adds is_interruptible granularity to individual messages (#386)

vocode/streaming/agent/anthropic_agent.py

@@ -91,7 +91,7 @@ async def generate_response(
         human_input,
         conversation_id: str,
         is_interrupt: bool = False,
-    ) -> AsyncGenerator[str, None]:
+    ) -> AsyncGenerator[Tuple[str, bool], None]:
         self.memory.chat_memory.messages.append(HumanMessage(content=human_input))
 
         bot_memory_message = AIMessage(content="")
@@ -115,7 +115,7 @@ async def generate_response(
             if sentence:
                 bot_memory_message.content = bot_memory_message.content + sentence
                 buffer = remainder
-                yield sentence
+                yield sentence, True
             continue
 
     def update_last_bot_message_on_cut_off(self, message: str):

vocode/streaming/agent/base_agent.py

@@ -66,6 +66,10 @@ class AgentInput(TypedModel, type=AgentInputType.BASE.value):
     conversation_id: str
     vonage_uuid: Optional[str]
     twilio_sid: Optional[str]
+    agent_response_tracker: Optional[asyncio.Event] = None
+
+    class Config:
+        arbitrary_types_allowed = True
 
 
 class TranscriptionAgentInput(AgentInput, type=AgentInputType.TRANSCRIPTION.value):
@@ -213,7 +217,7 @@ async def handle_generate_response(
         )
         is_first_response = True
         function_call = None
-        async for response in responses:
+        async for response, is_interruptible in responses:
             if isinstance(response, FunctionCall):
                 function_call = response
                 continue
@@ -222,7 +226,9 @@ async def handle_generate_response(
                 is_first_response = False
             self.produce_interruptible_agent_response_event_nonblocking(
                 AgentResponseMessage(message=BaseMessage(text=response)),
-                is_interruptible=self.agent_config.allow_agent_to_be_cut_off,
+                is_interruptible=self.agent_config.allow_agent_to_be_cut_off
+                and is_interruptible,
+                agent_response_tracker=agent_input.agent_response_tracker,
             )
         # TODO: implement should_stop for generate_responses
         agent_span.end()
@@ -429,5 +435,7 @@ def generate_response(
         human_input,
         conversation_id: str,
         is_interrupt: bool = False,
-    ) -> AsyncGenerator[Union[str, FunctionCall], None]:
+    ) -> AsyncGenerator[
+        Tuple[Union[str, FunctionCall], bool], None
+    ]:  # tuple of the content and whether it is interruptible
         raise NotImplementedError

vocode/streaming/agent/chat_gpt_agent.py

@@ -69,7 +69,9 @@ def get_functions(self):
             for action_config in self.agent_config.actions
         ]
 
-    def get_chat_parameters(self, messages: Optional[List] = None):
+    def get_chat_parameters(
+        self, messages: Optional[List] = None, use_functions: bool = True
+    ):
         assert self.transcript is not None
         messages = messages or format_openai_chat_messages_from_transcript(
             self.transcript, self.agent_config.prompt_preamble
@@ -86,7 +88,7 @@ def get_chat_parameters(self, messages: Optional[List] = None):
         else:
             parameters["model"] = self.agent_config.model_name
 
-        if self.functions:
+        if use_functions and self.functions:
             parameters["functions"] = self.functions
 
         return parameters
@@ -134,10 +136,10 @@ async def generate_response(
         human_input: str,
         conversation_id: str,
         is_interrupt: bool = False,
-    ) -> AsyncGenerator[Union[str, FunctionCall], None]:
+    ) -> AsyncGenerator[Tuple[Union[str, FunctionCall], bool], None]:
         if is_interrupt and self.agent_config.cut_off_response:
             cut_off_response = self.get_cut_off_response()
-            yield cut_off_response
+            yield cut_off_response, False
             return
         assert self.transcript is not None
 
@@ -174,4 +176,4 @@ async def generate_response(
         async for message in collate_response_async(
             openai_get_tokens(stream), get_functions=True
         ):
-            yield message
+            yield message, True

vocode/streaming/agent/echo_agent.py

@@ -17,8 +17,8 @@ async def generate_response(
         human_input,
         conversation_id: str,
         is_interrupt: bool = False,
-    ) -> AsyncGenerator[str, None]:
-        yield human_input
+    ) -> AsyncGenerator[Tuple[str, bool], None]:
+        yield human_input, True
 
     def update_last_bot_message_on_cut_off(self, message: str):
         pass

vocode/streaming/agent/llamacpp_agent.py

@@ -135,7 +135,7 @@ async def generate_response(
         human_input: str,
         conversation_id: str,
         is_interrupt: bool = False,
-    ) -> AsyncGenerator[str, None]:
+    ) -> AsyncGenerator[Tuple[str, bool], None]:
         asyncio.get_event_loop().run_in_executor(
             self.thread_pool_executor,
             lambda input: self.conversation.predict(input=input),
@@ -145,4 +145,4 @@ async def generate_response(
         async for message in collate_response_async(
             self.llamacpp_get_tokens(),
         ):
-            yield str(message)
+            yield str(message), True

vocode/streaming/agent/llm_agent.py

@@ -123,12 +123,12 @@ async def generate_response(
         human_input,
         conversation_id: str,
         is_interrupt: bool = False,
-    ) -> AsyncGenerator[str, None]:
+    ) -> AsyncGenerator[Tuple[str, bool], None]:
         self.logger.debug("LLM generating response to human input")
         if is_interrupt and self.agent_config.cut_off_response:
             cut_off_response = self.get_cut_off_response()
             self.memory.append(self.get_memory_entry(human_input, cut_off_response))
-            yield cut_off_response
+            yield cut_off_response, False
             return
         self.memory.append(self.get_memory_entry(human_input, ""))
         if self.is_first_response and self.first_response:
@@ -146,7 +146,7 @@ async def generate_response(
             sentence = re.sub(r"^\s+(.*)", r" \1", sentence)
             response_buffer += sentence
             self.memory[-1] = self.get_memory_entry(human_input, response_buffer)
-            yield sentence
+            yield sentence, True
 
     def update_last_bot_message_on_cut_off(self, message: str):
         last_message = self.memory[-1]