Your meds, supplies, insurance, appointments, labs, records, and family — scattered across ten apps, an inbox, and a spreadsheet today. DailyLifeOS connects them into one coherent system you own. Not a smarter chatbot — a system that quietly keeps you on top of life. Your records never leave home.
90% software · 10% AI · the model is one service, not the star
The center of gravity isn't the model — it's the workflow and the ownership. The thing you feel working for you every morning is the system helping you stay on top of life.
Every card below is generated from your vault, not a mockup: next appointments, today's meds, your A1C trend, supplies running low, what's due for refill, your documents, your emergency contact. No model touches it — it's the deterministic system, made visible.
Generated on your box, served on your box. The footer says it plainly: your records never left this box · N PHI-blind receipts · chain verified.
Most projects start with the technology and go hunting for a problem. We start with the daily realities of living with diabetes — and build only what earns its place.
Real value is mundane: did I take my meds? when's my appointment? what was my last A1C? how many strips are left? Those are data, workflow, and retrieval problems — deterministic software. The model is the friendly natural-language interface, not the thing that knows.
The hardware is commodity and the open-source organs already exist — Nextcloud, Immich, Home Assistant, Paperless, ntfy, calendars, OCR. Nobody has connected them into one coherent diabetic-life experience. That experience is the product. The moat isn't a box.
We build only what you'd notice if it vanished tomorrow — your refill reminder, your insurance documents, your appointment calendar, your family's emergency access, your records. A 10%-smarter chatbot? Nobody would notice. Utility over novelty.
Every layer is real — mapped to code that runs today or weights cooking right now. Honest status on each.
The foundation — rules, schedules, reminders, supply tracking, escalations, audit trails,
notifications, indexing, sync. None of it needs frontier AI.
Live: the reminder/nudge engine on the NAS (cron + ntfy + email), generic-off-box.
Records, insurance, contacts, appointments, labs, imaging, medication history — storage and
retrieval, on your box.
Live: the 15-folder Home Vault with a hard invariant. Next: typed records so
retrieval is a query, not a file hunt.
Purpose-built diabetic models that understand CGM data, insurance terms, wound-care, supply
management — domain knowledge beats parameter count.
Live: DiabeticDaily-4B / 9B / 27B on Hugging Face — run one on your box (setup guide) and DailyLifeOS auto-uses it, loopback-only. Cooking: DiabeticAnchor-27B (clinician-grade foundation).
Train once in the hive, distribute distilled: 27B → 14B → 9B → 4–7B, down to the box in your
home. Most people never need the 27B locally.
Live: the on-box edge brain (organize/summarize/remind; diagnosis refused).
Most users never need a frontier model running locally. They need "find my insurance card" and "remind me to refill insulin." Software problems first — the model is there when language understanding helps.
The 90% is mostly wiring proven tools together and teaching them to speak diabetes — behind the firewall, with receipts, under one conversational shell.
| Daily reality | Proven open-source organ | What DailyLifeOS adds |
|---|---|---|
| Records & documents | Paperless-ngx · Nextcloud | typed diabetic taxonomy · 10-second retrieval |
| Wound tracking | Immich | wound-progression timeline · clinician-share |
| Lab results (A1C) | OCR (tesseract) | structured history + trend |
| Meds & refills | ntfy + reminder engine ✓ live | med schedule · days-of-supply math |
| Appointments | CalDAV · Radicale | visit-prep packets · podiatry cadence |
| Supplies & inventory | Grocy | reorder-before-zero |
| Sensors & automation | Home Assistant | CGM / scale ingest · alerts |
| Family & emergency | thin custom | emergency access · escalation ladder |
A digital life OS holds your most private information. So it has exactly one inviolable rule.
Your vault, your records, your photos, your inference — all on hardware you own. Raw PHI is read in memory, used, and never persisted off-box.
PHI never leaves home.
Cooks the models, runs the heavy compute, receives only non-PHI receipts. It refuses any data that isn't typed open / synthetic / model.
Knows nothing about you.
Enforced in code, both sides: the hive runs assert_non_phi; the edge brain
refuses diagnosis with a hard 422 and never writes PHI. Every action emits a hash-chained,
PHI-blind receipt you can recompute yourself at diabeticledger.com.
The simplest test of a real product: would you notice if it vanished tomorrow?
A chatbot that's 10% smarter. A clever demo. A benchmark score. The novelty layer most AI products ship — and the thing a diabetic least needs.
✕we skip it
Your refill reminder. Your insurance documents. Your appointment calendar. Your family's emergency access. Your A1C history. Your supply count.
✓we build it
Same spirit as the mission itself: give people what they need, not what you have.
DailyLifeOS isn't a standalone app — it's the daily-life layer of the OpenDiabetic ecosystem. The model is one service among many.
A 2-bay NAS, a UGREEN, a ZimaCube, a Jetson — any is enough, because most of the value comes from math, code, storage, and execution. The model just makes the interface friendlier. The hardware is a commodity; the experience is the product.
No vaporware. What runs today, what's framework, what's cooking — stated plainly.
| Layer | Live today | Framework / in build |
|---|---|---|
| ① Math & Code | reminder / nudge engine (NAS + ntfy + email) | supply tracking · escalations · indexing |
| ② Data — vault | 15-folder Home Vault + hard invariant | typed records + indexed retrieval |
| ③ Models | foot-care & appeal edge models (proven) | DiabeticAnchor-27B (cooking) |
| ④ Edge | on-box edge brain (diagnosis refused) | 27B→14B→9B→4B distillation ladder |
First build: the typed vault + retrieval — so "find my insurance card" and "what was my last A1C?" return in ten seconds. Highest disappearance-test value, pure software, no model required. The model becomes the shell once the system underneath is solid.