Papers + PDFs into one cited URL.

The pain

• —You read 30 papers a quarter. By month two you can't remember which one had the argument you need.
• —Your AI assistant can summarize one paper at a time but can't synthesize across the whole stack.
• —Citations go stale the moment you close the tab — title alone doesn't help future-you.
• —Notion / Obsidian work, but they don't deploy to Claude or Cursor when you switch tools.

What you do in Memory.Wiki

1. 1
   Drop each PDF in

   Memory.Wiki extracts the text, runs the optional AI "clean up" pass, and saves it at a permanent URL. Source pages stay intact — you can re-quote with section anchors.
2. 2
   Let the concept index build

   Background Haiku extraction notes the concepts each paper raises. Cross-paper concepts (≥2 papers) get the orange dot. The Related-in-your-hub widget surfaces overlaps you didn't notice.
3. 3
   Bundle by question

   Group the 5-8 papers that bear on one open question into a Bundle. Set the bundle Intent ("Why does X happen when Y is held constant?"). The bundle's discoveries panel surfaces tensions across papers.
4. 4
   Deploy the hub URL

   Paste memory.wiki/hub/<you> into Claude, ChatGPT, or Cursor. It fetches the index + per-concept passages and answers from your actual citations, not training-data hallucinations.

What you get back

• ✓"What did Smith 2023 say about X?" → the AI quotes the passage with the citation.
• ✓Tensions across papers surface automatically — Compile a Brief and the synthesis names which paper disagrees with which.
• ✓When you switch from Cursor to Claude, the context is the URL. Zero re-priming.
• ✓Six months later, future-you can read the hub log and remember why you were investigating each thread.

Worked example

Try it with what's on your desk right now.

No signup. Drop in your first doc and the URL is yours.

Other shapes the URL takes