A replacement, not a restriction.

Tell FeedSip what eats your time.

Pick the apps and categories you lose time to. Optionally grant usage access so FeedSip can start from your real baseline instead of a guess, and optionally choose a few interests to seed the first feed.

Setup runs on your device and can be paused and resumed. A small on-device model is downloaded once during setup; after that, the feed works without a cloud service.

A feed with a floor.

When you open FeedSip you get a real feed you can swipe, with the same full-screen feel you already know. The difference is that it is finite. A session has a beginning, a middle, and an actual end.

It is mixed, not monotonous: short video alongside small active moments. The feed gently tapers toward the end, and when you reach it, that's it. Continuing starts a fresh session — stopping is never treated as a failure.

Not just another place to scroll.

A session mixes passive and active content, so it never becomes a pure novelty treadmill:

Puzzles and facts keep their authoritative content; the on-device model may present them, but can't replace a correct answer or a reviewed fact.

Interesting enough to be a real replacement.

A replacement feed only works if it's worth opening, so FeedSip learns what you actually finish. Ranking favors relevance, novelty, and variety — never raw watch time. A long, passive stare is not a reward here.

You can say more or less directly in the feed. A "less like this" suppresses that topic and creator for a while, and the interests you picked at setup fade as real behavior accumulates. The goal is a better guess, not an engagement trap.

Progress is time back, not time shifted.

Measurement is optional. When enabled, FeedSip compares your distracting app time (social, video, and games) against a frozen seven-day baseline of your own usage.

Crucially, FeedSip keeps its own time as a separate line. If other apps fall but FeedSip rises to replace them, that isn't counted as progress. Only time that leaves both feeds counts. A day above baseline is information, not a score or a failure.

A small model that runs on your phone.

A compact language model proposes how the feed should adapt; deterministic code measures behavior, validates every proposal, schedules the feed, and can always fall back. The model never sees raw logs, captions, or anything it could turn into a behavioral dossier.

The objective is explicit: less passive dependency over time, not more time spent.

What leaves your device.

Very little. Your usage, responses, preferences, and personalization stay local. There is no FeedSip account and no cloud behavioral profile.

FeedSip uses the network only to:

• load videos from public, federated sources;
• download the on-device model once during setup;
• open a source link, and only when you tap it.

Pricing that isn't built to keep you here.

FeedSip is not built on ads or an endless subscription relationship. The intent is simple: an app that helps you use it less should not get paid for keeping you. The exact pricing is still being worked out and will be stated plainly before launch.