Use saved api_base and only overwrite if not already configured.

src/client/content/config/tabs/models.py

@@ -149,15 +149,15 @@ def _render_model_selection(model: dict, provider_models: list, action: str) ->
 
 def _render_api_configuration(model: dict, provider_models: list, disable_for_oci: bool) -> dict:
     """Render API configuration UI and return updated model"""
-    api_base = next(
+    litellm_api_base = next(
         (m.get("api_base", "") for m in provider_models if m.get("key") == model["id"]), model.get("api_base", "")
     )
 
     model["api_base"] = st.text_input(
         "Provider URL:",
         help=help_text.help_dict["model_url"],
         key="add_model_url",
-        value=api_base,
+        value=model.get("api_base", litellm_api_base),
         disabled=disable_for_oci,
     )
 

tests/client/content/config/tabs/test_models.py

@@ -151,18 +151,12 @@ def test_page_content_verification(self, app_server, app_test):
 class TestModelFunctions:
     """Test individual functions from models.py"""
 
-    def _setup_function_test(self, app_test):
-        """Helper to set up function testing"""
-        at = app_test(ST_FILE)
-        at.run()  # Initialize session state
-        return at
-
     def test_clear_client_models_function(self, app_server, app_test):
         """Test clear_client_models function behavior"""
         from client.content.config.tabs.models import clear_client_models
 
         assert app_server is not None
-        at = self._setup_function_test(app_test)
+        at = app_test(ST_FILE).run()
 
         # Set up test client settings
         at.session_state.client_settings = {
@@ -189,7 +183,7 @@ def test_clear_client_models_no_matches(self, app_server, app_test):
         from client.content.config.tabs.models import clear_client_models
 
         assert app_server is not None
-        at = self._setup_function_test(app_test)
+        at = app_test(ST_FILE).run()
 
         original_settings = {
             "ll_model": {"model": "different/model"},
@@ -212,7 +206,7 @@ def test_get_models_function(self, app_server, app_test):
         from client.content.config.tabs.models import get_models
 
         assert app_server is not None
-        at = self._setup_function_test(app_test)
+        at = app_test(ST_FILE).run()
 
         # Clear existing model configs to test refresh
         at.session_state.model_configs = None
@@ -231,7 +225,7 @@ def test_get_models_force_refresh(self, app_server, app_test):
         from client.content.config.tabs.models import get_models
 
         assert app_server is not None
-        at = self._setup_function_test(app_test)
+        at = app_test(ST_FILE).run()
 
         # Set some existing data
         at.session_state.model_configs = ["old_data"]
@@ -249,7 +243,7 @@ def test_get_supported_models_function(self, app_server, app_test):
         from client.content.config.tabs.models import get_supported_models
 
         assert app_server is not None
-        at = self._setup_function_test(app_test)
+        at = app_test(ST_FILE).run()
 
         # Get providers using API context
         with patch("client.utils.api_call.state", at.session_state):
@@ -373,7 +367,7 @@ def test_render_model_selection_with_custom_model_id(self, app_server, app_test)
         from client.content.config.tabs.models import _render_model_selection, get_supported_models
 
         assert app_server is not None
-        at = self._setup_function_test(app_test)
+        at = app_test(ST_FILE).run()
 
         # Get actual supported models from API
         with patch("client.utils.api_call.state", at.session_state):
@@ -403,3 +397,97 @@ def test_render_model_selection_with_custom_model_id(self, app_server, app_test)
 
             # The model ID should be preserved
             assert result_model["id"] == custom_model_id
+
+    def test_render_api_configuration_preserves_saved_api_base(self, app_server, app_test):
+        """Test that _render_api_configuration preserves saved api_base over LiteLLM default"""
+        from client.content.config.tabs.models import _render_api_configuration, get_supported_models
+
+        assert app_server is not None
+        at = app_test(ST_FILE).run()
+
+        # Get actual supported models from API
+        with patch("client.utils.api_call.state", at.session_state):
+            supported_models = get_supported_models("ll")
+
+        # Find OpenAI provider
+        openai_provider = next((p for p in supported_models if p["provider"] == "openai"), None)
+        assert openai_provider is not None, "OpenAI provider should be available in supported models"
+
+        provider_models = openai_provider["models"]
+
+        # Find a model that has an api_base in the provider models
+        model_with_api_base = next((m for m in provider_models if m.get("api_base")), None)
+        if model_with_api_base:
+            litellm_api_base = model_with_api_base["api_base"]
+            model_id = model_with_api_base["key"]
+        else:
+            # Fallback if no model has api_base
+            litellm_api_base = "https://api.openai.com/v1"
+            model_id = "gpt-4"
+
+        # Create a model with a SAVED custom api_base (different from LiteLLM default)
+        saved_custom_api_base = "https://my-custom-api.example.com/v1"
+        model = {
+            "id": model_id,
+            "provider": "openai",
+            "type": "ll",
+            "api_base": saved_custom_api_base,  # This is the saved configuration
+        }
+
+        disable_for_oci = False
+
+        with patch("client.content.config.tabs.models.state", at.session_state):
+            # This should preserve the saved api_base, NOT overwrite with LiteLLM default
+            result_model = _render_api_configuration(model, provider_models, disable_for_oci)
+
+            # The saved api_base should be preserved (regression test for commit 5612888)
+            assert result_model["api_base"] == saved_custom_api_base
+            assert result_model["api_base"] != litellm_api_base or saved_custom_api_base == litellm_api_base
+
+    def test_render_api_configuration_uses_litellm_default_when_no_saved_value(self, app_server, app_test):
+        """Test that _render_api_configuration uses LiteLLM default when no saved api_base exists"""
+        from client.content.config.tabs.models import _render_api_configuration, get_supported_models
+
+        assert app_server is not None
+        at = app_test(ST_FILE).run()
+
+        # Get actual supported models from API
+        with patch("client.utils.api_call.state", at.session_state):
+            supported_models = get_supported_models("ll")
+
+        # Find OpenAI provider
+        openai_provider = next((p for p in supported_models if p["provider"] == "openai"), None)
+        assert openai_provider is not None, "OpenAI provider should be available in supported models"
+
+        provider_models = openai_provider["models"]
+
+        # Find a model that has an api_base in the provider models
+        model_with_api_base = next((m for m in provider_models if m.get("api_base")), None)
+        if model_with_api_base:
+            litellm_api_base = model_with_api_base["api_base"]
+            model_id = model_with_api_base["key"]
+        else:
+            # Fallback if no model has api_base
+            litellm_api_base = "https://api.openai.com/v1"
+            model_id = "gpt-4"
+
+        # Create a model WITHOUT a saved api_base (new model being added)
+        model = {
+            "id": model_id,
+            "provider": "openai",
+            "type": "ll",
+            # Note: No "api_base" key - simulating a new model being added
+        }
+
+        disable_for_oci = False
+
+        with patch("client.content.config.tabs.models.state", at.session_state):
+            # This should use the LiteLLM default api_base
+            result_model = _render_api_configuration(model, provider_models, disable_for_oci)
+
+            # Should fall back to LiteLLM default when no saved value exists
+            if model_with_api_base:
+                assert result_model["api_base"] == litellm_api_base
+            else:
+                # If no model has api_base, it should be empty string
+                assert result_model["api_base"] == ""