Merge pull request #957 from Pingdred/no-async

No async because Piero is paniking

Author: pieroit

core/cat/agents/base_agent.py

@@ -13,5 +13,5 @@ class AgentOutput(BaseModelDict):
 class BaseAgent(ABC):
 
     @abstractmethod
-    async def execute(*args, **kwargs) -> AgentOutput:
+    def execute(*args, **kwargs) -> AgentOutput:
         pass

core/cat/agents/form_agent.py

@@ -5,7 +5,7 @@
 
 class FormAgent(BaseAgent):
 
-    async def execute(self, stray) -> AgentOutput:
+    def execute(self, stray) -> AgentOutput:
 
         # get active form from working memory
         active_form = stray.working_memory.active_form

core/cat/agents/main_agent.py

@@ -26,7 +26,7 @@ def __init__(self):
         else:
             self.verbose = False
 
-    async def execute(self, stray) -> AgentOutput:
+    def execute(self, stray) -> AgentOutput:
         """Execute the agents.
 
         Returns
@@ -66,15 +66,15 @@ async def execute(self, stray) -> AgentOutput:
 
         # run tools and forms
         procedures_agent = ProceduresAgent()
-        procedures_agent_out : AgentOutput = await procedures_agent.execute(stray)
+        procedures_agent_out : AgentOutput = procedures_agent.execute(stray)
         if procedures_agent_out.return_direct:
             return procedures_agent_out
 
         # we run memory agent if:
         # - no procedures were recalled or selected or
         # - procedures have all return_direct=False
         memory_agent = MemoryAgent()
-        memory_agent_out : AgentOutput = await memory_agent.execute(
+        memory_agent_out : AgentOutput = memory_agent.execute(
             # TODO: should all agents only receive stray?
             stray, prompt_prefix, prompt_suffix
         )

core/cat/agents/memory_agent.py

@@ -11,7 +11,7 @@
 
 class MemoryAgent(BaseAgent):
 
-    async def execute(self, stray, prompt_prefix, prompt_suffix) -> AgentOutput:
+    def execute(self, stray, prompt_prefix, prompt_suffix) -> AgentOutput:
 
         prompt_variables = stray.working_memory.agent_input.model_dump()
         sys_prompt = prompt_prefix + prompt_suffix

core/cat/agents/procedures_agent.py

@@ -24,10 +24,10 @@ class ProceduresAgent(BaseAgent):
     form_agent = FormAgent()
     allowed_procedures: Dict[str, CatTool | CatForm] = {}
 
-    async def execute(self, stray) -> AgentOutput:
+    def execute(self, stray) -> AgentOutput:
 
         # Run active form if present
-        form_output: AgentOutput = await self.form_agent.execute(stray)
+        form_output: AgentOutput = self.form_agent.execute(stray)
         if form_output.return_direct:
             return form_output
 
@@ -38,7 +38,7 @@ async def execute(self, stray) -> AgentOutput:
             log.debug(f"Procedural memories retrived: {len(procedural_memories)}.")
 
             try:
-                procedures_result: AgentOutput = await self.execute_procedures(stray)
+                procedures_result: AgentOutput = self.execute_procedures(stray)
                 if procedures_result.return_direct:
                     # exit agent if a return_direct procedure was executed
                     return procedures_result
@@ -64,7 +64,7 @@ async def execute(self, stray) -> AgentOutput:
         return AgentOutput()
 
 
-    async def execute_procedures(self, stray):
+    def execute_procedures(self, stray):
 
         # using some hooks
         mad_hatter = MadHatter()
@@ -87,13 +87,13 @@ async def execute_procedures(self, stray):
             )
 
         # Execute chain and obtain a choice of procedure from the LLM
-        llm_action: LLMAction = await self.execute_chain(stray, procedures_prompt_template, allowed_procedures)
+        llm_action: LLMAction = self.execute_chain(stray, procedures_prompt_template, allowed_procedures)
 
         # route execution to subagents
-        return await self.execute_subagents(stray, llm_action, allowed_procedures)
+        return self.execute_subagents(stray, llm_action, allowed_procedures)
 
 
-    async def execute_chain(self, stray, procedures_prompt_template, allowed_procedures) -> LLMAction:
+    def execute_chain(self, stray, procedures_prompt_template, allowed_procedures) -> LLMAction:
 
         # Prepare info to fill up the prompt
         prompt_variables = {
@@ -136,15 +136,15 @@ async def execute_chain(self, stray, procedures_prompt_template, allowed_procedu
         return llm_action
 
 
-    async def execute_subagents(self, stray, llm_action, allowed_procedures):
+    def execute_subagents(self, stray, llm_action, allowed_procedures):
         # execute chosen tool / form
         # loop over allowed tools and forms
         if llm_action.action:
             chosen_procedure = allowed_procedures.get(llm_action.action, None)
             try:
                 if Plugin._is_cat_tool(chosen_procedure):
                     # execute tool
-                    tool_output = await chosen_procedure._arun(llm_action.action_input, stray=stray)
+                    tool_output = chosen_procedure.run(llm_action.action_input, stray=stray)
                     return AgentOutput(
                         output=tool_output,
                         return_direct=chosen_procedure.return_direct,
@@ -158,7 +158,7 @@ async def execute_subagents(self, stray, llm_action, allowed_procedures):
                     # store active form in working memory
                     stray.working_memory.active_form = form_instance
                     # execute form
-                    return await self.form_agent.execute(stray)
+                    return self.form_agent.execute(stray)
 
             except Exception as e:
                 log.error(f"Error executing {chosen_procedure.procedure_type} `{chosen_procedure.name}`")

core/cat/looking_glass/stray_cat.py

@@ -43,8 +43,6 @@ def __init__(
 
         self.__main_loop = main_loop
 
-        self.__loop = asyncio.new_event_loop()
-
     def __repr__(self):
         return f"StrayCat(user_id={self.user_id})"
 
@@ -342,7 +340,7 @@ def llm(self, prompt: str, stream: bool = False) -> str:
 
         return output
 
-    async def __call__(self, message_dict):
+    def __call__(self, message_dict):
         """Call the Cat instance.
 
         This method is called on the user's message received from the client.
@@ -408,7 +406,7 @@ async def __call__(self, message_dict):
 
         # reply with agent
         try:
-            agent_output: AgentOutput = await self.main_agent.execute(self)
+            agent_output: AgentOutput = self.main_agent.execute(self)
         except Exception as e:
             # This error happens when the LLM
             #   does not respect prompt instructions.
@@ -472,7 +470,7 @@ async def __call__(self, message_dict):
 
     def run(self, user_message_json, return_message=False):
         try:
-            cat_message = self.loop.run_until_complete(self.__call__(user_message_json))
+            cat_message = self.__call__(user_message_json)
             if return_message:
                 # return the message for HTTP usage
                 return cat_message
@@ -648,7 +646,3 @@ def main_agent(self):
     @property
     def white_rabbit(self):
         return CheshireCat().white_rabbit
-
-    @property
-    def loop(self):
-        return self.__loop

core/cat/mad_hatter/decorators/tool.py

@@ -9,7 +9,7 @@
 
 # All @tool decorated functions in plugins become a CatTool.
 # The difference between base langchain Tool and CatTool is that CatTool has an instance of the cat as attribute (set by the MadHatter)
-class CatTool(BaseTool):
+class CatTool:
     def __init__(
         self,
         name: str,
@@ -18,12 +18,7 @@ def __init__(
         examples: List[str] = [],
     ):
         description = func.__doc__.strip()
-
-        # call parent contructor
-        super().__init__(
-            name=name, func=func, description=description, return_direct=return_direct
-        )
-
+        
         self.func = func
         self.procedure_type = "tool"
         self.name = name
@@ -44,21 +39,8 @@ def start_examples(self):
     def __repr__(self) -> str:
         return f"CatTool(name={self.name}, return_direct={self.return_direct}, description={self.description})"
 
-    # we run tools always async, even if they are not defined so in a plugin
-    def _run(self, input_by_llm: str) -> str:
-        pass # do nothing
-    
-    # we run tools always async, even if they are not defined so in a plugin
-    async def _arun(self, input_by_llm, stray):
-
-        # await if the tool is async
-        if inspect.iscoroutinefunction(self.func):
-            return await self.func(input_by_llm, cat=stray)
-
-        # run in executor if the tool is not async
-        return await stray.loop.run_in_executor(
-            None, self.func, input_by_llm, stray
-        )
+    def run(self, input_by_llm: str, stray) -> str:
+        return self.func(input_by_llm, cat=stray)
 
     # override `extra = 'forbid'` for Tool pydantic model in langchain
     class Config:

core/tests/agents/test_main_agent.py

@@ -17,7 +17,7 @@ def test_main_agent_instantiation(main_agent):
 @pytest.mark.asyncio  # to test async functions
 async def test_execute_main_agent(main_agent, stray):
     # empty agent execution
-    out = await main_agent.execute(stray)
+    out = main_agent.execute(stray)
     assert isinstance(out, AgentOutput)
     assert not out.return_direct
     assert out.intermediate_steps == []

core/tests/looking_glass/test_stray_cat.py

@@ -29,7 +29,7 @@ def test_stray_nlp(stray):
 def test_stray_call(stray):
     msg = {"text": "Where do I go?", "user_id": "Alice"}
 
-    reply = stray.loop.run_until_complete(stray.__call__(msg))
+    reply = stray.__call__(msg)
 
     assert isinstance(reply, CatMessage)
     assert "You did not configure" in reply.content
@@ -57,7 +57,7 @@ def test_recall_to_working_memory(stray):
     msg = {"text": msg_text, "user_id": "Alice"}
 
     # send message
-    stray.loop.run_until_complete(stray.__call__(msg))
+    stray.__call__(msg)
 
     # recall after episodic memory was stored
     stray.recall_relevant_memories_to_working_memory(msg_text)