Feat: cache discovery results and extract shared get_model_info

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to `uipath_llm_client` (core package) will be documented in this file.
 
+## [1.8.4] - 2026-04-16
+
+### Added
+- `lru_cache` on `get_available_models()` — discovery endpoint results are cached per settings instance, avoiding redundant network calls when creating multiple models in a session
+- `get_model_info()` shared utility for looking up a model by name from the discovery endpoint results, with optional vendor and BYOM connection ID filters
+
 ## [1.8.3] - 2026-04-16
 
 ### Added

packages/uipath_langchain_client/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to `uipath_langchain_client` will be documented in this file.
 
+## [1.8.4] - 2026-04-16
+
+### Changed
+- Factory functions (`get_chat_model`, `get_embedding_model`) now use the shared `get_model_info()` utility from the core package instead of an inline implementation
+
 ## [1.8.3] - 2026-04-16
 
 ### Added

packages/uipath_langchain_client/pyproject.toml

@@ -6,7 +6,7 @@ readme = "README.md"
 requires-python = ">=3.11"
 dependencies = [
     "langchain>=1.2.15",
-    "uipath-llm-client>=1.8.3",
+    "uipath-llm-client>=1.8.4",
 ]
 
 [project.optional-dependencies]

packages/uipath_langchain_client/src/uipath_langchain_client/__version__.py

@@ -1,3 +1,3 @@
 __title__ = "UiPath LangChain Client"
 __description__ = "A Python client for interacting with UiPath's LLM services via LangChain."
-__version__ = "1.8.3"
+__version__ = "1.8.4"

packages/uipath_langchain_client/src/uipath_langchain_client/factory.py

@@ -22,6 +22,7 @@
 
 from typing import Any
 
+from uipath.llm_client.utils.discovery import get_model_info
 from uipath_langchain_client.base_client import (
     UiPathBaseChatModel,
     UiPathBaseEmbeddings,
@@ -37,49 +38,6 @@
 )
 
 
-def _get_model_info(
-    model_name: str,
-    *,
-    client_settings: UiPathBaseSettings,
-    byo_connection_id: str | None = None,
-    vendor_type: VendorType | str | None = None,
-) -> dict[str, Any]:
-    available_models = client_settings.get_available_models()
-
-    matching_models = [m for m in available_models if m["modelName"].lower() == model_name.lower()]
-
-    if vendor_type is not None:
-        matching_models = [
-            m for m in matching_models if m.get("vendor", "").lower() == str(vendor_type).lower()
-        ]
-
-    if byo_connection_id:
-        matching_models = [
-            m
-            for m in matching_models
-            if (byom_details := m.get("byomDetails"))
-            and byom_details.get("integrationServiceConnectionId", "").lower()
-            == byo_connection_id.lower()
-        ]
-
-    if not byo_connection_id and len(matching_models) > 1:
-        matching_models = [
-            m
-            for m in matching_models
-            if (
-                (m.get("modelSubscriptionType", "") == "UiPathOwned")
-                or (m.get("byomDetails") is None)
-            )
-        ]
-
-    if not matching_models:
-        raise ValueError(
-            f"Model {model_name} not found. Available models are: {[m['modelName'] for m in available_models]}"
-        )
-
-    return matching_models[0]
-
-
 def get_chat_model(
     model_name: str,
     *,
@@ -120,11 +78,11 @@ def get_chat_model(
         ValueError: If the model is not found in available models or vendor is not supported.
     """
     client_settings = client_settings or get_default_client_settings()
-    model_info = _get_model_info(
+    model_info = get_model_info(
+        client_settings.get_available_models(),
         model_name,
-        client_settings=client_settings,
+        vendor_type=str(vendor_type) if vendor_type is not None else None,
         byo_connection_id=byo_connection_id,
-        vendor_type=vendor_type,
     )
     model_family = model_info.get("modelFamily", None)
     if model_family is not None:
@@ -300,11 +258,11 @@ def get_embedding_model(
         >>> vectors = embeddings.embed_documents(["Hello world"])
     """
     client_settings = client_settings or get_default_client_settings()
-    model_info = _get_model_info(
+    model_info = get_model_info(
+        client_settings.get_available_models(),
         model_name,
-        client_settings=client_settings,
+        vendor_type=str(vendor_type) if vendor_type is not None else None,
         byo_connection_id=byo_connection_id,
-        vendor_type=vendor_type,
     )
     is_uipath_owned = model_info.get("modelSubscriptionType") == "UiPathOwned"
     if not is_uipath_owned:

src/uipath/llm_client/__init__.py

@@ -36,6 +36,7 @@
     PlatformSettings,
     get_default_client_settings,
 )
+from uipath.llm_client.utils.discovery import get_model_info
 from uipath.llm_client.utils.exceptions import (
     UiPathAPIError,
     UiPathAuthenticationError,
@@ -64,6 +65,8 @@
     # HTTPX clients
     "UiPathHttpxClient",
     "UiPathHttpxAsyncClient",
+    # Discovery
+    "get_model_info",
     # Retry
     "RetryConfig",
     # Exceptions

src/uipath/llm_client/__version__.py

@@ -1,3 +1,3 @@
 __title__ = "UiPath LLM Client"
 __description__ = "A Python client for interacting with UiPath's LLM services."
-__version__ = "1.8.3"
+__version__ = "1.8.4"

src/uipath/llm_client/clients/litellm/client.py

@@ -34,6 +34,7 @@
     RoutingMode,
     VendorType,
 )
+from uipath.llm_client.utils.discovery import get_model_info
 from uipath.llm_client.utils.retry import RetryConfig
 
 # Route OpenAI chat completions through base_llm_http_handler (accepts HTTPHandler)
@@ -189,22 +190,11 @@ def _discover_and_build_api_config(
         User-supplied ``vendor_type`` filters models during discovery.
         User-supplied ``api_flavor`` overrides the discovered value.
         """
-        available_models = self._client_settings.get_available_models()
-        matching = [
-            m for m in available_models if m["modelName"].lower() == self._model_name.lower()
-        ]
-
-        if vendor_type is not None:
-            matching = [
-                m for m in matching if m.get("vendor", "").lower() == str(vendor_type).lower()
-            ]
-
-        if not matching:
-            raise ValueError(
-                f"Model '{self._model_name}' not found. "
-                f"Available: {[m['modelName'] for m in available_models]}"
-            )
-        model_info = matching[0]
+        model_info = get_model_info(
+            self._client_settings.get_available_models(),
+            self._model_name,
+            vendor_type=str(vendor_type) if vendor_type is not None else None,
+        )
 
         model_family: str | None = None
         raw_family = model_info.get("modelFamily", None)

src/uipath/llm_client/settings/base.py

@@ -80,6 +80,10 @@ class UiPathBaseSettings(BaseSettings, ABC):
         extra="allow",
     )
 
+    # Pydantic models are not hashable by default; restore object identity
+    # hashing so that @lru_cache can be used on instance methods.
+    __hash__ = object.__hash__  # type: ignore[assignment]
+
     @abstractmethod
     def build_base_url(
         self,
@@ -132,13 +136,12 @@ def build_auth_pipeline(
         ...
 
     @abstractmethod
-    def get_available_models(
-        self,
-    ) -> list[dict[str, Any]]:
+    def get_available_models(self) -> list[dict[str, Any]]:
         """Get the list of available models from the backend.
 
         Subclasses must implement this method to query the backend's
-        model discovery endpoint.
+        model discovery endpoint. Implementations should use
+        ``@lru_cache`` to avoid redundant network calls.
 
         Returns:
             A list of dictionaries containing model information.

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

@@ -1,5 +1,6 @@
 import logging
 from collections.abc import Mapping
+from functools import lru_cache
 from typing import Any, Self
 
 from httpx import Client
@@ -114,7 +115,8 @@ def build_auth_headers(
         return headers
 
     @override
-    def get_available_models(self) -> list[dict[str, Any]]:
+    @lru_cache  # noqa: B019
+    def get_available_models(self) -> list[dict[str, Any]]:  # type: ignore[override]
         discovery_url = f"{self.base_url}/{self.org_id}/{self.tenant_id}/{LLMGatewayEndpoints.DISCOVERY_ENDPOINT.value}"
         with Client(
             auth=self.build_auth_pipeline(),

src/uipath/llm_client/settings/platform/settings.py

@@ -1,6 +1,7 @@
 """Base settings for UiPath Platform (AgentHub/Orchestrator) client."""
 
 from collections.abc import Mapping
+from functools import lru_cache
 from typing import Any, Self
 
 from pydantic import Field, SecretStr, model_validator
@@ -163,8 +164,8 @@ def build_auth_headers(
         return headers
 
     @override
-    def get_available_models(self) -> list[dict[str, Any]]:
-
+    @lru_cache  # noqa: B019
+    def get_available_models(self) -> list[dict[str, Any]]:  # type: ignore[override]
         models = UiPath().agenthub.get_available_llm_models(
             headers=dict(self.build_auth_headers()),
         )

src/uipath/llm_client/utils/discovery.py

@@ -0,0 +1,67 @@
+"""Shared model discovery helpers."""
+
+from typing import Any
+
+
+def get_model_info(
+    available_models: list[dict[str, Any]],
+    model_name: str,
+    *,
+    vendor_type: str | None = None,
+    byo_connection_id: str | None = None,
+) -> dict[str, Any]:
+    """Find and return a single model entry from the discovery endpoint results.
+
+    Applies the following filters in order:
+
+    1. Match by ``modelName`` (case-insensitive).
+    2. If ``vendor_type`` is given, keep only models whose ``vendor`` matches.
+    3. If ``byo_connection_id`` is given, keep only models whose
+       ``byomDetails.integrationServiceConnectionId`` matches.
+    4. When no ``byo_connection_id`` is provided and multiple candidates remain,
+       prefer UiPath-owned (non-BYOM) models.
+
+    Args:
+        available_models: Full list of model dictionaries from the discovery
+            endpoint (as returned by :meth:`UiPathBaseSettings.get_available_models`).
+        model_name: Name of the model to look up.
+        vendor_type: Optional vendor filter (e.g. ``"openai"``).
+        byo_connection_id: Optional BYOM connection ID filter.
+
+    Returns:
+        The first matching model dictionary.
+
+    Raises:
+        ValueError: If no model matches the given criteria.
+    """
+    matching = [m for m in available_models if m["modelName"].lower() == model_name.lower()]
+
+    if vendor_type is not None:
+        matching = [m for m in matching if m.get("vendor", "").lower() == str(vendor_type).lower()]
+
+    if byo_connection_id:
+        matching = [
+            m
+            for m in matching
+            if (byom_details := m.get("byomDetails"))
+            and byom_details.get("integrationServiceConnectionId", "").lower()
+            == byo_connection_id.lower()
+        ]
+
+    if not byo_connection_id and len(matching) > 1:
+        matching = [
+            m
+            for m in matching
+            if (
+                (m.get("modelSubscriptionType", "") == "UiPathOwned")
+                or (m.get("byomDetails") is None)
+            )
+        ]
+
+    if not matching:
+        raise ValueError(
+            f"Model '{model_name}' not found. "
+            f"Available models: {[m['modelName'] for m in available_models]}"
+        )
+
+    return matching[0]

tests/core/features/settings/test_llmgateway.py

@@ -153,6 +153,40 @@ def test_get_available_models_raises_on_unauthorized(self, llmgw_env_vars):
                 assert exc_info.value.status_code == 401
 
 
+class TestLLMGatewayDiscoveryCache:
+    """Tests for get_available_models lru_cache."""
+
+    def test_second_call_returns_cached_result(self, llmgw_env_vars):
+        """Second call should return the cached result without hitting the endpoint."""
+        with patch.dict(os.environ, llmgw_env_vars, clear=True):
+            settings = LLMGatewaySettings()
+
+            mock_response = MagicMock()
+            mock_response.is_error = False
+            mock_response.json.return_value = [{"modelName": "gpt-4o", "vendor": "openai"}]
+
+            with patch.object(Client, "get", return_value=mock_response) as mock_get:
+                first = settings.get_available_models()
+                second = settings.get_available_models()
+                assert first == second
+                mock_get.assert_called_once()
+
+    def test_cache_is_per_instance(self, llmgw_env_vars):
+        """Each settings instance should have its own independent cache."""
+        with patch.dict(os.environ, llmgw_env_vars, clear=True):
+            settings1 = LLMGatewaySettings()
+            settings2 = LLMGatewaySettings()
+
+            mock_response = MagicMock()
+            mock_response.is_error = False
+            mock_response.json.return_value = [{"modelName": "gpt-4o", "vendor": "openai"}]
+
+            with patch.object(Client, "get", return_value=mock_response) as mock_get:
+                settings1.get_available_models()
+                settings2.get_available_models()
+                assert mock_get.call_count == 2
+
+
 class TestLLMGatewayAuthRefresh:
     """Tests for LLMGatewayS2SAuth token refresh logic."""
 

tests/core/features/settings/test_platform.py

@@ -240,6 +240,27 @@ def test_validate_byo_model_is_noop(self, platform_env_vars, mock_platform_auth)
             assert result is None
 
 
+class TestPlatformDiscoveryCache:
+    """Tests for get_available_models lru_cache on PlatformSettings."""
+
+    def test_second_call_returns_cached_result(self, platform_env_vars, mock_platform_auth):
+        """Second call should return the cached result without querying the backend."""
+        with patch.dict(os.environ, platform_env_vars, clear=True):
+            settings = PlatformSettings()
+
+            mock_model = MagicMock()
+            mock_model.model_dump.return_value = {"modelName": "gpt-4o", "vendor": "openai"}
+
+            with patch("uipath.llm_client.settings.platform.settings.UiPath") as mock_uipath:
+                mock_uipath.return_value.agenthub.get_available_llm_models.return_value = [
+                    mock_model
+                ]
+                first = settings.get_available_models()
+                second = settings.get_available_models()
+                assert first == second
+                mock_uipath.return_value.agenthub.get_available_llm_models.assert_called_once()
+
+
 class TestPlatformAuthRefresh:
     """Tests for PlatformAuth token refresh logic."""