Defuddle extracts web content for LLMs
Defuddle is a lightweight TypeScript library and CLI tool built by Steph Ango to extract article content from the web while effectively stripping out ads, navigation, and other layout noise. Specifically engineered to preserve complex formatting like code blocks, LaTeX math, and footnotes, it is designed as a modern successor to legacy tools like Mozilla Readability, making it highly suitable for RAG pipelines, web clipping extensions, and LLM-based applications where clean, structured context is essential to minimize token waste.
Defuddle hits a sweet spot for modern AI development by solving the "junk extraction" problem with high precision for structured content.
* The "Why": Standard readability tools often fail at modern web constructs (callouts, math, code blocks), making them poor choices for feeding LLMs.
* The "So What": By producing clean, semantic Markdown/HTML, it dramatically improves data quality for RAG and context-window-constrained applications.
* The "Take": It's a pragmatic, developer-focused utility that fits perfectly into the current trend of local-first tooling and optimized AI data pipelines.
DISCOVERED
2h ago
2026-11-17
PUBLISHED
2h ago
2026-11-17
RELEVANCE
AUTHOR
GithubProjects