app.py

import streamlit as st
from streamlit_chat import message
import os
from utils import (
    parse_docx,
    parse_pdf,
    parse_txt,
    parse_csv,
    parse_pptx,
    search_docs,
    embed_docs,
    text_to_docs,
    get_answer,
    get_sources,
    wrap_text_in_html,
)
from openai.error import OpenAIError

def clear_submit():
    st.session_state["submit"] = False

def set_openai_api_key(api_key: str):
    st.session_state["OPENAI_API_KEY"] = api_key

st.markdown('<h1>File GPT 🤖<small> by <a href="https://codegpt.co">Code GPT</a></small></h1>', unsafe_allow_html=True)

# Sidebar
index = None
doc = None
with st.sidebar:
    user_secret = st.text_input(
        "OpenAI API Key",
        type="password",
        placeholder="Paste your OpenAI API key here (sk-...)",
        help="You can get your API key from https://platform.openai.com/account/api-keys.",
        value=st.session_state.get("OPENAI_API_KEY", ""),
    )
    if user_secret:
        set_openai_api_key(user_secret)

    uploaded_file = st.file_uploader(
        "Upload a pdf, docx, or txt file",
        type=["pdf", "docx", "txt", "csv", "pptx"],
        help="Scanned documents are not supported yet!",
        on_change=clear_submit,
    )

    if uploaded_file is not None:
        if uploaded_file.name.endswith(".pdf"):
            doc = parse_pdf(uploaded_file)
        elif uploaded_file.name.endswith(".docx"):
            doc = parse_docx(uploaded_file)
        elif uploaded_file.name.endswith(".csv"):
            doc = parse_csv(uploaded_file)
        elif uploaded_file.name.endswith(".txt"):
            doc = parse_txt(uploaded_file)
        elif uploaded_file.name.endswith(".pptx"):
            doc = parse_pptx(uploaded_file)
        else:
            st.error("File type not supported")
            doc = None
        text = text_to_docs(doc)
        st.write(text)
        try:
            with st.spinner("Indexing document... This may take a while⏳"):
                index = embed_docs(text)
                st.session_state["api_key_configured"] = True
        except OpenAIError as e:
            st.error(e._message)

tab1, tab2 = st.tabs(["Intro", "Chat with the File"])
with tab1:
    st.markdown("### How does it work?")
    st.write("File GPT is a tool that allows you to ask questions about a document and get answers from the document. The tool uses the OpenAI API to embed the document and then uses the Embedding API to find the most similar documents to the question. The tool then uses LangChain to obtain the answer from the most similar documents.")
    st.write("The tool is currently in beta and is not perfect. It is recommended to use it with short documents.")
    st.write("""---""")
    st.markdown("### How to use it?")
    st.write("To use the tool you must first add your OpenAI API Key and then upload a document. The tool currently supports the following file types: pdf, docx, txt, csv, pptx. Once the document is uploaded, the tool will index the document and embed it. This may take a while depending on the size of the document. Once the document is indexed, you can ask questions about the document. The tool will return the answer to the question and the source of the answer.")
    st.markdown('<p>Read the article to know more details: <a target="_blank" href="https://medium.com/@dan.avila7/file-gpt-conversaci%C3%B3n-por-chat-con-un-archivo-698d17570358">Medium Article (Spanish)</a></p>', unsafe_allow_html=True)
    st.write("## File GPT was written with the following tools:")
    st.markdown("#### Code GPT")
    st.write('All code was written with the help of Code GPT. Visit https://codegpt.co to get the extension.')
    st.markdown("#### Streamlit")
    st.write('The design was written with <a target="_blank" href="https://streamlit.io/">Streamlit</a>.', unsafe_allow_html=True)
    st.markdown("#### LangChain")
    st.write('Question answering with source <a target="_blank" href="https://langchain.readthedocs.io/en/latest/use_cases/question_answering.html#adding-in-sources">Langchain QA</a>.', unsafe_allow_html=True)
    st.markdown("#### Embedding")
    st.write('<a target="_blank" href="https://platform.openai.com/docs/guides/embeddings">Embedding</a> is done via the OpenAI API with "text-embedding-ada-002"', unsafe_allow_html=True)
    st.write("Please note that you must have credits in your OpenAI account to use this tool. Each file uploaded to the platform consumes credits for embedding and each query consumes credits to obtain the response.")
    st.markdown("""---""")
    st.write('Author:  <a target="_blank" href="https://www.linkedin.com/in/daniel-avila-arias/">Daniel Avila</a>', unsafe_allow_html=True)
    st.write('Repo: <a target="_blank" href="https://github.com/davila7/file-gpt">Github</a>', unsafe_allow_html=True)
    st.write("This software was developed with Code GPT, for more information visit: https://codegpt.co", unsafe_allow_html=True)

with tab2:
    st.write('To obtain an API Key you must create an OpenAI account at the following link: https://openai.com/api/')
    if 'generated' not in st.session_state:
        st.session_state['generated'] = []

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

    def get_text():
        if user_secret:
            st.header("Ask me something about the document:")
            input_text = st.text_area("You:", on_change=clear_submit)
            return input_text
    user_input = get_text()

    button = st.button("Submit")
    if button or st.session_state.get("submit"):
        if not user_input:
            st.error("Please enter a question!")
        else:
            st.session_state["submit"] = True
            sources = search_docs(index, user_input)
            try:
                answer = get_answer(sources, user_input)
                st.session_state.past.append(user_input)
                st.session_state.generated.append(answer["output_text"].split("SOURCES: ")[0])
            except OpenAIError as e:
                st.error(e._message)
            if st.session_state['generated']:
                for i in range(len(st.session_state['generated'])-1, -1, -1):
                    message(st.session_state["generated"][i], key=str(i))
                    message(st.session_state['past'][i], is_user=True, key=str(i) + '_user')