Merge e8e08b63f0 into 74050e03c5

litellm/proxy/_types.py

@@ -2585,6 +2585,10 @@ class ConfigGeneralSettings(LiteLLMPydanticObjectBase):
         None,
         description="If True, stores request messages and responses in spend logs. Default is False.",
     )
+    store_responses_in_spend_logs: bool | None = Field(
+        None,
+        description="Controls response content storage in spend logs independently. When unset, follows store_prompts_in_spend_logs for backward compatibility.",
+    )
     disable_auto_add_proxy_admin_to_teams: bool | None = Field(
         None,
         description="By default, the user calling /team/new is automatically added to the new team as a team admin. If True, proxy admins are no longer auto-added; members explicitly listed in members_with_roles are unaffected. Default is False.",

litellm/proxy/proxy_server.py

@@ -6546,6 +6546,21 @@ class ProxyConfig:
                 # For other types, convert to bool
                 general_settings["store_prompts_in_spend_logs"] = bool(value)
 
+        if "store_responses_in_spend_logs" in _general_settings:
+            response_storage_value: Final = (
+                general_settings.get("store_responses_in_spend_logs")
+                if "store_responses_in_spend_logs" in self._yaml_general_settings_keys
+                else _general_settings["store_responses_in_spend_logs"]
+            )
+            if response_storage_value is None:
+                general_settings["store_responses_in_spend_logs"] = None
+            elif isinstance(response_storage_value, bool):
+                general_settings["store_responses_in_spend_logs"] = response_storage_value
+            elif isinstance(response_storage_value, str):
+                general_settings["store_responses_in_spend_logs"] = response_storage_value.lower() == "true"
+            else:
+                general_settings["store_responses_in_spend_logs"] = bool(response_storage_value)
+
         if "disable_auto_add_proxy_admin_to_teams" in _general_settings:
             value = _general_settings["disable_auto_add_proxy_admin_to_teams"]
             if isinstance(value, str):
@@ -16061,6 +16076,7 @@ _GENERAL_SETTINGS_CONFIG_LIST_FIELD_TYPES: Final[Mapping[str, str]] = MappingPro
         "pass_through_endpoints": "PydanticModel",
         "store_model_in_db": "Boolean",
         "store_prompts_in_spend_logs": "Boolean",
+        "store_responses_in_spend_logs": "Boolean",
         "maximum_spend_logs_retention_period": "String",
         "maximum_health_check_retention_period": "String",
         "maximum_spend_logs_cleanup_batch_size": "Integer",

litellm/proxy/spend_tracking/spend_tracking_utils.py

@@ -1169,7 +1169,7 @@ def _get_response_for_spend_logs_payload(
 ) -> str:
     if payload is None:
         return "{}"
-    if _should_store_prompts_and_responses_in_spend_logs():
+    if _should_store_responses_in_spend_logs():
         response_obj: Any = payload.get("response")
         if response_obj is None:
             return "{}"
@@ -1236,6 +1236,18 @@ def _should_store_prompts_and_responses_in_spend_logs() -> bool:
     return get_secret_bool("STORE_PROMPTS_IN_SPEND_LOGS") is True
 
 
+def _should_store_responses_in_spend_logs() -> bool:
+    from litellm.proxy.proxy_server import general_settings
+
+    store_responses_value: Final = general_settings.get("store_responses_in_spend_logs")
+    if store_responses_value is not None:
+        if isinstance(store_responses_value, str):
+            return store_responses_value.lower() == "true"
+        return store_responses_value is True
+
+    return _should_store_prompts_and_responses_in_spend_logs()
+
+
 def _get_status_for_spend_log(
     metadata: dict,
 ) -> Literal["success", "failure"]:

tests/proxy_unit_tests/test_proxy_config_unit_test.py

@@ -353,3 +353,40 @@ class TestYamlStorePromptsDbOverride:
         """_yaml_general_settings_keys should be empty on init."""
         proxy_config = ProxyConfig()
         assert proxy_config._yaml_general_settings_keys == set()
+
+    @pytest.mark.asyncio
+    async def test_response_storage_yaml_value_takes_precedence_over_db(self):
+        proxy_config = self._make_proxy_config_with_yaml_keys({"store_responses_in_spend_logs"})
+        test_general_settings = {"store_responses_in_spend_logs": False}
+
+        with mock.patch("litellm.proxy.proxy_server.general_settings", test_general_settings):
+            await proxy_config._update_general_settings(
+                db_general_settings={"store_responses_in_spend_logs": True},
+            )
+
+        assert test_general_settings["store_responses_in_spend_logs"] is False
+
+    @pytest.mark.asyncio
+    @pytest.mark.parametrize(
+        ("db_value", "expected"),
+        [
+            (None, None),
+            (True, True),
+            ("TRUE", True),
+            (0, False),
+        ],
+    )
+    async def test_response_storage_db_value_is_normalized_when_yaml_omits_key(
+        self,
+        db_value,
+        expected,
+    ):
+        proxy_config = self._make_proxy_config_with_yaml_keys({"master_key"})
+        test_general_settings = {"master_key": "sk-test"}
+
+        with mock.patch("litellm.proxy.proxy_server.general_settings", test_general_settings):
+            await proxy_config._update_general_settings(
+                db_general_settings={"store_responses_in_spend_logs": db_value},
+            )
+
+        assert test_general_settings["store_responses_in_spend_logs"] is expected

tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py

@@ -29,6 +29,7 @@ from litellm.proxy.spend_tracking.spend_tracking_utils import (
     _sanitize_guardrail_information_for_spend_logs,
     _sanitize_request_body_for_spend_logs_payload,
     _should_store_prompts_and_responses_in_spend_logs,
+    _should_store_responses_in_spend_logs,
     get_logging_payload,
     get_spend_logs_id,
 )
@@ -1440,6 +1441,74 @@ def test_should_store_prompts_and_responses_in_spend_logs_case_insensitive_strin
         ), "Expected False (from env var) when key missing, got True"
 
 
+@pytest.mark.parametrize(
+    ("settings", "expected"),
+    [
+        ({"store_prompts_in_spend_logs": True}, True),
+        ({"store_prompts_in_spend_logs": False}, False),
+        (
+            {
+                "store_prompts_in_spend_logs": True,
+                "store_responses_in_spend_logs": False,
+            },
+            False,
+        ),
+        (
+            {
+                "store_prompts_in_spend_logs": False,
+                "store_responses_in_spend_logs": True,
+            },
+            True,
+        ),
+        (
+            {
+                "store_prompts_in_spend_logs": True,
+                "store_responses_in_spend_logs": "FALSE",
+            },
+            False,
+        ),
+    ],
+)
+def test_should_store_responses_in_spend_logs(
+    settings,
+    expected,
+):
+    with patch("litellm.proxy.proxy_server.general_settings", settings):
+        assert _should_store_responses_in_spend_logs() is expected
+
+
+def test_spend_logs_can_store_request_without_response():
+    settings = {
+        "store_prompts_in_spend_logs": True,
+        "store_responses_in_spend_logs": False,
+    }
+    kwargs = {
+        "litellm_params": {
+            "proxy_server_request": {
+                "body": {
+                    "model": "gpt-5-mini",
+                    "messages": [{"role": "user", "content": "Hello!"}],
+                }
+            }
+        }
+    }
+    payload = cast(
+        StandardLoggingPayload,
+        {"response": {"role": "assistant", "content": "Hi there!"}},
+    )
+
+    with patch("litellm.proxy.proxy_server.general_settings", settings):
+        request_result = _get_proxy_server_request_for_spend_logs_payload(
+            metadata={},
+            litellm_params=kwargs["litellm_params"],
+            kwargs=kwargs,
+        )
+        response_result = _get_response_for_spend_logs_payload(payload=payload, kwargs=kwargs)
+
+    assert json.loads(request_result)["messages"] == [{"role": "user", "content": "Hello!"}]
+    assert response_result == "{}"
+
+
 def test_get_spend_logs_metadata_guardrail_info_fallback_from_metadata():
     """
     When standard_logging_payload is None (e.g. guardrail blocks before LLM call),

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -25152,6 +25152,11 @@ export interface components {
              * @description If True, stores request messages and responses in spend logs. Default is False.
              */
             store_prompts_in_spend_logs?: boolean | null;
+            /**
+             * Store Responses In Spend Logs
+             * @description Controls response content storage in spend logs independently. When unset, follows store_prompts_in_spend_logs for backward compatibility.
+             */
+            store_responses_in_spend_logs?: boolean | null;
             /**
              * Supported Db Objects
              * @description Fine-grained control over which object types to load from the database when store_model_in_db is True. Available types: 'models', 'mcp', 'guardrails', 'vector_stores', 'pass_through_endpoints', 'prompts', 'model_cost_map', 'tools', 'config_overrides'. If not set, all objects are loaded (default behavior).