app.py

from openai import OpenAI
import gradio as gr
import json
from bot_actions import functions_dictionary
import os
 
CSS ="""
.contain { display: flex; flex-direction: column; }
.svelte-vt1mxs div:first-child { flex-grow: 1; overflow: auto;}
#chatbot { flex-grow: 1; overflow: auto;}
footer {display: none !important;}
.app.svelte-182fdeq.svelte-182fdeq {
  max-width: 100vw !important;
}
#main_container {
  height: 95vh;
}
#markup_container {
  height: 100%;
  overflow:auto;
}
"""

openAIToken = os.environ['openAIToken']
assistantId = os.environ['assistantId']
initial_message = os.environ['initialMessage']

client = OpenAI(api_key=openAIToken)

def handle_requires_action(data):
  actions_results = []
  for tool in data.required_action.submit_tool_outputs.tool_calls:
    function_name = tool.function.name
    function_args = json.loads(tool.function.arguments)
    print(function_name)
    print(function_args)
    try:
      result = functions_dictionary[tool.function.name](**function_args)
      print("Function result:", result)
      actions_results.append({"tool_output" : {"tool_call_id": tool.id, "output": result["message"]}})
    except Exception as e:
      print(e)

    
  # Submit all tool_outputs at the same time
  return actions_results 


def create_thread_openai(sessionStorage):
  streaming_thread = client.beta.threads.create()
  sessionStorage["threadId"] = streaming_thread.id
  return sessionStorage

def add_message_to_openai(text, threadId):
  print("User message: ", text)
  return client.beta.threads.messages.create(
    thread_id=threadId,
    role="user",
    content=text
  )


def process_text_chunk(text, storage):
  print(text, end="", flush=True)
  local_message = None
  accumulative_string = storage["accumulative_string"] + text
  local_message = accumulative_string
  return local_message, storage

def handle_events(threadId, chat_history, storage):
  storage.update({
    "accumulative_string" : "",
    "markup_string": "",
  })
  try:
    with client.beta.threads.runs.stream(
      thread_id=threadId,
      assistant_id=assistantId
    ) as stream:
      for event in stream:
        if event.event == "thread.message.delta" and event.data.delta.content:
          text = event.data.delta.content[0].text.value
          local_message, storage = process_text_chunk(text, storage)
          if local_message is not None:
            chat_history[-1][1] += local_message 
          yield [chat_history,  storage]
        if event.event == 'thread.run.requires_action':
          result = handle_requires_action(event.data)
          tool_outputs = [x["tool_output"] for x in result]
          with client.beta.threads.runs.submit_tool_outputs_stream(
            thread_id=stream.current_run.thread_id,
            run_id=event.data.id,
            tool_outputs=tool_outputs,
          ) as action_stream:
            for text in action_stream.text_deltas:
              local_message, storage = process_text_chunk(text, storage)
              if local_message is not None:
                chat_history[-1][1] += local_message
              yield [chat_history, storage]
            action_stream.close()
      stream.until_done()
      print("")
      return [chat_history, storage]
  except Exception as e:
    print(e)
    chat_history[-1][1] = "Error occured during processing your message. Please try again"
    yield [chat_history, storage]

def check_moderation_flag(message):
  moderation_response = client.moderations.create(input=message, model="omni-moderation-latest")
  print("Moderation respones: ", moderation_response)
  flagged = moderation_response.results[0].flagged
  return flagged

def process_user_input(text, thread_id, chat_history, storage):
  print("User input: ", text)
  is_flagged = check_moderation_flag(text)
  print("Check is flagged:", is_flagged)
  if is_flagged:
    chat_history[-1][1] = "Your request contains some inappropriate information. We cannot proceed with it."
    yield [chat_history, storage]
  else:
    add_message_to_openai(text, thread_id)
    for response in handle_events(thread_id, chat_history, storage):
      yield response

def initiate_chatting(chat_history, storage):
  threadId = storage["threadId"]
  chat_history = [[None, ""]]
  for response in process_user_input(initial_message, threadId, chat_history, storage):
    yield response

def respond_on_user_msg(chat_history, storage):
  message = chat_history[-1][0]
  threadId = storage["threadId"]
  print("Responding for threadId: ", threadId)
  chat_history[-1][1] = ""
  for response in process_user_input(message, threadId, chat_history, storage):
    yield response

def create_chat_tab():
  msg = gr.Textbox(label="Answer")
  storage = gr.State({"accumulative_string": ""})
  chatbot = gr.Chatbot(label="Board of Advisors Assistant", line_breaks=False, height=300, show_label=False, show_share_button=False, elem_id="chatbot")

  def user(user_message, history):
    return "", history + [[user_message, None]]

  def disable_msg():
    message_box = gr.Textbox(value=None, interactive=False)
    return message_box
  
  def enable_msg():
    message_box = gr.Textbox(value=None, interactive=True)
    return message_box

  add_user_message_flow = [user, [msg,chatbot],  [msg,chatbot]]
  chat_response_flow = [respond_on_user_msg, [chatbot, storage], [chatbot, storage]]
  disable_msg_flow = [disable_msg, None, msg] 
  enable_msg_flow = [enable_msg, None, msg]

  with gr.Blocks(css=CSS, fill_height=True) as chat_view:
    storage.render()
    with gr.Row(elem_id="main_container"):
      with gr.Column(scale=4):
        chatbot.render()
        examples = gr.Examples(examples=[
          "I need someone that can help me with real estate in Texas",
          "I'm looking for help with payment system for my business",
          "I need help to develop my leadership skills"], 
          inputs=msg, 
          )
        msg.render()
    print(gr.Request)

    msg.submit(*add_user_message_flow
        ).then(*disable_msg_flow
        ).then(*chat_response_flow
        ).then(*enable_msg_flow)

    examples.load_input_event.then(*add_user_message_flow
        ).then(*disable_msg_flow
        ).then(*chat_response_flow
        ).then(*enable_msg_flow)
    
    chat_view.load(*disable_msg_flow
        ).then(create_thread_openai, inputs=storage, outputs=storage
        ).then(initiate_chatting, inputs=[chatbot, storage], outputs=[chatbot, storage]
        ).then(*enable_msg_flow)
  return chat_view

if __name__ == "__main__":
  chat_view = create_chat_tab()
  chat_view.launch()