Trace - F**k The Algorithm. Make Your Own

@tarat Congrats on the launch Tarun. Is there a negative keyword rule or allowed/disallowed times? Like "Keep Justin Bieber away"/categories like politics, sport, trash (cat videos)? allowing for focus vs fun?

@cruise_chen Trace has two agents. The reflection agent that runs every midnight to figure out what you liked and what didn't based on your interactions with the feed and a feed generator agent that runs every morning to curate a new feed daily for you :)

@zeiki_yu Thank you! I am glad you find it useful

@long_wang4 Yes — that’s very much part of the direction.

The goal isn’t to stop at big, public platforms. A lot of the best signal lives in smaller, niche places: industry forums, paid publications, research blogs, even academic sources.

The tricky part is doing this without turning Trace into a giant manual setup. My bias is to add support for deeper and more specialized sources once there’s enough signal to personalize them properly, so they don’t just become another noisy firehose.

So short answer: yes, niche and professional sources are planned. The focus is to bring them in thoughtfully, so they actually raise signal instead of increasing volume.

@chilarai using firebase right now and there is no limit to store the info :)

@andreitudor14 It's both! Thumbs up/down give strong signals, but it also learns from what you scroll past. If you see something but don't interact, those topics get slight decay.

Notes on downvotes get analyzed to understand why you didn't like something and create personal signals. Basically, the more you use it, the better it gets at surfacing signals (not distractions) that matter to you.

@andreitudor14 Glad that resonated 🙂

The feedback loop is mostly implicit by default. Trace pays attention to what you actually do: what you open, what you consistently skip, what you come back to. That signal compounds over time.

There are light explicit controls (like marking something as useful / not useful), but they’re meant as nudges, not constant training. You shouldn’t have to babysit it.

The guiding idea is:
your behavior > your clicks > your settings.

If you read something once, it won’t take over your feed. If you keep engaging with a type of content over time, it slowly earns more space.

The loop is designed to be quiet, gradual, and reversible — so curiosity doesn’t turn into noise.

@vik_sh Thanks — really glad that landed.

On the “what not to show” side, Trace is deliberately conservative. It doesn’t hard-block things just because of one signal. Instead, content that you consistently skip, don’t open, or bounce from quickly just loses priority over time and quietly fades out.

There’s also a bias against repetition and volume. Even if a topic is relevant, it won’t dominate the feed unless your engagement stays steady across days. The idea is to reduce noise by de-emphasizing what isn’t useful, not aggressively pruning the feed and risking blind spots.

So it’s less “you told me no once” and more “you’ve shown me, over time, that this isn’t worth your attention.”

@devonlastapp This is a real concern, and it’s something I’m deliberately trying to avoid.

Trace doesn’t react hard to single interactions. One click or one read won’t suddenly flood your feed with more of the same. Signals are weighted over time, not instantly amplified.

Think of it more like:

• repeated interest → gradual increase

• one-off curiosity → mostly ignored

There’s also an explicit bias toward stability. The feed is damped so it doesn’t swing wildly just because something caught your eye once. If you stop engaging with a topic, it naturally fades without you having to “fix” anything.

The goal is that you never feel punished for being curious. You shouldn’t have to babysit the algorithm or undo accidental clicks. If that ever starts happening, that’s a bug, not expected behavior.

@mykyta_semenov_ Great question — and you’re basically describing where this can go long-term.

Right now, Trace is opinionated on purpose: fewer knobs, more learning from what you actually read. I’m trying to avoid turning it into something you have to constantly configure to get value.

That said:

• Source control, topic depth, and format preferences are all things I’m thinking about, especially once there’s enough signal to make them useful instead of overwhelming.

• Audio summaries / alternative formats are very much on the radar, but only if they stay “open → get value → leave,” not another thing demanding attention.

• Delivery to other surfaces (email, etc.) makes sense for some people, though I want to be careful about not recreating newsletter overload in a new form.

So yes — the vision overlaps with a personalized news aggregator, but biased toward less setup and less noise rather than maximum control from day one.

If you’re open to it, I’d love to hear which part of that matters most to you first:
source control, formats (audio/video), or delivery outside the app.