Every AI chat exporter ends up rebuilding the same fragile parsers for ChatGPT, Claude, Gemini, Perplexity and others. Web pages aren't much easier. decant-core provides reusable extraction, parsing and detection for AI conversations and regular web content, so developers can build exporters, archivers, web clippers and AI tools without maintaining their own parser stack. 17 AI platforms, web extraction, HTML→Markdown, and a normalised foundation — in one open-source dependency. AGPL-3.0.

I kept seeing the same problem: every AI chat exporter or browser extension ends up maintaining its own collection of fragile parsers. Then ChatGPT changes something. Or Claude. Or Gemini. Or some new AI platform appears. So I extracted the parsing layer from my own projects and made it reusable. decant-core currently handles 17 AI platforms as well as regular web content extraction and HTML → Markdown. The idea is simple: **don't make every developer reverse-engineer the same websites again.** It's AGPL-3.0 because I'd rather improvements to this infrastructure remain available to everyone using it. If you're building an exporter, web clipper, archiver, research tool, or anything else that needs to extract content from modern websites, I'd particularly like to hear what you're currently using for parsing.

decant-core

The extraction layer for AI chats and the web

The extraction layer for AI chats and the web