Adds OpenAIConversationalAgent (#746)

src/backend/langflow/components/agents/OpenAIConversationalAgent.py

@@ -0,0 +1,76 @@
+from langflow import CustomComponent
+from typing import Optional
+from langchain.prompts import SystemMessagePromptTemplate
+from langchain.tools import Tool
+from langchain.schema.memory import BaseMemory
+from langchain.chat_models import ChatOpenAI
+
+from langchain.agents.agent import AgentExecutor
+from langchain.agents.openai_functions_agent.base import OpenAIFunctionsAgent
+from langchain.memory.token_buffer import ConversationTokenBufferMemory
+from langchain.prompts.chat import MessagesPlaceholder
+from langchain.agents.agent_toolkits.conversational_retrieval.openai_functions import (
+    _get_default_system_message,
+)
+
+
+class ConversationalAgent(CustomComponent):
+    display_name: str = "OpenaAI Conversational Agent"
+    description: str = "Conversational Agent that can use OpenAI's function calling API"
+
+    def build_config(self):
+        openai_function_models = [
+            "gpt-3.5-turbo-0613",
+            "gpt-3.5-turbo-16k-0613",
+            "gpt-4-0613",
+            "gpt-4-32k-0613",
+        ]
+        return {
+            "tools": {"is_list": True, "display_name": "Tools"},
+            "memory": {"display_name": "Memory"},
+            "system_message": {"display_name": "System Message"},
+            "max_token_limit": {"display_name": "Max Token Limit"},
+            "model_name": {
+                "display_name": "Model Name",
+                "options": openai_function_models,
+                "value": openai_function_models[0],
+            },
+            "code": {"show": False},
+        }
+
+    def build(
+        self,
+        model_name: str,
+        tools: Tool,
+        memory: Optional[BaseMemory] = None,
+        system_message: Optional[SystemMessagePromptTemplate] = None,
+        max_token_limit: int = 2000,
+    ) -> AgentExecutor:
+        llm = ChatOpenAI(model=model_name)
+        if not memory:
+            memory_key = "chat_history"
+            memory = ConversationTokenBufferMemory(
+                memory_key=memory_key,
+                return_messages=True,
+                output_key="output",
+                llm=llm,
+                max_token_limit=max_token_limit,
+            )
+        else:
+            memory_key = memory.memory_key  # type: ignore
+
+        _system_message = system_message or _get_default_system_message()
+        prompt = OpenAIFunctionsAgent.create_prompt(
+            system_message=_system_message,  # type: ignore
+            extra_prompt_messages=[MessagesPlaceholder(variable_name=memory_key)],
+        )
+        agent = OpenAIFunctionsAgent(
+            llm=llm, tools=tools, prompt=prompt  # type: ignore
+        )
+        return AgentExecutor(
+            agent=agent,
+            tools=tools,  # type: ignore
+            memory=memory,
+            verbose=True,
+            return_intermediate_steps=True,
+        )

src/backend/langflow/interface/custom/constants.py

@@ -8,6 +8,8 @@ from langchain.text_splitter import TextSplitter
 from langchain.tools import Tool
 from langchain.vectorstores.base import VectorStore
 from langchain.schema import BaseOutputParser
+from langchain.schema.memory import BaseMemory
+from langchain.memory.chat_memory import BaseChatMemory
 from langchain.agents.agent import AgentExecutor
 
 LANGCHAIN_BASE_TYPES = {
@@ -23,6 +25,8 @@ LANGCHAIN_BASE_TYPES = {
     "Embeddings": Embeddings,
     "BaseRetriever": BaseRetriever,
     "BaseOutputParser": BaseOutputParser,
+    "BaseMemory": BaseMemory,
+    "BaseChatMemory": BaseChatMemory,
 }
 
 # Langchain base types plus Python base types

src/backend/langflow/processing/base.py

@@ -5,6 +5,7 @@ from langflow.api.v1.callback import (
 )
 from langflow.processing.process import fix_memory_inputs, format_actions
 from langflow.utils.logger import logger
+from langchain.agents.agent import AgentExecutor
 
 
 async def get_result_and_steps(langchain_object, inputs: Union[dict, str], **kwargs):
@@ -20,6 +21,7 @@ async def get_result_and_steps(langchain_object, inputs: Union[dict, str], **kwa
             # to display intermediate steps
             langchain_object.return_intermediate_steps = True
         try:
+            if not isinstance(langchain_object, AgentExecutor):
                 fix_memory_inputs(langchain_object)
         except Exception as exc:
             logger.error(f"Error fixing memory inputs: {exc}")