dissorial / pdf-chatbot

Chat app that supports multiple PDF files, topics, chat windows, and chat history — powered by GPT.

ChatPDF Plus: GPT-4 x Pinecone x LangChain x MongoDB

Features

• Create multiple topics to chat about
• Google OAuth to log in and store chats in MongoDB associated with your email
• Upload any number of PDF files to each topic
• Create any number of chats (chat windows) for each topic
• Upload files, convert them to embeddings, store the embeddings in a namespace and upload to Pinecone, and delete Pinecone namespaces from within the browser
• Upload and automatically retrieve chat history for all chats using MongoDB

+ LangChain and Pinecone

---

Landing page

---

Main chat area

---

Settings page

---

Mobile layout

---

Disclaimer

This repository was originally a fork of GPT-4 & LangChain repository by mayooear. I planned on making small changes to satisfy my needs but ended up:

• adding several api routes (both for Pinecone and MongoDB)
• creating a bunch of new (large and small) components
• completely changing the styling
• setting up MongoDB
• converting all npm scripts to run within the browser
• and more misc changes

That's why I published this as a standalone repo. Nevertheless, some parts of this readme.md are borrowed from there too.

---

Local setup

Clone the repo

git clone https://github.com/dissorial/pdf-chatbot.git

This repository has two branches: master and old-master. The old-master branch contains working code that you can run locally and has all the features except for Google OAuth. If you want to avoid setting up Google OAuth, clone that branch with:

$ git clone https://github.com/dissorial/pdf-chatbot.git -b old-master

Pinecone setup

Create an account on Pinecone. Go to Indexes and Create index. Enter any name, put 1536 for Dimensions and leave the rest on default. Then go to API keys and Create API key.

MongoDB setup

Create an account on MongoDB. In your dashboard, click New Project and name it. Then, click Create project. You'll be redirected to Database Deployments page. Click Build a database, and choose M0 FREE. Choose any provider and region, but you can leave it as-is. At the bottom, give your cluster a name, and click Create.

Username and password should be pre-filled here. Click Create user. In IP access list, click Add my current API address, then Finish and close and Go to databases. Next to your cluster, there will be a Connect button. Click in, then go to Drivers. You'll find your MongoDB_URI under Add your connection string into your application code. Copy this to your env and replace <password> with your password (the one used when you clicked 'Create user'). Done.

Install packages

yarn install

3. Set up your .env file

• Rename .env.example to .env
• Your .env file should look like this:

OPENAI_API_KEY=

PINECONE_API_KEY=
PINECONE_ENVIRONMENT=
PINECONE_INDEX_NAME=

MONGODB_URI=''

GOOGLE_CLIENT_ID=
GOOGLE_CLIENT_SECRET=

NEXTAUTH_URL=http://localhost:3000
NEXTAUTH_SECRET=

JWT_SECRET=

ALLOWED_EMAIL_DOMAIN='@gmail.com'

NODE_ENV=development

OpenAI API

• Visit OpenAI to retrieve API keys and insert into your .env file

Pinecone

• Visit Pinecone to create and retrieve your API keys, and also retrieve your environment and index name from the dashboard.

MongoDB

• Visit MongoDB, create a database (free one is enough for these purposes) and retrieve your URI in Connect -> Drivers

Google OAuth (Client ID and secret)

• Read this guide from Google

NextAuth Secret

• You can generate this by running npx nextauth secret in your terminal.

JWT Secret

• You can generate this by running npx nextauth jwt-secret in your terminal.

NextAuth URL

• Default is http://localhost:3000. In production, this should be the URL of your deployed app.

Allowed email domain

• '@gmail.com' by default. If you want to use any email domain, remove callback from options in pages/api/auth/[...nextauth].ts

Node environment

• development by default. In production, set this to 'production' (without the quotes)

Other

• In utils/makechain.ts, adjust the QA_PROMPT for your own usecase. Change modelName in new OpenAI to gpt-4, if you have access to gpt-4 api.

---

Run the app

Run npm run dev. Once the local dev environment launches, log in with Google, and then go to Settings in the bottom left corner. Upload your PDF files and give them a 'namespace'. Here, 'namespace' is synonymous with being the topic of your conversation. This way, you can upload multiple files to multiple namespaces, and maintain several conversations about different topics and documents.

---

Chatting with PDF files

If you retrun to the home page now, you should see your namespace(s) on the left sidebar. Click on it, create a new chat and have a conversation with the PDF files embedded for that particular namespace.

---

Troubleshooting

General errors

• Make sure that you are running the latest version of Node. To check your version run node -v.
• If you're encountering issues with a specific PDF, try converting it to text first or try a different PDF file. It's possible that the PDF is corrupted, scanned, or requires OCR to be converted to text.
• Consider logging the env variables and ensure that they are properly exposed.
• Confirm that you're using the same versions of LangChain and Pinecone as this repository.
• Check that you've created an .env file that contains your valid API keys, environment, and index name.
• Ensure that you do not have multiple OPENAPI keys in your global environment. If multiple keys exist, the local env file in the project will be overwritten by the system's env variable.

Pinecone errors

• Make sure that your Pinecone dashboard environment and index match the values in the .env file.
• Confirm that you've set the vector dimensions to 1536.
• Note that Pinecone indexes for users on the Starter (free) plan are deleted after 7 days of inactivity. To prevent this, send an API request to Pinecone to reset the counter before 7 days.
• If issues persist, consider starting fresh with a new Pinecone project, index, and cloned repository.

---

Credit

Frontend of this repo is inspired by ChatGPT.