Documents, Loaders and Text Splitters

Document = text + metadata

A LangChain Document holds page_content (text) and metadata (source, page, date, permissions). Loaders read files, web pages, databases and cloud drives into Documents. A text splitter divides long documents into chunks sized for embedding and prompting. RecursiveCharacterTextSplitter tries separators in order (paragraph breaks, line breaks, spaces) so chunks end at natural boundaries, with chunk_size and chunk_overlap in characters (or tokens, if you provide a token-based length function). Test sizes on your own questions; very small chunks lose context and very large ones blur topics.

Split, embed, retrieve, answer

Documents become chunks, chunks become vectors, and retrieved chunks go into the prompt.

Splitting with overlap, run

I ran this offline in a Python virtual environment with langchain-core 1.6.6, langchain-text-splitters 1.1.2 and llama-index-core 0.14.25. No API key or network call is needed because a fake model or a toy embedding stands in for the real one. With chunk_size=90 and chunk_overlap=20, the text becomes 4 chunks. The second chunk is cut at 86 characters and the third ("keep context across boundaries.") starts with words repeated from the end of the second: that is the overlap.

from langchain_text_splitters import RecursiveCharacterTextSplitter

text = ("LangChain splits long documents into chunks.\n\n"
        "Each chunk keeps some overlap with the previous one. "
        "This helps retrieval keep context across boundaries.\n\n"
        "Smaller chunks are more precise; larger chunks carry more context.")
splitter = RecursiveCharacterTextSplitter(chunk_size=90, chunk_overlap=20)
chunks = splitter.split_text(text)
for i, c in enumerate(chunks):
    print(i, len(c), repr(c))

Output:

0 44 'LangChain splits long documents into chunks.'
1 86 'Each chunk keeps some overlap with the previous one. This helps retrieval keep context'
2 31 'keep context across boundaries.'
3 66 'Smaller chunks are more precise; larger chunks carry more context.'

Quick check: What is chunk overlap for?

• Keeping context across chunk boundaries
• Doubling the cost
• Encrypting text
• Deleting duplicates