Lunar Memory: built a feature that remembers what users forget. No AI chat, just data they already have.
Ok so quick backstory first because I think it explains why I built this the way I did.
Lunar is an ADHD task app for women that adjusts based on energy + cycle phase. been building it solo for like 2-3 months now, nights and weekends, day job is as an Engineer.
Few weeks back I pulled my PostHog data and it was rough. 70% of people dropping off mid onboarding.
But the ones who DID stick around kept doing this one thing: Re-adding the same task over and over. Some examples:
- "meal prep" added 4 times, completed once.
- "organize desk" added 6 times, completed zero.
And it hit me that the app had ZERO memory of this. every time someone added "meal prep" again it was a brand new task to the system. no context. no "hey you've tried this before."
Which is actually kind of the whole problem with ADHD tools in general imo. you fail at something, you feel guilty, you try again, you fail again, repeat forever, and the app just watches you do this with a completely straight face like it's not noticing the pattern.
So I started designing something I'm calling Lunar Memory.
First instinct was to make it some chatty AI thing. like "ask your tasks anything!" type deal.
Built a rough version, hated it within a day. for ADHD users specifically, a free text chat interface is just... more decision fatigue. more typing. more thinking about what to ask. exactly the opposite of what the audience needs.
So I scrapped that completely.
What I landed on instead is way simpler and honestly I think more useful:
When you type a task, it fuzzy-matches against your history and shows you straight up — "you added this 4 weeks ago, last
completed Sat 12:14pm, this month 1 of 4
done." no commentary. no AI telling you What to do about it. Just the receipt.
And then a stats tab that groups everything you've ever added and shows completion rate, sortable, so you can actually see what's working and what's a recurring lie you tell yourself every Sunday night.
Zero LLM calls needed for any of this honestly, which surprised me. it's just
string similarity matching + aggregation queries on data the app already collects (every task already has a timestamp, cycle phase, energy level attached). I was overcomplicating it by assuming I needed embeddings or an AI layer. turns out the boring SQL version is actually the better product.
Its not over yet though. The next version (not built yet, but designed) ties completion rate to cycle phase specifically, which is the bit I'm actually most excited about, because nobody else has this data shape. Think of it like:
"you complete deep work tasks 85% of the time in your follicular phase, 12% in luteal" is not something any generic task app can ever tell you, because
they don't track cycle data at all. that's the part where I'd eventually bring in
something smarter, maybe a machine learning model on top of the pattern data, but the v1 doesn't need it and I didn't want to wait to ship something real.
Put up a coming soon page for it if anyone wants to poke at the actual UI mockups:
https://trylunar.in/features/lunar-memory
TBH, I'm genuinely curious what this community thinks though, a few honest questions for you all:
Q - Does "the app remembers your patterns" sound like a real reason to upgrade to paid, or does it feel like a feature nobody asks for until they see it?
Q - If you've built something similar (task history, pattern tracking, anything in that direction) what actually got it used vs what sat there ignored?
Happy to add anyone interested to the early beta list when it ships, just drop a comment or DM, no pressure either way. mostly just wanted to think out loud about this build decision with people who'd actually have an opinion worth hearing.
If you're interested in becoming an early adopter and a pro member, you can try the app at:
https://apps.apple.com/us/app/lunar-adhd-period-planner/id6773340318
