not deep clone
     }
 
     async fn put_environment(&self, environment_key: &str, document: Value) -> bool {
@@ -89,7 +93,7 @@ impl EnvironmentsCache for LocalMemEnvironmentsCache {
             if let Ok(environment) = serde_json::from_value::<Environment>(document.clone()) {
                 let flagsmith_env: FlagsmithEnvironment = environment.to_flagsmith_environment();
                 let context = environment_to_context(flagsmith_env);
-                contexts.insert(environment_key.to_string(), context);
+                contexts.insert(environment_key.to_string(), Arc::new(context));
             }
 
             environments.insert(environment_key.to_string(), Arc::new(document));
diff --git a/src/routes/flags.rs b/src/routes/flags.rs
index 136f868..0cb9681 100644
--- a/src/routes/flags.rs
+++ b/src/routes/flags.rs
@@ -1,10 +1,12 @@
 use crate::error::Result;
+use crate::models::APIFeatureState;
 use crate::routes::extractors::extract_environment_key;
 use crate::state::AppState;
 use axum::{
     Json,
     extract::{Query, State},
     http::HeaderMap,
+    response::IntoResponse,
 };
 use serde::Deserialize;
 
@@ -13,22 +15,34 @@ pub struct FlagsQuery {
     pub feature: Option<String>,
 }
 
+pub enum FlagsResponse {
+    Single(APIFeatureState),
+    Multiple(Vec<APIFeatureState>),
+}
+
+impl IntoResponse for FlagsResponse {
+    fn into_response(self) -> axum::response::Response {
+        match self {
+            FlagsResponse::Single(f) => Json(f).into_response(),
+            FlagsResponse::Multiple(f) => Json(f).into_response(),
+        }
+    }
+}
+
 pub async fn get_flags(
     State(service): State<AppState>,
     headers: HeaderMap,
     Query(query): Query<FlagsQuery>,
-) -> Result<Json<serde_json::Value>> {
+) -> Result<FlagsResponse> {
     let environment_key = extract_environment_key(&headers)?;
 
-    let flags = service
+    let mut flags = service
         .get_flags_response_data(&environment_key, query.feature.as_deref())
         .await?;
 
     if query.feature.is_some() && flags.len() == 1 {
-        // Return single feature
-        Ok(Json(serde_json::to_value(&flags[0])?))
+        Ok(FlagsResponse::Single(flags.swap_remove(0)))
     } else {
-        // Return all features
-        Ok(Json(serde_json::to_value(flags)?))
+        Ok(FlagsResponse::Multiple(flags))
     }
 }