future app · harnesses X bookmarks idea AI-enhanced
scroll
Your X bookmarks, auto-clustered by theme. The thread you saved at 2am isn't lost in the heap — scroll ships the cluster, not the single bookmark.
"The 'saving this for later' reflex, finally readable."
The hook
Heavy bookmarkers never re-read their bookmarks. The pile is too big. scroll makes the pile mean something by clustering by topic and surfacing the headline.
The klip-feed
| Replaces | The "saving this for later" reflex (currently a black hole) |
| Klip payload | Cluster headline + 3 representative tweets + their X URLs |
| Profile signal | Topical map of "what I'm chewing on" |
AI layer
| What it does | Embeddings model encodes each bookmark; UMAP/HDBSCAN finds clusters; an LLM names each cluster ("post-AGI labor economics", "espresso machine porn", "your friend's threads about her divorce") and picks the 3 most representative tweets. |
| Input | Full bookmark corpus (text + media) |
| Output | Named clusters + 3-tweet representatives + size + recency |
| Why with > without | Without AI, you have a chronological pile. With AI, you have a topical library that ships ready-to-klip. |
Open questions
- X API — bookmarks endpoint exists but is paywalled (Basic tier). Cost per active user is a real number — model it before MVP.
- X political risk — platform terms shift fast. Build with a portable scraper that could swap in Bluesky / Mastodon bookmarks if needed.
- Cluster naming — naming each cluster is the actual hard part. LLM-driven, user-editable.
Status
idea Filed 2026-05-15 from brainstorm session.
Lino Labs · future-apps · scroll