Founder's Notes & FAQs

Simpl.day is a private, local-first app that holds a whole life in one place — tasks, notes, goals, life areas, people, mood, health logs, faith practice — and an AI Companion that can talk about all of it without your life leaving your phone. The one-line version I keep coming back to:

A trellis for a human life — it carries the small things, so you can go and live the large ones.

Everything is organized under six pillars of a flourishing life — Body, Stability, Belonging, Growth, Purpose, Spirit — so the app has one lens for balance rather than a dozen disconnected lists. Your data lives on your device in app-private storage plus a backup file you own; no account and no cloud are required to use it. The name is the thesis: a simple day — the ordinary one, not the highlight reel — is the thing worth protecting.

The most consistent finding across decades of research on wellbeing is that the quality of your close relationships is the best predictor of health and happiness. Meanwhile, most of the software built for "wellness" runs on the attention economy's one move: the longer you stay on the screen, the better the company did. The harm is displacement — screens quietly crowding out sleep, focus, presence, and embodied human contact.

Simpl.day is built against that. Its governing image is the trellis: the app is not the plant, it is the support structure — and the better it works, the more it disappears beneath the growth. Its north-star metric is deliberately inverted:

Time-to-Human ÷ Time-on-Glass — real human contact created, divided by the screen time it cost. The dream is for that ratio to be enormous, and for you to need the app a little less every month.

A feature that increases screen time is treated as a regression here, even if people would enjoy it. Every path in the app is meant to end in a human, a place, or a piece of paper — never another screen. The app's proudest moments are the ones where it sends you off itself.

• The ordinary person reaching for more — not in crisis, just aware their days are being eaten; someone who has tried productivity and wellness apps and come away a little emptier. This is the heart of it.
• The person in a hard season — grief, burnout, recovery. They need a safety floor that doesn't depend on luck, one-tap reach to a real person, and a clean handoff to actual help — never a chatbot pretending to be that help.
• The seeker — someone after meaning, not just order: reflection, examen, a rule of life, faith practice woven into the same place their tasks live.
• The relationally hungry — people on the wrong side of the loneliness epidemic. The app's deepest work is not to be their company but to help them rebuild a real human network.
• The practitioner — therapists, spiritual directors, coaches who might one day use it alongside the people they serve. A first-class user in the long-term model.

By default the app names no religion — the seeded life areas work for any believer or any secular user — and a faith tradition is a setting you choose, never an assumption.

One connected model, on your device

The spine is Pillar → Area: goals, tasks, notes, people, and logs all hang off life areas, so everything is connected in one graph instead of siloed features. All of it lives in on-device storage plus a plain JSON backup file in your Documents folder that survives reinstalls — a file you can read, move, and own.

An AI Companion that asks before it speaks to the internet

The Companion has selectable engines: a deterministic no-AI mode (rules and retrievers, no network), an offline on-device model, a hosted mode run by Simpl.day, or your own API key. A fresh install always starts in the no-AI mode — there is no silent online default. The first time you message the Companion on a build that offers the hosted mode, your message is held, unsent, until you choose. Every online call is logged in an on-device audit log — provider, model, token counts, never content — that you can read and clear in Settings.

Deterministic first, model second

Every turn is attempted deterministically before any model is invoked: crisis phrases match first and return real hotline information with no AI call at all; reads of your own data are answered by a keyword router that always wins over the model; and destructive operations cannot be triggered by AI output — nothing mutates until you tap a confirmation chip. Safety here is designed to be predictable, or it isn't safety.

Quiet by default

Almost all reminder rules ship off. Notifications sit in three channels so you can mute the app's suggestions while keeping what you asked for. And if you stop tapping a class of nudge for two weeks, the app sends one more and then goes quiet — silence is not consent.

More than a prototype, less than a launch. The codebase is roughly six hundred Dart files with over two thousand automated tests. The major shipped capabilities:

• The Companion — pluggable engines, ~33 deterministic read intents and ~20 confirm-gated write intents, a continuous conversation stream that files things into your real life areas with a correctable receipt, on-device dictation, and a crisis gate that runs before everything else.
• Tasks, areas, goals, notes — 24 task types, routines, dependencies, a day timeline; goals with why-chains and implementation intentions; notes with dictation, sketching, photos, on-device OCR, and semantic search via on-device embeddings.
• People — circles of intimacy, expected rhythms, conversation packs, accountability pacts, and a "night watch" for keeping quiet vigil with someone you trust.
• End-to-end encrypted collaboration — two phones can connect by QR or invite link and share a 1:1 chat, shared tasks, and a wake-together alarm; the server stores only ciphertext and membership, and cannot read any of it.
• Backups — a versioned on-device backup with full export/import, plus opt-in encrypted cloud backup where the file is encrypted on your device before it leaves and only you hold the recovery key.
• Assessments — chronotype, tempo, Big Five, values, strengths, love languages and more, composed into a relational pattern derived fresh on-device and shown with honest confidence bands, never fake-precise percentages.
• Faith — curated Scripture, Catechism, and Summa fact cards; all 150 psalm meditations with on-device narration; morning prayer; a daily verse. All curated and citeable — never model-generated.
• Insights and Home — a bounded (never infinite) life mosaic, a north-star trend card, "one next thing", and a reflect-enough card that sends you off the app.
• An optional Android launcher — a still, monochrome home screen that takes two deliberate steps to enable and never seeds a feed.
• A public feature board — the app's one deliberately world-readable surface, where the most-voted requests steer the roadmap.

The most valuable lessons are the ones that cost something. A few, honestly recorded:

• No silent online default, ever. The AI starts in a no-network mode on every build, and the first message is held unsent until you choose an engine — because asking after sending is not consent.
• The AI cannot destroy anything. No code path routes model output to a delete operation. Every write waits for your tap; dangerous operations live only behind deterministic settings the model cannot reach.
• Crisis support is deterministic. Self-harm and abuse phrases are matched on-device before any model call and answered with real hotlines. No AI, no logging, no cleverness — predictability is the point.
• Keys stay server-side; content is never stored. The hosted AI's key exists only on the server, and neither prompts nor responses are stored anywhere — only a metadata audit row on your own device.
• Sensitive stores are encrypted and kept out of the plaintext backup. Chat transcripts and the app's model of you are encrypted at rest and excluded from the readable backup file by default, with one explicit switch if you prefer completeness.
• Faith content is curated, never generated. Scripture, Catechism, and Summa are citeable, display-only sources; a test pins psalm text verbatim. An AI paraphrasing sacred text is a line this app does not cross.
• Numbers are a design decision. Confidence is never shown as a percentage; relational drift never shows a count ("it's been a while" carries the signal without giving anyone a weapon); procrastination counts are banned from motivation copy. A score derived from what got detected would quietly punish the person who journals honestly.
• Threat models come before features. The relational-memory feature had its abuse threat model written first, and shipped without "what they told you" — a dossier on a third party who never consented — until protective controls exist.
• The beta takes in no money. For legal and personal reasons the app currently has zero monetization anywhere — no ads, subscriptions, or donations. That constraint forced a clearer long-term philosophy: retain the practitioner, release the user. Grow by being trusted, not by being needed.
• Becoming your launcher takes two deliberate steps, ships disabled, and never seeds a feed — the app will not put Instagram one tap from prayer on your behalf.

These are real and intentional caveats, not placeholders:

• Store launch. The app is in private beta; the Play Store path (closed testing with a required tester period, listing assets, policy declarations) is in progress. There is no iOS build yet — Android is the primary platform today.
• Broader shared accountability over the encrypted transport is built underneath but not fully surfaced; check-ins and digests still go out via WhatsApp, SMS, or email hand-off.
• "The loop" — a Context → Reflect → Plan → Act → Close cycle at every scale from a task to a life — is in development and ships dark until it is whole. The self-model's belief pipeline is similarly built but switched off.
• Platform skew. Phone-usage coaching, the session overlay, and the Documents backup are Android-only; iOS will lean on iCloud and local notifications.
• Accessibility — low-vision scaling and labels are done; full screen-reader coverage of charts and custom canvases is partial.
• Crisis phrase matching is English-only today, which blocks a non-English launch.
• The privacy policy has a known correction list pending — the code moved faster than the copy, and the copy is being brought back in line before wider release.

Immediately: finish the launch path — closed testing with real pairs of people (much of what makes this app different needs two humans to exercise), then a slow, personal send-out to the first hundred users, one at a time.

Soon after:

• Raise the share of Companion turns answered deterministically — every such turn is faster, free, and creates no network egress at all.
• Give the offline model its full power, so a phone with no connection loses almost nothing.
• Light up the shared seams of the life graph — the consented, encrypted edges between two people's lives.
• Push toward invisible AI: the best AI features don't look like AI; they look like the app quietly filing, recalling, and phrasing things well.

Longer term: the practitioner partnership model — therapists, directors, and coaches using Simpl.day alongside the people they serve — and a measured version of the ownership question: is the user becoming more capable, or merely more dependent? Roadmap steering stays public: the feature board's most-voted requests are what get built next.

Start with two documents in the repository: the README, which is the canonical map of what the app actually does, and the project's hard rules, which encode the ethics as engineering constraints. Read them before touching code — the rules exist because each one was paid for.

Preserve these, or you've built a different product:

• The consent architecture — no silent online default, the held first message, the audit log, the bounded snapshot.
• Deterministic crisis routing that runs before any model, always.
• No AI-triggered destructive operations; the confirm chip is not optional.
• The data-safety discipline: unique storage type IDs, the backup schema checklist, the safe install path.
• The inverted metric. If a change makes people spend more time on glass and less with humans, it failed, whatever the engagement chart says.

Feel free to rethink: the visual design, the specific feature set, the Flutter/Hive stack, the six-pillar naming, even the AI engines — those are implementations, not principles. The test for any new feature is already written: does this make a human relationship stronger, or does it substitute for one? Strengthen — maybe ship. Substitute — never.

Ideas that would survive this particular app:

• The trellis, not the plant. A tool's success is how invisible it becomes under the growth it supports. A good trellis is invisible by late summer.
• Invert the objective function. Measure real-life outcomes divided by screen time, and treat engagement gains as regressions.
• Make ethics architectural. A privacy promise should be a property of the system — a pure function, a test, a server that never holds the data — not a paragraph on a policy page.
• Deterministic first; the model is the fallback, never the front door. Speed, privacy, and cost all point the same way — and the deterministic path should win every safety arbitration.
• Optimize capability gained minus dependence created. Assistance should shrink as capability grows — training wheels, by design.
• No silent defaults for anything irreversible. Hold the message unsent. Ask before the first send, and mean the answer.
• Numbers are a design decision. Some true numbers should never be shown, because the number is the thing that stings — and the thing an abuser quotes.
• Write the abuse threat model before the feature, and let it veto engineering conveniences.
• Bounded surfaces beat infinite ones. Feeds should end. "You're caught up" is a feature.
• Know your lane and name its edge out loud. This app holds the safety plan your therapist helped you write; it does not do therapy. It walks you to the door of the room where the real thing happens — and waits outside.

If you take one thing from this project, let it be that software can point away from itself. Most of what I'm proud of here is restraint: the notification that was never sent, the number that was never shown, the message that was held until you said yes, the feed that was never built. Those absences took more engineering than most of the features.

I hope people use Simpl.day and then, gradually, need it less — that it carries their small things faithfully enough that they spend the recovered attention on a person, a table, a quiet hour no screen interrupts. And if you're a builder reading this after me: take the ideas. They're the durable part. The code is just the current trellis; the life it's for is yours.

— Clinton, builder of Simpl.day

Want to try it? Join the beta list.