fix(litellm): forward ttl field from CachePoint in _format_system_messages

src/strands/models/litellm.py

@@ -220,11 +220,14 @@ def _format_system_messages(
         for block in system_prompt_content or []:
             if "text" in block:
                 system_content.append({"type": "text", "text": block["text"]})
-            elif "cachePoint" in block and block["cachePoint"].get("type") == "default":
+            elif "cachePoint" in block and block["cachePoint"]["type"] == "default":
                 # Apply cache control to the immediately preceding content block
                 # for LiteLLM/Anthropic compatibility
                 if system_content:
-                    system_content[-1]["cache_control"] = {"type": "ephemeral"}
+                    cache_control: dict[str, Any] = {"type": "ephemeral"}
+                    if ttl := block["cachePoint"].get("ttl"):
+                        cache_control["ttl"] = ttl
+                    system_content[-1]["cache_control"] = cache_control
 
         # Create single system message with content array rather than mulitple system messages
         return [{"role": "system", "content": system_content}] if system_content else []

src/strands/types/content.py

@@ -67,9 +67,12 @@ class CachePoint(TypedDict):
 
     Attributes:
         type: The type of cache point, typically "default".
+        ttl: Optional cache TTL duration (e.g. "5m", "1h"). Supported by providers
+            that accept Anthropic-compatible cache_control fields.
     """
 
     type: str
+    ttl: NotRequired[str]
 
 
 class ContentBlock(TypedDict, total=False):

tests/strands/models/test_litellm.py

@@ -955,6 +955,39 @@ def test_format_request_message_tool_call_no_reasoning_signature():
     assert "__thought__" not in result["id"]
 
 
+def test_format_system_messages_preserves_cache_point_ttl():
+    """CachePoint with ttl="1h" should produce cache_control with ttl field."""
+    result = LiteLLMModel._format_system_messages(
+        system_prompt_content=[
+            {"text": "You are a helpful assistant."},
+            {"cachePoint": {"type": "default", "ttl": "1h"}},
+        ]
+    )
+    assert result[0]["content"][0]["cache_control"] == {"type": "ephemeral", "ttl": "1h"}
+
+
+def test_format_system_messages_cache_point_without_ttl():
+    """CachePoint without ttl should produce cache_control with no ttl key (backward compat)."""
+    result = LiteLLMModel._format_system_messages(
+        system_prompt_content=[
+            {"text": "You are a helpful assistant."},
+            {"cachePoint": {"type": "default"}},
+        ]
+    )
+    assert result[0]["content"][0]["cache_control"] == {"type": "ephemeral"}
+    assert "ttl" not in result[0]["content"][0]["cache_control"]
+
+
+def test_format_system_messages_cache_point_with_no_preceding_content():
+    """CachePoint with no preceding text block should be silently ignored."""
+    result = LiteLLMModel._format_system_messages(
+        system_prompt_content=[
+            {"cachePoint": {"type": "default", "ttl": "1h"}},
+        ]
+    )
+    assert result == []
+
+
 def test_thought_signature_round_trip():
     """Test that thought signature is preserved through a full response -> internal -> request cycle."""
     model = LiteLLMModel(model_id="test")