ViewPulse
Open Source →

USE CASE

Agentic recommendation engine

YouTube’s default rec stack is tuned to keep you watching. ViewPulse flips the script: a local word map helps you choose what still matters, then the paid workflow searches and verifies current channels and videos for those topics.

Open ViewPulse — upload TakeoutAll use casesSource on GitHub

What “agentic” means here

Not autoplay chains chosen for watch time — a model (or workflow you build around the exported Markdown) that answers a clear prompt: “Given my real mix of channels and title keywords, what should I try next?” You stay the editor: reject, tweak the count, regenerate, or hand the list to a coding agent that books calendar time to actually watch one pick.

Healthier than the feed alone

The feed rewards urgency. A short, explicit list rewards intention. Pair this with opening YouTube from search or subscriptions, muting notifications, or using wellbeing patterns from the rest of ViewPulse if you want fewer accidental sessions. The app does not block YouTube — it gives you another way to decide what deserves your attention.

Privacy-first by design

ViewPulse parses your Google Takeout watch history in the browser. Your export is not uploaded for core analytics. Account AI requires sign-in, uses deployment-managed provider credentials, and runs only when you explicitly trigger a metered request.

Review the open-source codeSelf Degree — ViewPulse