fixes

Author: cosminacho

packages/uipath_langchain_client/src/uipath_langchain_client/base_client.py

@@ -102,7 +102,7 @@ class UiPathBaseLLMClient(BaseModel, ABC):
     )
 
     default_headers: Mapping[str, str] | None = Field(
-        default={
+        default_factory=lambda: {
             "X-UiPath-LLMGateway-TimeoutSeconds": "295",  # server side timeout, default is 10, maximum is 300
             "X-UiPath-LLMGateway-AllowFull4xxResponse": "true",  # allow full 4xx responses (default is false)
         },
@@ -210,7 +210,7 @@ def uipath_request(
     async def uipath_arequest(
         self,
         method: Literal["POST", "GET"] = "POST",
-        url: str = "/",
+        url: URL | str = "/",
         *,
         request_body: dict[str, Any] | None = None,
         **kwargs: Any,

packages/uipath_langchain_client/src/uipath_langchain_client/clients/normalized/chat_models.py

@@ -140,7 +140,7 @@ def _llm_type(self) -> str:
 
     @property
     def _default_params(self) -> dict[str, Any]:
-        """Get the default parameters for calling OpenAI API."""
+        """Get the default parameters for the normalized API request."""
         exclude_if_none = {
             "frequency_penalty": self.frequency_penalty,
             "presence_penalty": self.presence_penalty,
@@ -175,7 +175,7 @@ def _get_usage_metadata(self, json_data: dict[str, Any]) -> UsageMetadata:
                 cache_creation=json_data.get("cache_creation_input_tokens", 0),
             ),
             output_token_details=OutputTokenDetails(
-                audio=json_data.get("audio_tokens", 0),
+                audio=json_data.get("output_audio_tokens", 0),
                 reasoning=json_data.get("thoughts_tokens", 0),
             ),
         )
@@ -241,7 +241,11 @@ def _preprocess_request(
                         {
                             "id": tool_call["id"],
                             "name": tool_call["function"]["name"],
-                            "arguments": json.loads(tool_call["function"]["arguments"]),
+                            "arguments": (
+                                tool_call["function"]["arguments"]
+                                if isinstance(tool_call["function"]["arguments"], dict)
+                                else (json.loads(tool_call["function"]["arguments"]) if tool_call["function"]["arguments"] else {})
+                            ),
                         }
                         for tool_call in converted_message["tool_calls"]
                     ]

src/uipath/llm_client/__version__.py

@@ -1,3 +1,3 @@
-__titile__ = "UiPath LLM Client"
+__title__ = "UiPath LLM Client"
 __description__ = "A Python client for interacting with UiPath's LLM services."
 __version__ = "1.3.0"

src/uipath/llm_client/settings/llmgateway/auth.py

@@ -30,8 +30,8 @@ def get_llmgw_token_header(
     ) -> str:
         """Retrieve a new access token from the LLM Gateway identity endpoint."""
         url_get_token = f"{self.settings.base_url}/{LLMGatewayEndpoints.IDENTITY_ENDPOINT.value}"
-        assert self.settings.client_id is not None
-        assert self.settings.client_secret is not None
+        if self.settings.client_id is None or self.settings.client_secret is None:
+            raise ValueError("client_id and client_secret are required for S2S authentication")
         token_credentials = dict(
             client_id=self.settings.client_id.get_secret_value(),
             client_secret=self.settings.client_secret.get_secret_value(),
@@ -40,11 +40,15 @@ def get_llmgw_token_header(
         with Client() as http_client:
             response = http_client.post(url_get_token, data=token_credentials)
             if response.is_client_error:
+                try:
+                    body = response.json()
+                except Exception:
+                    body = response.text
                 raise UiPathAuthenticationError(
                     message="Failed to authenticate with LLM Gateway, invalid credentials",
                     request=response.request,
                     response=response,
-                    body=response.json(),
+                    body=body,
                 )
             elif response.is_error:
                 raise UiPathAPIError.from_response(response)

src/uipath/llm_client/settings/llmgateway/settings.py

@@ -75,9 +75,11 @@ def build_base_url(
     ) -> str:
         base_url = f"{self.base_url}/{self.org_id}/{self.tenant_id}"
         if api_config is not None and api_config.client_type == "normalized":
-            url = f"{base_url}/{LLMGatewayEndpoints.NORMALIZED_ENDPOOINT.value.format(api_type='chat/completions' if api_config.api_type == 'completions' else 'embeddings')}"
+            url = f"{base_url}/{LLMGatewayEndpoints.NORMALIZED_ENDPOINT.value.format(api_type='chat/completions' if api_config.api_type == 'completions' else 'embeddings')}"
         else:
-            url = f"{base_url}/{LLMGatewayEndpoints.PASSTHROUGH_ENDPOOINT.value.format(vendor=api_config.vendor_type if api_config is not None else None, model=model_name, api_type=api_config.api_type if api_config is not None else None)}"
+            if api_config is None:
+                raise ValueError("api_config is required for passthrough client_type")
+            url = f"{base_url}/{LLMGatewayEndpoints.PASSTHROUGH_ENDPOINT.value.format(vendor=api_config.vendor_type, model=model_name, api_type=api_config.api_type)}"
         return url
 
     @override
@@ -115,16 +117,16 @@ def get_available_models(self) -> list[dict[str, Any]]:
     @override
     def validate_byo_model(self, model_info: dict[str, Any]) -> None:
         byom_details = model_info.get("byomDetails", {})
-        operaion_codes = byom_details.get("operationCodes", [])
-        if self.operation_code and self.operation_code not in operaion_codes:
+        operation_codes = byom_details.get("operationCodes", [])
+        if self.operation_code and self.operation_code not in operation_codes:
             raise ValueError(
                 f"The operation code {self.operation_code} is not allowed for the model {model_info['modelName']}"
             )
-        if not self.operation_code and len(operaion_codes) > 0:
-            if len(operaion_codes) > 1:
+        if not self.operation_code and len(operation_codes) > 0:
+            if len(operation_codes) > 1:
                 logging.warning(
                     "Multiple operation codes are allowed for the model %s, but no operation code was provided, picking the first one available: %s",
                     model_info["modelName"],
-                    operaion_codes[0],
+                    operation_codes[0],
                 )
-            self.operation_code = operaion_codes[0]
+            self.operation_code = operation_codes[0]

src/uipath/llm_client/settings/llmgateway/utils.py

@@ -10,5 +10,5 @@ class LLMGatewayEndpoints(StrEnum):
 
     IDENTITY_ENDPOINT = "identity_/connect/token"
     DISCOVERY_ENDPOINT = "llmgateway_/api/discovery"
-    NORMALIZED_ENDPOOINT = "llmgateway_/api/{api_type}"
-    PASSTHROUGH_ENDPOOINT = "llmgateway_/api/raw/vendor/{vendor}/model/{model}/{api_type}"
+    NORMALIZED_ENDPOINT = "llmgateway_/api/{api_type}"
+    PASSTHROUGH_ENDPOINT = "llmgateway_/api/raw/vendor/{vendor}/model/{model}/{api_type}"

src/uipath/llm_client/settings/utils.py

@@ -1,13 +1,17 @@
+import threading
 from typing import Any
 
 
 class SingletonMeta(type):
     """Metaclass for creating singleton classes. Used to keep global configs shared between instances."""
 
     _instances: dict[type, Any] = {}
+    _lock: threading.Lock = threading.Lock()
 
     def __call__(cls, *args: Any, **kwargs: Any) -> Any:
         if cls not in SingletonMeta._instances:
-            instance = super().__call__(*args, **kwargs)
-            SingletonMeta._instances[cls] = instance
+            with SingletonMeta._lock:
+                if cls not in SingletonMeta._instances:
+                    instance = super().__call__(*args, **kwargs)
+                    SingletonMeta._instances[cls] = instance
         return SingletonMeta._instances[cls]

src/uipath/llm_client/utils/exceptions.py

@@ -70,6 +70,7 @@ def from_response(cls, response: Response, request: Request | None = None) -> "U
 
         Args:
             response: The httpx Response object.
+            request: The original httpx Request object. Falls back to response.request if None.
 
         Returns:
             A UiPathAPIError instance (or subclass) matching the response status code.

src/uipath/llm_client/utils/retry.py

@@ -47,10 +47,17 @@
 from tenacity.wait import wait_base
 from typing_extensions import TypedDict
 
-from uipath.llm_client.utils.exceptions import UiPathAPIError, UiPathRateLimitError
+from uipath.llm_client.utils.exceptions import (
+    UiPathAPIError,
+    UiPathRateLimitError,
+    UiPathTooManyRequestsError,
+)
 
 # Default retry configuration values
-_DEFAULT_RETRY_ON_EXCEPTIONS: tuple[type[Exception], ...] = (UiPathRateLimitError,)
+_DEFAULT_RETRY_ON_EXCEPTIONS: tuple[type[Exception], ...] = (
+    UiPathRateLimitError,
+    UiPathTooManyRequestsError,
+)
 _DEFAULT_INITIAL_DELAY: float = 2.0
 _DEFAULT_MAX_DELAY: float = 60.0
 _DEFAULT_EXP_BASE: float = 2.0
@@ -157,7 +164,7 @@ def _build_retryer(
     """Build a tenacity retryer from configuration.
 
     Args:
-        max_retries: Maximum number of retry attempts. Returns None if <= 1.
+        max_retries: Maximum number of retry attempts. Returns None if < 1 (i.e., 0 or negative).
         retry_config: Configuration for retry behavior. Uses defaults if not provided.
         logger: Logger for retry attempt warnings.
         async_mode: If True, returns AsyncRetrying; otherwise returns Retrying.
@@ -217,7 +224,7 @@ def __init__(
         """Initialize the retryable transport.
 
         Args:
-            max_retries: Maximum number of retry attempts. Set to 1 to disable retries.
+            max_retries: Maximum number of retry attempts. Set to 0 (default) to disable retries.
             retry_config: Configuration for retry behavior. Uses defaults if not provided.
             logger: Logger for retry attempt warnings.
             *args: Positional arguments passed to HTTPTransport.
@@ -281,7 +288,7 @@ def __init__(
         """Initialize the retryable async transport.
 
         Args:
-            max_retries: Maximum number of retry attempts. Set to 1 to disable retries.
+            max_retries: Maximum number of retry attempts. Set to 0 (default) to disable retries.
             retry_config: Configuration for retry behavior. Uses defaults if not provided.
             logger: Logger for retry attempt warnings.
             *args: Positional arguments passed to AsyncHTTPTransport.