Life Garden

A gentle, local-first PWA that helps people understand their body and emotions, then choose one tiny, optional act of care.

Comment

Inspiration

Life Garden began with a personal contradiction: I wanted to live well, but after leaving my job and losing the structure of a normal routine, I could not make myself do the things I already knew were healthy.

I did not need another app to tell me to sleep earlier, exercise, eat properly, or stop scrolling. I already knew those answers. What I lacked was the clarity and energy to turn intention into action. When I woke up in the afternoon feeling foggy and believed the day had already been wasted, streaks and productivity scores only made the problem feel like a failure of character.

I first imagined a weight-management product, but gradually realized that weight was only a visible branch. Sleep, energy, physical discomfort, emotion, desire, food, movement, hormones, and life context all grow from the same underlying life.

That became the central metaphor: every part of a person is a different flower, and every person needs a different way of being cared for.

Life Garden asks a different question from most wellness trackers:

What if a person could feel understood before being asked to improve?

What it does

Life Garden is a gentle, local-first progressive web app for observing and caring for yourself.

The core experience begins with a simple question: “What kind of companionship do you need right now?”

A person can:

  • record something quickly without explaining it;
  • slowly clarify a confused state;
  • ask for one concrete “handle” to help them begin;
  • or simply be accompanied without trying to solve anything.

The app then helps the person focus on one current signal, such as sleep, physical discomfort, energy, emotion, or desire.

Before offering advice, Life Garden asks how much capacity the person actually has. Some days may need rest rather than a restart. If the person wants help taking action, the app offers one tiny, concrete, and optional step.

A later follow-up records whether that action helped, changed nothing, did not fit, or was not tried. It never labels the result as success or failure.

Each moment becomes a Footprint. The seven-day Growth view separates recorded facts from cautious possible patterns and possible care strategies. It only speaks when enough information exists, and it never presents correlation as diagnosis.

All personal records stay in IndexedDB on the current device and browser. There is no account, health-data backend, analytics pipeline, or automatic upload. Users can export and restore a versioned JSON backup.

For judging, a clearly labeled demo URL provides fictional in-memory records without reading or changing the visitor’s personal local data.

How we built it

Life Garden was created through one continuous collaboration between a solo humanities-trained creator and Codex using GPT-5.6.

GPT-5.6 participated before any code was written. It helped me transform a broad idea about weight and health into a product philosophy, develop the garden metaphor, distinguish support from pressure, design the low-point decision flow, establish privacy and medical boundaries, and decide how the product should express uncertainty honestly.

Codex translated those decisions into a milestone-based implementation, including:

  • a Next.js and React interface;
  • a TypeScript data model;
  • IndexedDB local persistence;
  • a guided check-in and low-point branching flow;
  • optional tiny actions and follow-ups;
  • the Footprints timeline and seven-day Growth view;
  • backup and restore;
  • service-worker and installable PWA support;
  • mobile safe-area adaptation;
  • and Vercel deployment.

The visual direction also evolved collaboratively. We moved from an early illustrated interface toward a brighter Monet-inspired watercolor garden, combining paper texture, glass-material controls, p5.js ambient light, and restrained flower-themed motion.

During the submission pass, Codex audited the project from a judge’s perspective: repository state, build and lint results, PWA assets, mobile behavior, sensitive information, fresh-user experience, and the complete demonstration loop.

The current product does not call GPT-5.6 at runtime. GPT-5.6 and Codex were the product, design, and engineering collaborators used to create it—not a fictional feature added to make the app appear more AI-powered.

Challenges we ran into

The hardest challenge was not technical. It was preserving the product philosophy inside every interaction.

A “tiny action” can easily become another command. A follow-up can feel like checking homework. A weekly summary can turn into a score. Even a beautiful garden can become a punishment if it withers when the user stops recording.

We repeatedly tested the product against one rule:

It should increase understanding, agency, or care—not shame.

The second challenge was expressing possible patterns responsibly. Health records contain many tempting correlations, but a few repeated entries are not evidence of causation. The Growth view therefore separates recorded facts, cautious observations, and unknown areas. It is intentionally willing to say that there is not enough information.

The third challenge was demonstrating long-term value without violating local-first privacy. A new browser naturally contains no personal records, but an empty garden cannot demonstrate seven-day reflection in three minutes.

We solved this with a safe demo mode. It generates clearly fictional records in memory, resets on refresh, and never writes them into the visitor’s IndexedDB.

Finally, shipping an offline-capable PWA required careful handling of service-worker updates, mobile safe areas, backup warnings, and a server/client timezone mismatch that only appeared after deployment.

Accomplishments that we're proud of

  • Shipped a coherent working product rather than only a visual concept.
  • Built the complete loop from recording a moment to receiving a response, choosing an optional action, returning for a follow-up, and viewing longer-term reflections.
  • Created a low-point flow that distinguishes rest from activation instead of always pushing productivity.
  • Preserved local-first privacy without a health-data backend or hidden upload.
  • Made uncertainty visible instead of generating confident-sounding pseudo-insight.
  • Built an installable mobile PWA with backup and restore.
  • Created a privacy-safe judge demo using fictional data.
  • Maintained one product philosophy across architecture, writing, visual design, interaction, testing, and submission materials.

Most importantly, I used Life Garden during a genuinely difficult period and found that it felt helpful rather than demanding. That was the first and most meaningful validation of the idea.

What we learned

I learned that knowing what is healthy and being able to do it today are completely different problems. A useful product must respond to the second one.

I also learned that gentleness is an engineering constraint, not only a tone of voice. It changes state models, defaults, labels, reminders, reward systems, thresholds for pattern detection, and what the product chooses not to build.

Working with GPT-5.6 and Codex showed me that AI can contribute far beyond generating code. The most valuable work happened when product philosophy, UX writing, architecture, visual design, implementation, and testing remained in the same continuous reasoning context.

As a humanities-trained creator who had only been vibe coding for about two months, I learned that a clear product belief can become a practical engineering compass. I did not need to know every implementation detail before beginning. I needed to understand what the product should protect—and what it must never become.

What's next

Life Garden will remain focused on understanding rather than expanding without limits.

Possible next steps include a gentler first-entry experience and optional additional “flowers,” such as menstrual cycles, food, medication and supplements, movement, and other body signals.

If synchronization is added later, it will remain optional and transparent. Local-first use, exportability, and user control will remain the default.

The long-term goal is not to create the most comprehensive health dashboard. It is to help a person gradually understand:

  • How does my body express fatigue?
  • What genuinely restores me?
  • What repeatedly drains me?
  • What kind of support helps me return?
  • How can I care for myself without turning care into punishment?

Life Garden is not asking people to become more disciplined.

It helps them feel held, understand themselves, and find one small way to care for the life that is already here.

Built With

Share this project:

Updates