future app · harnesses YouTube history idea AI-enhanced
rerun
Your YouTube watch history, filtered to videos you've rewatched 3+ times. The actual obsessions, not the algorithm's guess.
"The 5–10 videos you've watched a dozen times."
The hook
Everyone has those 5–10 videos they've watched a dozen times. rerun finds them in your history and makes the pattern visible — and klippable.
The klip-feed
| Replaces | The YouTube share button |
| Klip payload | Thumbnail + YouTube URL with ?t= timestamp picking the highest-replayed segment |
| Profile signal | "this month's loops" tile — diagnostically honest taste signal |
AI layer
| What it does | Since YouTube doesn't expose per-segment watch counts, an LLM analyzes the video's transcript + comment-section "timestamp comments" to infer the most-rewatched segment, and writes a one-line "why people loop this". |
| Input | Video transcript + top "0:42" timestamp comments + video metadata |
| Output | Suggested ?t= + one-line hook caption |
| Why with > without | Without AI, every klip points at t=0. With AI, the deep-link drops you into the part people actually loop. |
Open questions
- YouTube history access — Google Takeout exports watch history but not in real time. YouTube Data API doesn't expose per-user watch counts. May need a periodic Takeout pull.
- 3-times threshold — arbitrary. Tunable per-user.
Status
idea Filed 2026-05-15 from brainstorm session.
Lino Labs · future-apps · rerun