app.py

import streamlit as st
import os
from llama_index import (GPTSimpleVectorIndex, LLMPredictor, GPTListIndex)
from llama_index.composability import ComposableGraph
from langchain.llms import OpenAIChat
from langchain.agents import initialize_agent
from llama_index import GPTListIndex
from llama_index.langchain_helpers.memory_wrapper import GPTIndexChatMemory
from llama_index.indices.query.query_transform.base import StepDecomposeQueryTransform
from langchain.agents import Tool
from langchain.chat_models import ChatOpenAI

from streamlit_chat import message

##os.environ['OPENAI_API_KEY'] = ""

# Log
#logging.basicConfig(stream=sys.stdout, level=logging.INFO)
#logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

def build_Messari_agent():
    # Construindo e consultando os index e graphs
    conv_index = GPTListIndex([])
    messari_index = GPTSimpleVectorIndex.load_from_disk('indexMessari.json')
    messari_graph = ComposableGraph.load_from_disk("crypto_index_graph.json")

    messari_index.set_text("Used to answer questions about the Messari Crypto Theses for 2023, "
    "written by Ryan Selkis. It provides predictions and 2023 trends about the crypto industry regarding DeFi, NFTs, "
    "CeFi, People to Watch, Crypto Policy, Ethereum, L1, L2, DAOs and Web3.")

    #Definindo que o LLM usado é o GPT3, para o LlamaIndex e Langchain
    llm = OpenAIChat(model_name="gpt-3.5-turbo")
    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo"))
    
    #adicionando um objeto para quebrar a query
    step_decompose_transform_gpt3 = StepDecomposeQueryTransform(
    llm_predictor, verbose=True)
    
    # Construindo as "tools" do Langchain a serem usadas pelo agent.
    # No momento só tem tool do report da Messari
    tools = [
        Tool(
            name = "Messari Index",
            func=lambda q: str(messari_index.query(q, query_transform=step_decompose_transform_gpt3)),
            description="Useful to answer questions about the Messari Crypto Theses for 2023, "
"written by Ryan Selkis. It provides predictions and 2023 trends about the crypto industry regarding DeFi, NFTs, "
"CeFi, People to Watch, Crypto Policy, Ethereum, L1, L2, DAOs and Web3.",
            return_direct=True
        ),
    ]

    # Objeto de memória, que tb usa o objeto de index da LlamaIndex para guardar a memória do chat
    memory = GPTIndexChatMemory(
        index=conv_index,
        memory_key="chat_history", 
        query_kwargs={"response_mode": "compact"},
        # return_source returns source nodes instead of querying index
        return_source=True,
        # return_messages returns context in message format
        return_messages=True
    )

    #Construindo o agente do Langchain, passando o objeto de memoria, as ferramentas e o llm.
    #Verbose=true mostra o passo a passo do raciocinio da AI
    v_agent_chain = initialize_agent(tools, llm, agent="conversational-react-description", memory=memory, verbose=True)
    return v_agent_chain

# Inicializando os agentes de chat, e os inputs/outputs do chat
if 'messariAgent' not in st.session_state:
    st.session_state.messariAgent = build_Messari_agent()

if 'replies' not in st.session_state:
    st.session_state.replies = []

if 'prompts' not in st.session_state:
    st.session_state.prompts = []

st.title('Messari Report Chat\n')

with st.sidebar.form("chat_form"):
   prompt = st.text_area('Ask something about Messari´s report.')
   # Every form must have a submit button.
   submitted = st.form_submit_button("Submit")

if submitted:
    st.session_state.prompts.append(prompt)
    reply = st.session_state.messariAgent.run(input=prompt)
    st.session_state.replies.append(reply)

if 'replies' in st.session_state:
    
    for i in range(len(st.session_state.replies)-1, -1, -1):
        message(st.session_state.replies[i], key=str(i))
        message(st.session_state.prompts[i], is_user=True, key=str(i) + '_user')