diff --git a/pyproject.toml b/pyproject.toml
index 56c365ee..d3c41f32 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "uipath"
-version = "2.1.106"
+version = "2.1.107"
 description = "Python SDK and CLI for UiPath Platform, enabling programmatic interaction with automation services, process management, and deployment tools."
 readme = { file = "README.md", content-type = "text/markdown" }
 requires-python = ">=3.10"
diff --git a/src/uipath/_services/llm_gateway_service.py b/src/uipath/_services/llm_gateway_service.py
index fad16478..c552bf13 100644
--- a/src/uipath/_services/llm_gateway_service.py
+++ b/src/uipath/_services/llm_gateway_service.py
@@ -344,7 +344,7 @@ def __init__(self, config: Config, execution_context: ExecutionContext) -> None:
     @traced(name="llm_chat_completions", run_type="uipath")
     async def chat_completions(
         self,
-        messages: List[Dict[str, str]],
+        messages: Union[List[Dict[str, str]], List[tuple[str, str]]],
         model: str = ChatModels.gpt_4o_mini_2024_07_18,
         max_tokens: int = 4096,
         temperature: float = 0,
@@ -475,13 +475,26 @@ class Country(BaseModel):
             This service uses UiPath's normalized API format which provides consistent
             behavior across different underlying model providers and enhanced enterprise features.
         """
+        converted_messages = []
+
+        for message in messages:
+            if isinstance(message, tuple) and len(message) == 2:
+                role, content = message
+                converted_messages.append({"role": role, "content": content})
+            elif isinstance(message, dict):
+                converted_messages.append(message)
+            else:
+                raise ValueError(
+                    f"Invalid message format: {message}. Expected tuple (role, content) or dict with 'role' and 'content' keys."
+                )
+
         endpoint = EndpointManager.get_normalized_endpoint().format(
             model=model, api_version=api_version
         )
         endpoint = Endpoint("/" + endpoint)
 
         request_body = {
-            "messages": messages,
+            "messages": converted_messages,
             "max_tokens": max_tokens,
             "temperature": temperature,
             "n": n,
diff --git a/src/uipath/tracing/_utils.py b/src/uipath/tracing/_utils.py
index d225d13a..6711c694 100644
--- a/src/uipath/tracing/_utils.py
+++ b/src/uipath/tracing/_utils.py
@@ -13,16 +13,26 @@
 
 from opentelemetry.sdk.trace import ReadableSpan
 from opentelemetry.trace import StatusCode
+from pydantic import BaseModel
 
 logger = logging.getLogger(__name__)
 
 
 def _simple_serialize_defaults(obj):
-    if hasattr(obj, "model_dump"):
+    # Handle Pydantic BaseModel instances
+    if hasattr(obj, "model_dump") and not isinstance(obj, type):
         return obj.model_dump(exclude_none=True, mode="json")
-    if hasattr(obj, "dict"):
+
+    # Handle classes - convert to schema representation
+    if isinstance(obj, type) and issubclass(obj, BaseModel):
+        return {
+            "__class__": obj.__name__,
+            "__module__": obj.__module__,
+            "schema": obj.model_json_schema(),
+        }
+    if hasattr(obj, "dict") and not isinstance(obj, type):
         return obj.dict()
-    if hasattr(obj, "to_dict"):
+    if hasattr(obj, "to_dict") and not isinstance(obj, type):
         return obj.to_dict()
 
     # Handle dataclasses
@@ -31,7 +41,7 @@ def _simple_serialize_defaults(obj):
 
     # Handle enums
     if isinstance(obj, Enum):
-        return obj.value
+        return _simple_serialize_defaults(obj.value)
 
     if isinstance(obj, (set, tuple)):
         if hasattr(obj, "_asdict") and callable(obj._asdict):
@@ -44,6 +54,10 @@ def _simple_serialize_defaults(obj):
     if isinstance(obj, (timezone, ZoneInfo)):
         return obj.tzname(None)
 
+    # Allow JSON-serializable primitives to pass through unchanged
+    if obj is None or isinstance(obj, (bool, int, float, str)):
+        return obj
+
     return str(obj)
 
 
diff --git a/tests/tracing/test_traced.py b/tests/tracing/test_traced.py
index 1237fada..b6b73ac9 100644
--- a/tests/tracing/test_traced.py
+++ b/tests/tracing/test_traced.py
@@ -650,3 +650,81 @@ def test_complex_input(input: CalculatorInput) -> CalculatorOutput:
     assert output["result"] == 54.6  # 10.5 * 5.2 = 54.6
     # Verify the enum is serialized as its value
     assert output["operator"] == "*"
+
+
+@pytest.mark.asyncio
+async def test_traced_with_pydantic_basemodel_class(setup_tracer):
+    """Test that Pydantic BaseModel classes can be serialized in tracing.
+
+    This tests the fix for the issue where passing a Pydantic BaseModel class
+    as a parameter (like response_format=OutputFormat) would cause JSON
+    serialization errors in tracing.
+    """
+    from pydantic import BaseModel
+
+    exporter, provider = setup_tracer
+
+    class OutputFormat(BaseModel):
+        result: str
+        confidence: float = 0.95
+
+    @traced()
+    async def llm_chat_completions(messages: List[Any], response_format=None):
+        """Simulate LLM function with BaseModel class as response_format."""
+        if response_format:
+            mock_content = '{"result": "hi!", "confidence": 0.95}'
+            return {"choices": [{"message": {"content": mock_content}}]}
+        return {"choices": [{"message": {"content": "hi!"}}]}
+
+    # Test with tuple message format and BaseModel class as parameter
+    messages = [("human", "repeat this: hi!")]
+    result = await llm_chat_completions(messages, response_format=OutputFormat)
+
+    assert result is not None
+    assert "choices" in result
+
+    provider.shutdown()  # Ensure spans are flushed
+    spans = exporter.get_exported_spans()
+
+    assert len(spans) == 1
+    span = spans[0]
+    assert span.name == "llm_chat_completions"
+    assert span.attributes["span_type"] == "function_call_async"
+
+    # Verify inputs are properly serialized as JSON, including BaseModel class
+    assert "input.value" in span.attributes
+    inputs_json = span.attributes["input.value"]
+    inputs = json.loads(inputs_json)
+
+    # Check BaseModel class is properly serialized with schema representation
+    assert "response_format" in inputs
+    response_format_data = inputs["response_format"]
+
+    # Verify the BaseModel class is serialized as a schema representation
+    assert "__class__" in response_format_data
+    assert "__module__" in response_format_data
+    assert "schema" in response_format_data
+    assert response_format_data["__class__"] == "OutputFormat"
+
+    # Verify the schema contains expected structure
+    schema = response_format_data["schema"]
+    assert "properties" in schema
+    assert "result" in schema["properties"]
+    assert "confidence" in schema["properties"]
+    assert schema["properties"]["result"]["type"] == "string"
+    assert schema["properties"]["confidence"]["type"] == "number"
+
+    # Verify that tuple messages are also properly serialized
+    assert "messages" in inputs
+    messages_data = inputs["messages"]
+    assert isinstance(messages_data, list)
+    assert len(messages_data) == 1
+    assert messages_data[0] == ["human", "repeat this: hi!"]
+
+    # Verify that outputs are properly serialized as JSON
+    assert "output.value" in span.attributes
+    output_json = span.attributes["output.value"]
+    output = json.loads(output_json)
+
+    assert "choices" in output
+    assert len(output["choices"]) == 1