Chunk any web page for RAG

Extract clean, structured content from any URL — heading-aware chunks, accurate BPE token counts, and source metadata. Built for RAG pipelines, LLM context windows, and SEO research.

  • No account — paste a URL and go
  • Accurate token counts on every chunk
  • Export JSON, Markdown, JSONL, or LangChain

Extract content

One URL at a time
Determines target size for content chunks
Auto
Words of overlap between chunks (prevents context loss)
How content is divided into chunks
Main-content extraction approach
Affects copy + download only; views stay JSON

The chunker fetches and structures the URL you submit. Nothing is stored — results appear only in your browser.

Processing Your URL...

Extracting and structuring content. This may take a few seconds depending on page size.

⚠️ Processing Error

📄 Extracted Content Results

How the chunker works

Every page you submit runs through the same pipeline automatically. No configuration required — though advanced options are there when you want them.

Auto-detection

Analyzes each page to pick the optimal chunk size, overlap, and splitting strategy — heading-based, recursive, or fixed — based on structure and length.

Size-aware chunking

Small (100–200), medium (200–500), or large (500–1000 word) chunks. Recursive splitting keeps every chunk under your limit; smart merging prevents tiny fragments.

Token counts & metadata

Accurate BPE token counts per chunk, heading breadcrumbs, position offsets, and source metadata (JSON-LD, OpenGraph). Export as JSON, Markdown, JSONL, or LangChain.

Frequently asked questions

How the chunker extracts, splits, and exports web content.

Who is Web Content Chunker for?

SEO professionals, content analysts, and RAG builders who need clean, structured content from web pages. It strips navigation, ads, and boilerplate while preserving the content hierarchy — ready for retrieval pipelines, LLM context, competitive research, or migration.

How does the extraction work?

Defuddle isolates the main content (with a cheerio fallback), then the page is split by one of three strategies — heading-based, recursive size-aware, or fixed — chosen automatically for each page. Small chunks merge, oversized ones split, and every chunk gets word and BPE token counts.

What sites can I extract from?

Any publicly accessible page — news articles, blog posts, documentation, product pages. It works best on content-heavy pages with clear heading structure. JavaScript-rendered SPAs and password-protected pages aren't supported.

Is my data secure?

Extraction happens on our servers in real time, and we don't keep the extracted content — results live only in your browser session. Like most web tools, submitted URLs appear in our usage analytics and short-lived server logs, which we use to keep the service running.