Unwindly

Shipped an Apple Watch app. Started April 29, had a working watch app by May 6. Mood logging from your wrist, voice journal entries with on-device transcription, a breathing companion with haptic guidance, and a complication for quick access. The watch talks to the phone over WatchConnectivity — entries sync when they're in range, queue offline when they're not. Took about a week of intense work, then two more weeks of edge cases: extended runtime sessions, workout session handling for breathing, notification dedup between watch and phone.

Breathing exercise with heart rate tracking. The watch breathing exercise registers as an HKWorkoutSession, which means it gets heart rate data from the sensor. When you finish a session, it shows your HR delta — how much your heart rate dropped during the exercise. Seeing "-12 bpm" after a 3-minute breathing session is more convincing than any marketing copy.

iCloud backup. Users kept asking for a safety net. Built a full backup system — journal entries, media attachments, everything — compressed and exported to iCloud. One tap to back up, one tap to restore. Peace of mind without giving up privacy.

Paywall rework. Rewrote the paywall with personalized copy based on what the user selected during onboarding. If you said your goal is "manage anxiety," the paywall speaks to that. If you said "build a journaling habit," different copy. Early numbers suggest it converts better than the generic version, but still testing.

New voice pack. Replaced all the guided audio — breathing cues, body scan narration — with a professional voice pack. The old clips were functional but lifeless. The new ones actually feel calming.

And a first subscription!

Built three things in March that changed the trajectory.

Voice memos in journal entries. The biggest feature request I kept hearing: "I want to journal but I don't want to type." Added voice recording with on-device transcription directly into journal entries. You hit record, talk through your thoughts, and the transcript becomes your journal entry. No audio leaves the device. Took a week to get right — speech recognition edge cases, microphone permissions, making it feel natural inside the existing thought record flow.

iOS 26 Liquid Glass redesign. Apple announced the new design language and I shipped a full redesign in v1.2.0 the same month. Solo dev advantage — no design committee, no sprint planning, just rebuild and ship. The whole journal system got a state machine refactor underneath, which fixed a class of bugs where entries could get into weird half-saved states.

8 blog articles in 3 days. The AI pipeline I built in January paid off. Published articles targeting "worry journal", "CBT worksheets", "mood tracking vs journaling", "digital vs paper journal", "journaling prompts for anxiety", "journaling for stress relief", and more. Each one went through the 4-agent quality gate (SEO writer → editor → CBT consultant → fact checker). This sprint doubled the blog's content and covered most of the remaining high-value keywords I'd identified.

Three months post-launch: the app has voice, a modern design, and real organic search presence. Still solo, still bootstrapped. Still no users 😄

Launched unwindly.app/blog in late January. Now have 46 articles across 4 languages. Here's how.

I'm a developer, not a writer. Writing one blog post used to take me an entire day. For a mental health app competing on SEO, I needed volume without sacrificing quality — especially since CBT content can be genuinely harmful if it's clinically wrong.

So I built a multi-agent pipeline in Claude Code. One slash command — /blog:new "cognitive distortions journal" — kicks off the whole thing:

1. SEO Writer agent drafts the article targeting the keyword, structures it for featured snippets, adds internal cross-links

2. Chief Editor agent scores it on 4 dimensions (interest, readability, composure, flow) — needs 36/40+ to pass

3. CBT Consultant agent checks clinical accuracy — binary GO/NO-GO. This catches things like misexplaining a cognitive distortion or making claims that journaling "cures" anxiety

4. Fact Checker agent verifies every link is live and matches its context

5. Loop until all four pass. Max 10 iterations. Most articles converge in 3-4.

The whole pipeline runs in my terminal. I review the output, make edits where my voice is missing, and publish. Each article takes ~30 minutes of my time instead of a full day.

What actually matters: the CBT Consultant. Mental health misinformation is a real risk with AI-generated content. Having an automated clinical review that blocks publication until accuracy is verified is the part I'm most glad I built. It's caught subtle errors I would have missed — like conflating thought suppression with thought challenging, which sound similar but are opposite approaches.

Stack: Astro blog on Cloudflare Pages, 4 specialized Claude agents, articles in EN/DE/RU/UK. All published to unwindly.app/blog for domain SEO rather than Medium.

Five months after the first commit, Unwindly is live on the App Store.

The last two weeks were chaos. Submitted on December 9. Got feedback from App Store review the next day — fixed it in hours. Then spent a week chasing audio bugs (breathing exercise voice guidance kept breaking on iOS), RevenueCat subscription edge cases, and a bundle size optimization that cut ~20MB.

What shipped:

• Guided thought records (the core — situation → automatic thought → distortion → evidence → balanced thought)

• Mood tracking with emoji picker

• Breathing exercises with voice guidance

• Body scan meditation

• Clinical assessments (PHQ-9, GAD-7, PSS-10, WHO-5)

• On-device ML crisis detection (ONNX model, runs locally, nothing sent to a server)

• Database encryption

• 4 languages (EN, DE, RU, UK)

• Landing page, help center, press kit

• 23 short texts about the main CBT concepts

And no true monetization. I really wanted to release in 2025 and didn't manage to do a business model and monetization, so I just made the app paid (9.99) and prayed.

The "preparing for release" commit was at midnight. Some things don't change.

There is a joke that the first 80% is the hardest, the second 80% are fine. Well, a lot of work has been done to move Unwindly closer to release:

• Onboarding to explain new users the value of the app

• Walkthrough to explain the main features

• Crisis detection. Trusted contacts.

• RevenueCat integration

• Landing page, marketing docs

The most important part - I got a lot of feedback by my friends and that helped a lot to structure and polish what I have.

Not only journalling but also some breathing practices! This one I built for fun - and it's screenshot is still with me!