Skip to content

A quickly customizable retrieval chatbot designed to help you prototype your own application

License

Notifications You must be signed in to change notification settings

Nick777-Pixel/WinterDev-GPT

 
 

Repository files navigation

DocGobbler - A Customizable Retrieval Chatbot

Mr. Doc Gobbler

Welcome to DocGobbler! This project allows you to quickly build out a customizable chatbot trained on your files. Just ingest your documents, edit the `src/configs` files, and deploy! 100% free and open to use

Examples

Key Features of DocGobbler

  • ✅ Chat with Multiple Documents & File Types
  • ✅ API Streaming
  • ✅ View Context Sources & Metadata
  • ✅ Easy to Customize the Entire Project via the src/configs Folder
  • ✅ Light/Dark Mode Theme Switching

📚 Tech Stack

🪜 API Setup

Prerequisites

  1. Copy the .env example file: cp .env.example .env
  2. Get an OpenAI API Key - Link
    • Enter as OPENAI_API_KEY in .env
  3. Make a Pinecone Account (free) - Link

Pinecone Index Setup

Create a new Index:

  • Give it a name

    • Enter as PINECONE_INDEX_NAME in .env
  • Set Dimensions to exactly 1536 (required for embedding models)

  • Set Metric to cosine

  • Select the free starter Pod Type

On the side panel, Click on "API Keys"

  • Copy the ENVIRONMENT location
    • Enter as PINECONE_ENVIRONMENT in .env
  • Reveal and Copy the Value
    • Enter as PINECONE_API_KEY in .env

🖥️ Project Install & Document Ingest

1.) Clone or Fork Repository

git clone https://github.com/PaulGriz/DocGobbler

2.) Install Dependancies

pnpm install

3.) Copy your Documents into the public/docs Directory

  • Note: The public/docs folder is reccomended because it allows Next.js to easily link to sources files.

4.) Run pnpm run ingest to chuck and vectorize docs into Pincone Index

  • Note: This can take a few minutes. Do not abort the command!
  • You should see Loading X chunks into pinecone... where "X" is the number of vectors from your docs.

5.) Go to src/configs/ai-configs.ts and edit the QA_TEMPLATE

  • The QA_TEMPLATE is the "System Prompt" for the AI. Customize it to fit your documents or objective for your own goals.
  • The CONDENSE_TEMPLATE is not as important, and I'm using the standard prompt from other Langchain projects.

6.) 🏃‍♂️ Run the Project

  • pnpm run dev

🎨 How to Customize

Use the src/configs/ folder to make this project your own!

  • ai-configs.ts: For Model Parameters, Ingest Settings, and Prompts

  • ui-configs.ts: For Project Titles, UI Messages, Placeholders, and More

  • metadata.ts: For the Next.js App's Metadata

  • env.ts: Uses zod to validate env variables. Not needed, just used for basic type/error checking.

📋 Commands

All commands are run from the root of the project, from a terminal:

Command Action
pnpm i Installs dependencies
pnpm ingest Chucks docs in the public/docs/ folder and uploads vectors to Pinecone
pnpm dev Starts the local development server at localhost:3000

🚸 Roadmap

  • Add more Langchain file loaders
  • Improve metadata typing, error handling, missing data fallbacks
  • Write documentation how to edit and use metadata from different file types in the public/docs folder
  • Allow users to upload files directly from the web interface
  • Add an "About" Page
  • Add a "Prompting Tips" Page
  • Add User Authentication & Multiple Chat Histories/Sessions

💬 Questions?

  • Reach out to me on Twitter: @PaulGrizII
  • Submit a PR or Feature Request

📓 Learn More

About

A quickly customizable retrieval chatbot designed to help you prototype your own application

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • TypeScript 88.0%
  • JavaScript 6.7%
  • CSS 5.3%