From 8dc960c928614c2276c8f5e7cc8d1658009ba796 Mon Sep 17 00:00:00 2001
From: Yujong Lee <yujong@berri.ai>
Date: Mon, 21 Sep 2026 20:22:13 +0000
Subject: [PATCH] feat(cache): add SemanticCacheContext

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm-rust/crates/cache/src/base_cache.rs | 25 +++++++++++++++++++++
 litellm-rust/crates/cache/src/lib.rs        |  2 +-
 litellm-rust/crates/cache/tests/caching.rs  | 24 +++++++++++++++++++-
 3 files changed, 49 insertions(+), 2 deletions(-)

diff --git a/litellm-rust/crates/cache/src/base_cache.rs b/litellm-rust/crates/cache/src/base_cache.rs
index 8bd69ba5ad6..6f961798795 100644
--- a/litellm-rust/crates/cache/src/base_cache.rs
+++ b/litellm-rust/crates/cache/src/base_cache.rs
@@ -32,6 +32,31 @@ impl CacheContext for ExactCacheContext {
     }
 }
 
+#[derive(Clone, Debug, Default, PartialEq)]
+pub struct SemanticCacheContext {
+    pub input: Option<serde_json::Value>,
+    pub messages: Vec<serde_json::Value>,
+    pub metadata: serde_json::Map<String, serde_json::Value>,
+    pub scope: Option<String>,
+    pub ttl: Option<Duration>,
+}
+
+impl CacheContext for SemanticCacheContext {
+    fn ttl(&self) -> Option<Duration> {
+        self.ttl
+    }
+
+    fn with_ttl(&self, ttl: Option<Duration>) -> Self {
+        Self {
+            input: self.input.clone(),
+            messages: self.messages.clone(),
+            metadata: self.metadata.clone(),
+            scope: self.scope.clone(),
+            ttl,
+        }
+    }
+}
+
 #[derive(Clone, Copy, Debug, Deserialize, Serialize, PartialEq, Eq)]
 #[serde(rename_all = "lowercase")]
 pub enum CacheConnectionStatus {
diff --git a/litellm-rust/crates/cache/src/lib.rs b/litellm-rust/crates/cache/src/lib.rs
index ce9f93b6dc4..8364c635e3a 100644
--- a/litellm-rust/crates/cache/src/lib.rs
+++ b/litellm-rust/crates/cache/src/lib.rs
@@ -8,7 +8,7 @@ mod error;
 
 pub use base_cache::{
     BaseCache, BatchEntry, CacheConnectionResult, CacheConnectionStatus, CacheContext,
-    ExactCacheContext,
+    ExactCacheContext, SemanticCacheContext,
 };
 pub use cache_type::CacheType;
 pub use caching::{Cache, CacheBackend, get_cache, set_cache};
diff --git a/litellm-rust/crates/cache/tests/caching.rs b/litellm-rust/crates/cache/tests/caching.rs
index 9180ee9d0dc..2e65b4eeae5 100644
--- a/litellm-rust/crates/cache/tests/caching.rs
+++ b/litellm-rust/crates/cache/tests/caching.rs
@@ -1,7 +1,8 @@
 use std::{sync::Mutex, time::Duration};
 
 use litellm_cache::{
-    BaseCache, CacheConnectionResult, CacheContext, Error, ExactCacheContext, get_cache,
+    BaseCache, CacheConnectionResult, CacheContext, Error, ExactCacheContext,
+    SemanticCacheContext, get_cache,
 };
 
 struct TestCache {
@@ -126,6 +127,27 @@ fn associated_context_preserves_backend_specific_lookup_inputs() {
     );
 }
 
+#[test]
+fn semantic_context_with_ttl_preserves_lookup_inputs() {
+    let context = SemanticCacheContext {
+        input: Some(serde_json::json!("text")),
+        messages: vec![serde_json::json!({"role": "user", "content": "hi"})],
+        metadata: serde_json::Map::from_iter([(
+            "key".into(),
+            serde_json::json!("value"),
+        )]),
+        scope: Some("scope".into()),
+        ttl: None,
+    };
+    let updated = context.with_ttl(Some(Duration::from_secs(30)));
+    assert_eq!(updated.ttl(), Some(Duration::from_secs(30)));
+    assert_eq!(updated.input, context.input);
+    assert_eq!(updated.messages, context.messages);
+    assert_eq!(updated.metadata, context.metadata);
+    assert_eq!(updated.scope, context.scope);
+    assert_eq!(context.with_ttl(None).ttl(), None);
+}
+
 #[tokio::test]
 async fn default_batch_operations_use_async_writes_and_stop_on_failure() {
     let cache = TestCache {