This is a small demo project illustrating how to create a chatbot that can query a scraped website. It uses LangChain to manage the chatbot's framework, Gradio for a user friendly interface, OpenAI's gpt-3.5-turbo LLM model, and ChromaDB for as a vector store.

This project accompanies a blog post on my website, and can be read here.

This project supports both pip and pipenv. I recommend using pipenv for the best (and least error prone) experience.

Run

pip install -r requirements.txt

if using pip.

Run

pipenv install

if using pipenv, followed by pipenv shell to start a shell with the installed packages.

You need to create a new .env file from the .env.example file with your OPENAI_API_KEY. You can create one of these on OpenAI's platform. This will require an OpenAI developer account.

To scrape a site, run

python scrape.py --site <site_url> --depth <int>

This will scrape a url and all links found at that url recursively up to the specified depth. This will only scrape sites with the same origin as the given <site_url>, so for example scraping https://python.langchain.com/docs will only scrape sites at https://python.langchain.com.

The data will be stored in a new scrape/ directory.

To generate and persist the embeddings and create a vector store, run

python embed.py

A new persisted vector store will be created in the chroma/ directory.

To launch the chatbot, you can run

python main.py

This will start a Gradio server at http://127.0.0.1:7860, allowing you to chat to the scraped website and data store.

NOTE: you must both first scrape a site and persist a vector store in order for this to work.