An API key is no longer required if base_url is pointed to a non-default endpoint

tests/test_framework_llm.py

@@ -199,7 +199,7 @@ def test_client_cache_identity_uses_timeout(monkeypatch):
 def test_missing_api_key_error_is_provider_neutral():
     client = OpenAICompatibleStructuredLLMClient()
 
-    with pytest.raises(AuditaLLMError, match="AUDITA_VALIDATION_LLM_API_KEY"):
+    with pytest.raises(AuditaLLMError, match="OpenRouter endpoint"):
         client.run_structured(
             stage_name="test-stage",
             messages=[{"role": "user", "content": "Hello"}],
@@ -208,6 +208,32 @@ def test_missing_api_key_error_is_provider_neutral():
         )
 
 
+def test_missing_api_key_is_allowed_for_nondefault_endpoint(monkeypatch):
+    create_calls, openai_inits = _install_fake_llm_modules(monkeypatch)
+    client = OpenAICompatibleStructuredLLMClient()
+
+    client.run_structured(
+        stage_name="test-stage",
+        messages=[{"role": "user", "content": "Hello"}],
+        response_model=DummyResponseModel,
+        config=_config(
+            api_key=None,
+            model="meta-llama/Llama-3.1-8B-Instruct",
+            base_url="http://localhost:8000/v1",
+        ),
+    )
+
+    assert openai_inits == [{"api_key": None, "base_url": "http://localhost:8000/v1", "timeout": 600}]
+    assert create_calls == [
+        {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "messages": [{"role": "user", "content": "Hello"}],
+            "response_model": DummyResponseModel,
+            "max_retries": 3,
+        }
+    ]
+
+
 def test_client_initialization_is_safe_under_concurrent_calls(monkeypatch):
     _, openai_inits = _install_fake_llm_modules(monkeypatch)
     client = OpenAICompatibleStructuredLLMClient()