feat: allow overriding llmgw url for local development

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to `uipath_llm_client` (core package) will be documented in this file.
 
+## [1.20.0] - 2026-09-10
+
+### Added
+- `LLMGatewaySettings` honours `UIPATH_SERVICE_URL_LLMGATEWAY`, the per-service local override the Platform backend already uses. Request and discovery URLs go to that host as `/api/...`, without the `{org}/{tenant}/llmgateway_` prefix the cloud front door adds; the S2S token is still minted at `LLMGW_URL`. Unset, URLs are unchanged.
+
 ## [1.19.0] - 2026-09-09
 
 ### Added

README.md

@@ -237,6 +237,9 @@ settings = LLMGatewaySettings(
 - Either `access_token` OR both `client_id` and `client_secret` must be provided
 - S2S authentication uses `client_id`/`client_secret` to obtain tokens automatically
 
+**Local development:**
+- `UIPATH_SERVICE_URL_LLMGATEWAY=http://localhost:7091` sends request and discovery URLs to a standalone gateway as `/api/...`, without the `{org}/{tenant}/llmgateway_` prefix the cloud front door adds. The S2S token is still minted at `LLMGW_URL`.
+
 ## Usage Examples
 
 ### Quick Start with Direct Client Classes

packages/uipath_langchain_client/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to `uipath_langchain_client` will be documented in this file.
 
+## [1.20.0] - 2026-09-10
+
+### Changed
+- Bumped the `uipath-llm-client` floor to `>=1.20.0`, which adds `UIPATH_SERVICE_URL_LLMGATEWAY` for pointing `LLMGatewaySettings` at a standalone (e.g. local) gateway.
+
 ## [1.19.0] - 2026-09-09
 
 ### Added

packages/uipath_langchain_client/pyproject.toml

@@ -6,7 +6,7 @@ readme = "README.md"
 requires-python = ">=3.11"
 dependencies = [
     "langchain>=1.2.15,<2.0.0",
-    "uipath-llm-client>=1.19.0,<2.0.0",
+    "uipath-llm-client>=1.20.0,<2.0.0",
 ]
 
 [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__ = "1.19.0"
+__version__ = "1.20.0"

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__ = "1.19.0"
+__version__ = "1.20.0"

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

@@ -5,6 +5,7 @@
 from httpx import Client
 from pydantic import Field, SecretStr, model_validator
 from typing_extensions import override
+from uipath.platform.common import resolve_service_url
 
 from uipath.llm_client.settings.base import UiPathAPIConfig, UiPathBaseSettings
 from uipath.llm_client.settings.constants import ApiType, RoutingMode
@@ -59,6 +60,17 @@ class LLMGatewayBaseSettings(UiPathBaseSettings):
         default_factory=dict, validation_alias="LLMGW_ADDITIONAL_HEADERS"
     )
 
+    def _build_gateway_url(self, path: str) -> str:
+        """Build a gateway URL, honouring ``UIPATH_SERVICE_URL_LLMGATEWAY``.
+
+        The ``{org}/{tenant}/llmgateway_`` prefix is added by the cloud front door,
+        so a standalone gateway serves ``/api/...`` directly.
+        """
+        override_url = resolve_service_url(path)
+        if override_url:
+            return override_url
+        return f"{self.base_url}/{self.org_id}/{self.tenant_id}/{path}"
+
     @model_validator(mode="after")
     def validate_auth_settings(self) -> Self:
         """Validate that either access_token or S2S credentials are provided."""
@@ -81,14 +93,19 @@ def build_base_url(
             raise ValueError(
                 "api_config.routing_mode is required for LLMGatewaySettings.build_base_url"
             )
-        base_url = f"{self.base_url}/{self.org_id}/{self.tenant_id}"
         if api_config.routing_mode == RoutingMode.NORMALIZED:
-            url = f"{base_url}/{LLMGatewayEndpoints.NORMALIZED_ENDPOINT.value.format(api_type='chat/completions' if api_config.api_type == ApiType.COMPLETIONS else 'embeddings')}"
+            path = LLMGatewayEndpoints.NORMALIZED_ENDPOINT.value.format(
+                api_type="chat/completions"
+                if api_config.api_type == ApiType.COMPLETIONS
+                else "embeddings"
+            )
         elif api_config.routing_mode == RoutingMode.PASSTHROUGH:
-            url = f"{base_url}/{LLMGatewayEndpoints.PASSTHROUGH_ENDPOINT.value.format(vendor=api_config.vendor_type, model=model_name, api_type=api_config.api_type)}"
+            path = LLMGatewayEndpoints.PASSTHROUGH_ENDPOINT.value.format(
+                vendor=api_config.vendor_type, model=model_name, api_type=api_config.api_type
+            )
         else:
             raise ValueError(f"Unsupported routing_mode: {api_config.routing_mode}")
-        return url
+        return self._build_gateway_url(path)
 
     @override
     def build_auth_headers(
@@ -115,11 +132,17 @@ def build_auth_headers(
 
     @override
     def _discovery_cache_key(self) -> tuple[str, ...]:
-        return (self.base_url, self.org_id, self.tenant_id, self.requesting_product)
+        # Effective URL, so an active service override gets its own cache entry.
+        return (
+            self._build_gateway_url(LLMGatewayEndpoints.DISCOVERY_ENDPOINT.value),
+            self.org_id,
+            self.tenant_id,
+            self.requesting_product,
+        )
 
     @override
     def _fetch_available_models(self) -> list[dict[str, Any]]:
-        discovery_url = f"{self.base_url}/{self.org_id}/{self.tenant_id}/{LLMGatewayEndpoints.DISCOVERY_ENDPOINT.value}"
+        discovery_url = self._build_gateway_url(LLMGatewayEndpoints.DISCOVERY_ENDPOINT.value)
         with Client(
             auth=self.build_auth_pipeline(),
             headers=self.build_auth_headers(),

tests/core/features/settings/test_llmgateway.py

@@ -158,6 +158,96 @@ def test_get_available_models_raises_on_unauthorized(self, llmgw_env_vars):
                 assert exc_info.value.status_code == 401
 
 
+class TestLLMGatewayServiceUrlOverride:
+    """Tests for UIPATH_SERVICE_URL_LLMGATEWAY."""
+
+    LOCAL = "http://localhost:7091"
+
+    def test_urls_unchanged_without_override(self, llmgw_env_vars, passthrough_api_config):
+        with patch.dict(os.environ, llmgw_env_vars, clear=True):
+            settings = LLMGatewaySettings()
+            url = settings.build_base_url(model_name="gpt-4o", api_config=passthrough_api_config)
+
+        assert url == (
+            "https://cloud.uipath.com/test-org-id/test-tenant-id/"
+            "llmgateway_/api/raw/vendor/openai/model/gpt-4o/completions"
+        )
+
+    def test_override_redirects_passthrough_url(self, llmgw_env_vars, passthrough_api_config):
+        env = {**llmgw_env_vars, "UIPATH_SERVICE_URL_LLMGATEWAY": self.LOCAL}
+        with patch.dict(os.environ, env, clear=True):
+            settings = LLMGatewaySettings()
+            url = settings.build_base_url(model_name="gpt-4o", api_config=passthrough_api_config)
+
+        assert url == f"{self.LOCAL}/api/raw/vendor/openai/model/gpt-4o/completions"
+
+    def test_override_redirects_normalized_url(self, llmgw_env_vars, normalized_api_config):
+        env = {**llmgw_env_vars, "UIPATH_SERVICE_URL_LLMGATEWAY": self.LOCAL}
+        with patch.dict(os.environ, env, clear=True):
+            settings = LLMGatewaySettings()
+            url = settings.build_base_url(model_name="gpt-4o", api_config=normalized_api_config)
+
+        assert url == f"{self.LOCAL}/api/chat/completions"
+
+    def test_override_redirects_discovery_url(self, llmgw_env_vars):
+        UiPathBaseSettings._discovery_cache.clear()
+        env = {**llmgw_env_vars, "UIPATH_SERVICE_URL_LLMGATEWAY": self.LOCAL}
+        mock_response = MagicMock()
+        mock_response.is_error = False
+        mock_response.json.return_value = []
+
+        with patch.dict(os.environ, env, clear=True):
+            settings = LLMGatewaySettings()
+            with patch.object(Client, "get", return_value=mock_response) as mock_get:
+                settings.get_available_models(refresh=True)
+
+        assert mock_get.call_args.args[0] == f"{self.LOCAL}/api/discovery"
+
+    def test_override_and_cloud_do_not_share_discovery_cache(self, llmgw_env_vars):
+        UiPathBaseSettings._discovery_cache.clear()
+        mock_response = MagicMock()
+        mock_response.is_error = False
+        mock_response.json.return_value = []
+
+        with patch.object(Client, "get", return_value=mock_response) as mock_get:
+            with patch.dict(os.environ, llmgw_env_vars, clear=True):
+                LLMGatewaySettings().get_available_models()
+            env = {**llmgw_env_vars, "UIPATH_SERVICE_URL_LLMGATEWAY": self.LOCAL}
+            with patch.dict(os.environ, env, clear=True):
+                LLMGatewaySettings().get_available_models()
+
+        assert [c.args[0] for c in mock_get.call_args_list] == [
+            "https://cloud.uipath.com/test-org-id/test-tenant-id/llmgateway_/api/discovery",
+            f"{self.LOCAL}/api/discovery",
+        ]
+
+    def test_unrelated_service_override_is_ignored(self, llmgw_env_vars, passthrough_api_config):
+        env = {**llmgw_env_vars, "UIPATH_SERVICE_URL_AGENTHUB": self.LOCAL}
+        with patch.dict(os.environ, env, clear=True):
+            settings = LLMGatewaySettings()
+            url = settings.build_base_url(model_name="gpt-4o", api_config=passthrough_api_config)
+
+        assert "localhost" not in url
+        assert url.startswith("https://cloud.uipath.com/test-org-id/test-tenant-id/llmgateway_/")
+
+    def test_s2s_token_is_still_minted_at_base_url(self, llmgw_s2s_env_vars):
+        """A standalone gateway has no identity_ endpoint."""
+        from uipath.llm_client.settings.llmgateway.auth import LLMGatewayS2SAuth
+
+        env = {**llmgw_s2s_env_vars, "UIPATH_SERVICE_URL_LLMGATEWAY": self.LOCAL}
+        mock_response = MagicMock()
+        mock_response.is_error = False
+        mock_response.json.return_value = {"access_token": "s2s-token-value"}
+
+        with patch.dict(os.environ, env, clear=True):
+            settings = LLMGatewaySettings()
+            with patch.object(Client, "post", return_value=mock_response) as mock_post:
+                auth = LLMGatewayS2SAuth(settings=settings)
+
+        assert auth.access_token == "s2s-token-value"
+        assert mock_post.call_args.args[0] == "https://cloud.uipath.com/identity_/connect/token"
+
+
 class TestLLMGatewayAuthRefresh:
     """Tests for LLMGatewayS2SAuth token refresh logic."""