Document Loading

Before you run

Open a terminal in your langchain-course folder. Activate the venv from Project Setup.

langchain-community is already installed. PDF and web loaders need two more packages — install them once:

pip install pypdf beautifulsoup4

The Document object

A loader always returns a list of Document objects. Two fields matter: page_content (the text) and metadata (source path, page number, and similar). See the LangChain docs on Documents if you want the full type definition.

from langchain_core.documents import Document

doc = Document(
    page_content="The <a> tag creates a hyperlink.",
    metadata={"source": "html_basics.txt"},
)

print(doc.page_content)
print(doc.metadata)

Which loader to use

Import from langchain_community.document_loaders. Match the loader to your source type:

Text files

TextLoader reads a .txt file into one document. On Windows, pass encoding="utf-8" if accented characters print as garbage.

from langchain_community.document_loaders import TextLoader

docs = TextLoader("document_loading_samples/html_basics.txt", encoding="utf-8").load()

print(docs[0].page_content)
print(docs[0].metadata)

PDF files

PyPDFLoader returns one document per page. A 10-page file gives you a list of 10 items. It reads embedded text only — scanned photos of pages come back empty unless you run OCR separately.

from langchain_community.document_loaders import PyPDFLoader

docs = PyPDFLoader("document_loading_samples/html_basics.pdf").load()

for page in docs:
    print(page.metadata)   # page number, source path
    print(page.page_content[:120])

CSV files

CSVLoader makes one document per row. The row text looks like tag: a on one line and description: Creates a hyperlink… on the next — column names stay attached to their values.

from langchain_community.document_loaders import CSVLoader

docs = CSVLoader("document_loading_samples/html_tags.csv").load()

for row in docs:
    print(row.page_content)   # one CSV row as text
    print(row.metadata)

Web pages

WebBaseLoader downloads the HTML at a URL and keeps the visible text. You need a working internet connection. We use https://www.google.com in the demo — a real site most students already know.

from langchain_community.document_loaders import WebBaseLoader

docs = WebBaseLoader("https://www.google.com").load()

print(docs[0].page_content[:300])
print(docs[0].metadata)

Run the demo

Download the ZIP below (script and sample files together), unzip into your project folder, then run:

python document_loading_demo.py

If it fails

• FileNotFoundError — the sample folder must sit next to the script. Check the path printed in the error.
• ModuleNotFoundError: pypdf — run pip install pypdf beautifulsoup4.
• Connection error on WebBaseLoader — check your network, or comment out that block and run the file loaders only.
• Blank PDF output — the file is probably image-based. Try the included html_basics.pdf first.

More loaders are listed in the LangChain document loaders docs.

What's Next

You can load files into Document objects. Next up: split long text into smaller chunks.