I save everything. Links, articles, threads, random things that catch my eye. I've done it for years - Pocket, bookmarks, Notes app, messaging myself. Hundreds of links, scattered everywhere, never looked at again.
One day I realised: all those saves tell a story. Not about what I read, but about what I'm drawn to. The patterns in what you collect reveal something about how you see the world. That's taste. And nobody was building a product that helped you understand it.
So I built Trove.
Not a bookmarking app
This is the hardest thing to communicate about Trove. It looks like a bookmarking app. You save links to it. But the product isn't the saving - it's what happens after.
Every link you save gets analysed by AI. Not to summarise it (though it does that too), but to extract signals: themes, entities, tone, intent. Over time, your "Patina" emerges - a taste profile built from your actual behaviour, not from what you say you like.
I spent weeks getting the positioning right. The one-line test for every feature decision: "Does this help the user understand their taste better?" If no, don't build it.
What I cut (and why it matters)
The most important product decisions weren't what I added - they were what I deliberately removed.
RSS feeds - gone. Turns Trove into a feed reader. That's a different product (and I already built that one: it's called CultureTerminal).
Reading goals - gone. The moment you add "you haven't saved this week" notifications, you've turned a reflective tool into a guilt machine. Trove should feel calm, not demanding.
Smart folders - gone. Filing cabinet thinking. The whole point is that patterns emerge from AI analysis, not from you manually organising things.
Ranked feed / trending - gone. "What's popular" is social signal. That's Curio's job, not Trove's.
Each of these features made logical sense in isolation. Together, they'd have turned Trove into Pocket-meets-Feedly-meets-newsletter-app. A nothing product. The discipline was saying no.
The features that stayed
Patina - your taste profile dashboard. Activity heatmap, theme constellation, reading stats, intent breakdown (why you save things), tone breakdown (the voice of what you read). It's a mirror.
Connections - AI finds non-obvious links between your saves. An article about Japanese design connected to one about ad agency culture, because both explore how constraint produces better creative output. That's a Trove connection. Rated by strength: strong, subtle, or surprising.
Briefing - AI reflecting your taste back to you. "This week you've been drawn to [themes]. Here's what that tells us." Not a newsletter. A personal insight.
Taste DNA - a shareable card that visualises your taste profile. Not for social media metrics. For identity. "This is what I'm into."
Curio looks out. Trove looks in.
I've got a sister product called Curio - a social signal aggregator. "Your network is the algorithm." It shows you what your world is talking about. Outward-facing.
Trove faces the opposite direction. It's not about what's trending or what other people care about. It's about you. Your saves, your patterns, your taste.
The hard rule: they can never converge. If a feature is about collective signal, it goes in Curio. If it's about personal pattern, it goes in Trove. If it's both, I pick one or don't build it at all.
The language matters
Trove says "saves" not "bookmarks." "Patina" not "profile." "Themes" not "categories." "Signals" not "data."
This isn't marketing fluff - it's product design. The vocabulary shapes how people think about the tool. "Bookmarks" implies storage and retrieval. "Saves" implies something worth keeping. "Profile" is data. "Patina" is texture, layered over time. Every word choice reinforces what the product actually is.
Save from anywhere
A taste engine is only as good as the signals you feed it. If saving a link feels like work, you won't do it. So I built capture tools for every context.
Chrome extension - one click from your browser. It authenticates with Trove, lets you pick an intent (Read Later, Research, Remember, Act On, Reference) and add tags. The link saves instantly and the AI analysis kicks off in the background. You never leave the page you're on.
iPhone web app - this was the interesting one: iOS doesn't let you build native share sheet extensions without a full App Store submission. So I built a mobile-optimised save page as a Progressive Web App. Add it to your home screen, and when you want to save a link: copy it, tap the Trove icon, one tap to save from your clipboard. The page reads your clipboard, hits the API, and you're done in under two seconds.
Both tools feed the same pipeline: save the URL instantly, then extract content, run AI analysis (themes, entities, tone), and generate embeddings for semantic search - all in the background. The user sees "Saved!" before any of that finishes.
Getting the iPhone flow right took iteration. The Clipboard API is unreliable on iOS, service workers cache aggressively, and PWA manifests fight you on start URLs. The final solution is simple but getting there wasn't. That's the pattern with most product work - the simplest experience requires the most engineering patience.
Trove taught me that product clarity is everything. Not clarity of design (though that helps). Clarity of purpose. When you can articulate what something is in one sentence - "a taste engine that helps you understand what your saves say about you" - every decision becomes easier. Add this feature? Does it serve the sentence? No? Don't build it.
It also taught me that sometimes the best product is the one that does less. In a world of feature-bloated everything-apps, doing one thing deeply is a radical act.
