MiDosis.

Data flow, step by step

1. Signing up creates the account in Supabase Auth; a database trigger creates the matching profile row.
2. Each new medication is saved to Postgres, where row-level security lets every user read and write only their own rows.
3. At the same time, the app schedules one daily local notification per dose time — reminders don’t need a server.
4. Home builds today’s list from the medications and the dose log; “taken” inserts or updates a row in medication_logs.
5. The chat sends the question and the current medication list to OpenRouter; if three attempts fail, the app answers from its offline set.

Engineering notes

• Reminders without a backendDaily expo-notifications triggers on a max-importance Android channel: they fire offline and with the screen locked.
• Resilient AI callsA 12-second timeout, three attempts with growing pauses (1.5 s, then 3 s) and keyword-based offline answers as a last resort.
• A careful, context-aware promptThe system prompt adds the user’s medications, keeps answers to about 150 words and always says it doesn’t replace a doctor.
• Free model routingThe openrouter/free router picks an available free model, so the assistant costs nothing per question.