Tawen
Daily health score, on-device AI, no subscription
Every recovery-score app I tried wanted one of three things: a $30/month membership, a proprietary band I'd have to keep buying, or my sleep and heart-rate data shipped to someone's cloud so they could "personalize" it. I already had the data – my watch writes it to Health Connect every night. I just wanted one honest number in the morning telling me whether to push or rest. Without renting it forever, and without handing my health data to a stranger.
So I built the inverse. It's called Tawen.
It reads the sleep, HRV, and activity already on your phone via Health Connect, computes a readiness score (0–100) on the device, and explains it in plain English – written on the device by Gemini Nano. Nothing leaves the phone. No account, no subscription, no ads. One-time $4.99.
The one decision that shaped the whole app: the score is deterministic; the AI only narrates it.
I could have let the model produce the number. I didn't, for a reason that turned out to matter more than I expected: different versions of Gemini Nano give different output for the same prompt. If the model owned the score, your number could drift between devices and model updates – and you'd have no way to inspect why. So a plain rule-based engine computes the score (same inputs, same number, fully explainable), and Nano writes the explanation about it. The bonus: on phones without Nano, a rule-based explanation takes its place and the score is identical. I only call it "AI" when AI actually wrote it.
(If you want the gory engineering details – the foreground-only constraint, AICore returning BUSY under concurrency, why I keep a deterministic core – I wrote it up here: https://dev.to/d_v_/shipping-gemini-nano-on-device-in-a-real-app-the-constraints-nobody-warns-you-about-2pdi)
The bootstrapper reality: I'm solo. I run a dev studio (Eltexsoft) and have spent years shipping other people's products; Tawen is the first thing I own end to end. There's no venture money, no growth team, and – this is the part I like – no one to sell your data to, because I never built a server to send it to. The pricing follows from that: one-time $4.99, not a subscription. I'd rather earn a place on your status bar once than rent it from you monthly.
That's also the bet I'm least sure about, and where I'd love this community's read: is one-time pricing a mistake in 2026? Everyone says subscriptions are the only sane model for a solo dev. I think there's an audience that's tired of subscriptions and surveillance and just wants the number. I'm about to find out if I'm right.
Where I honestly am: just launched on Google Play (US/Canada for now – health-data compliance is a lot to scope, so I started narrow). Early days, no traction story yet. I'm here for the part before the hockey stick, not after.
What I'd genuinely value feedback on:
The one-time-vs-subscription bet for a privacy-first health app.
Whether "on-device AI" actually lands as a benefit for normal users, or if it's a dev-only flex.
The "alternative to expensive wearable subscriptions" positioning – too niche, or a real wedge?
Happy to answer anything about the build, the on-device Nano work, the Health Connect integration, or the bootstrapping side. Roast freely – that's what I am here for!
Tawen: https://tawen.app
Tawen on Play: https://play.google.com/store/apps/details?id=com.icemint.tawen
5 Comments
The decision that stood out to me wasn't the on-device AI—it was keeping the score deterministic and letting AI explain it instead of generate it. That's a subtle product decision, but it builds trust because users can rely on the number staying consistent over time. I'm curious whether you've considered making that explainability part of the positioning itself, rather than leading with "on-device AI." It feels like trust might be the bigger differentiator than the technology.
You're right, and honestly it took a decent amount of time to actually make that deterministic score true, and aligned with the user's actual overall wellbeing. My fair guess, most won't really care about the internal scoring logic, but they would blame the app if the score doesn't align with how they feel. Thank you for actually diving in, and spotting this – I might describe the scoring logic with more details, as I polish it up in the next releases.
That's exactly the part I found interesting.
I don't think the biggest implication is the scoring system itself.
It's the business decision that quietly follows once users start trusting the score—and I don't think I can unpack that properly in a thread without oversimplifying it.
Happy to explain what I mean if it's useful. What's the best email to reach you on?
Thanks! I'm at hello@tawen(dot)app
Just sent it over by email.
Looking forward to hearing your thoughts once you've had a chance to read it.