Improve docstrings and test clarity as requested in PR review

Co-Authored-By: Joe Moura <joao@crewai.com>

src/crewai/crew.py

@@ -589,6 +589,23 @@ class Crew(BaseModel):
         self,
         inputs: Optional[Dict[str, Any]] = None,
     ) -> CrewOutput:
+        """
+        Starts the crew to work on its assigned tasks.
+        
+        This method initializes all agents, sets up their configurations, and executes
+        the tasks according to the specified process (sequential or hierarchical).
+        
+        For each agent, if no function_calling_llm is specified:
+        - Uses the crew's function_calling_llm if available
+        - Otherwise uses the agent's own LLM for function calling, enabling
+          non-OpenAI models to work without requiring OpenAI credentials
+        
+        Args:
+            inputs: Optional dictionary of inputs to be used in task execution
+            
+        Returns:
+            CrewOutput: The result of the crew's execution
+        """
         try:
             for before_callback in self.before_kickoff_callbacks:
                 if inputs is None:

tests/agent_test.py

@@ -1827,6 +1827,9 @@ def test_agent_uses_own_llm_for_function_calling_when_not_specified():
     Test that an agent uses its own LLM for function calling when no function_calling_llm
     is specified, ensuring that non-OpenAI models like Gemini can be used without
     requiring OpenAI API keys.
+    
+    This test verifies the fix for issue #2517, where users would get OpenAI authentication
+    errors even when using non-OpenAI models like Gemini.
     """
     @tool
     def simple_tool(input_text: str) -> str:
@@ -1853,8 +1856,8 @@ def test_agent_uses_own_llm_for_function_calling_when_not_specified():
                 agent.execute_task(task, tools=[simple_tool])
                 
                 args, kwargs = mock_tool_usage.call_args
-                assert kwargs['function_calling_llm'] == agent.llm
+                assert kwargs['function_calling_llm'] == agent.llm, "Agent should use its own LLM for function calling"
-                assert kwargs['function_calling_llm'].model.startswith("gemini")
+                assert kwargs['function_calling_llm'].model.startswith("gemini"), "Function calling LLM should be Gemini"
             except Exception as e:
                 if "OPENAI_API_KEY" in str(e):
                     pytest.fail("Test failed with OpenAI API key error despite using Gemini model")

tests/crew_test.py

@@ -4127,6 +4127,10 @@ def test_crew_agents_use_own_llm_for_function_calling():
     """
     Test that agents in a crew use their own LLM for function calling when no
     function_calling_llm is specified for either the agent or the crew.
+    
+    This test verifies the fix for issue #2517, where users would get OpenAI authentication
+    errors even when using non-OpenAI models like Gemini. The fix ensures that when no
+    function_calling_llm is specified, the agent uses its own LLM for function calling.
     """
     @tool
     def simple_tool(input_text: str) -> str:
@@ -4160,8 +4164,8 @@ def test_crew_agents_use_own_llm_for_function_calling():
                 crew.kickoff()
                 
                 args, kwargs = mock_tool_usage.call_args
-                assert kwargs['function_calling_llm'] == gemini_agent.llm
+                assert kwargs['function_calling_llm'] == gemini_agent.llm, "Agent should use its own LLM for function calling"
-                assert kwargs['function_calling_llm'].model.startswith("gemini")
+                assert kwargs['function_calling_llm'].model.startswith("gemini"), "Function calling LLM should be Gemini"
             except Exception as e:
                 if "OPENAI_API_KEY" in str(e):
                     pytest.fail("Test failed with OpenAI API key error despite using Gemini model")