feat: implement fallback LLMs for agent execution

src/crewai/agent.py

@@ -91,6 +91,9 @@ class Agent(BaseAgent):
     function_calling_llm: Optional[Union[str, InstanceOf[BaseLLM], Any]] = Field(
         description="Language model that will run the agent.", default=None
     )
+    fallback_llms: Optional[List[Union[str, InstanceOf[BaseLLM], Any]]] = Field(
+        default=None, description="List of fallback language models to try if the primary LLM fails."
+    )
     system_template: Optional[str] = Field(
         default=None, description="System format for the agent."
     )
@@ -174,6 +177,8 @@ def post_init_setup(self):
         self.agent_ops_agent_name = self.role
 
         self.llm = create_llm(self.llm)
+        if self.fallback_llms:
+            self.fallback_llms = [create_llm(fallback_llm) for fallback_llm in self.fallback_llms]
         if self.function_calling_llm and not isinstance(
             self.function_calling_llm, BaseLLM
         ):

src/crewai/agents/crew_agent_executor.py

@@ -159,6 +159,7 @@ def _invoke_loop(self) -> AgentFinish:
                     messages=self.messages,
                     callbacks=self.callbacks,
                     printer=self._printer,
+                    fallback_llms=getattr(self.agent, 'fallback_llms', None),
                 )
                 formatted_answer = process_llm_response(answer, self.use_stop_words)
 

src/crewai/lite_agent.py

@@ -526,6 +526,7 @@ def _invoke_loop(self) -> AgentFinish:
                         messages=self._messages,
                         callbacks=self._callbacks,
                         printer=self._printer,
+                        fallback_llms=getattr(self, 'fallback_llms', None),
                     )
 
                     # Emit LLM call completed event

src/crewai/utilities/agent_utils.py

@@ -145,27 +145,52 @@ def get_llm_response(
     messages: List[Dict[str, str]],
     callbacks: List[Any],
     printer: Printer,
+    fallback_llms: Optional[List[Union[LLM, BaseLLM]]] = None,
 ) -> str:
-    """Call the LLM and return the response, handling any invalid responses."""
-    try:
-        answer = llm.call(
-            messages,
-            callbacks=callbacks,
-        )
-    except Exception as e:
-        printer.print(
-            content=f"Error during LLM call: {e}",
-            color="red",
-        )
-        raise e
-    if not answer:
-        printer.print(
-            content="Received None or empty response from LLM call.",
-            color="red",
-        )
-        raise ValueError("Invalid response from LLM call - None or empty.")
-
-    return answer
+    """Call the LLM and return the response, handling any invalid responses and trying fallbacks if available."""
+    llms_to_try = [llm]
+    if fallback_llms:
+        llms_to_try.extend(fallback_llms)
+
+    last_exception = None
+
+    for i, current_llm in enumerate(llms_to_try):
+        try:
+            answer = current_llm.call(
+                messages,
+                callbacks=callbacks,
+            )
+            if not answer:
+                error_msg = "Received None or empty response from LLM call."
+                printer.print(content=error_msg, color="red")
+                if i < len(llms_to_try) - 1:
+                    printer.print(content=f"Trying fallback LLM {i+1}...", color="yellow")
+                    continue
+                else:
+                    raise ValueError("Invalid response from LLM call - None or empty.")
+            return answer
+        except Exception as e:
+            last_exception = e
+            if i == 0:
+                printer.print(content=f"Primary LLM failed: {e}", color="red")
+            else:
+                printer.print(content=f"Fallback LLM {i} failed: {e}", color="red")
+
+            if e.__class__.__module__.startswith("litellm"):
+                error_str = str(e).lower()
+                if any(term in error_str for term in ["authentication", "api key", "unauthorized", "forbidden"]):
+                    printer.print(content="Authentication error detected, skipping remaining fallbacks", color="red")
+                    raise e
+
+            if i < len(llms_to_try) - 1:
+                printer.print(content=f"Trying fallback LLM {i+1}...", color="yellow")
+                continue
+
+    printer.print(content="All LLMs failed, raising last exception", color="red")
+    if last_exception is not None:
+        raise last_exception
+    else:
+        raise RuntimeError("All LLMs failed but no exception was captured")
 
 
 def process_llm_response(

tests/agent_test.py

@@ -1984,7 +1984,7 @@ def test_crew_agent_executor_litellm_auth_error():
         )
 
     # Verify error handling messages
-    error_message = f"Error during LLM call: {str(mock_llm_call.side_effect)}"
+    error_message = f"Primary LLM failed: {str(mock_llm_call.side_effect)}"
     mock_printer.assert_any_call(
         content=error_message,
         color="red",