Describes observable correlations in your own therapy history — for example, whether your leaks rise on nights you sleep on your back. It works only on your data, with your consent.
Where it gets what it says
Exclusively your therapy journal history. It doesn't use other users' data or improvised medical knowledge.
What it refuses to do
- It doesn't diagnose: it describes correlations ("X is observed alongside Y"), never clinical causes.
- It doesn't recommend pressure changes, machine modes, medication, or therapy interruptions.
- With fewer than 7 logged nights, it doesn't opine — better no advice than bad advice.
- Its output passes a post-filter that blocks any drift toward diagnostic language.
Guides you through practical therapy problems — leaks, dryness, aerophagia, marks — with concrete solutions, one at a time.
Where it gets what it says
Only the product's knowledge base, with every entry cited to primary sources (AASM, ERS, SEPAR). If the tree doesn't cover your problem, it says so instead of improvising.
What it refuses to do
- It doesn't answer diagnosis, medication, pressure, or therapy-interruption questions — every message passes a blocked-intents filter, before and after.
- On any clinical signal (chest pain, fainting, sudden worsening), it refers you to your doctor or emergency services instead of continuing the practical tree.
- It doesn't invent remedies: if a solution isn't in the knowledge base, it doesn't exist for it.
Will draft encyclopedia articles from primary sources. It's in preparation: nothing will be published until a licensed medical reviewer is in the loop.
Where it gets what it says
A cited knowledge base (AASM, ERS, SEPAR) and primary bibliography. Every claim must map to a source or gets flagged as unverified.
What it refuses to do
- It cannot publish: it doesn't have that tool. Everything it produces is born as a draft.
- No medical article gets published without an identified medical reviewer, review date, and zero unsourced claims — enforced at the data layer, not in the prompt.
- It doesn't diagnose or recommend treatment within content.
Will measure weekly whether answer engines (ChatGPT, Perplexity, Claude) cite Apnalia when answering questions about apnea and CPAP.
Where it gets what it says
A fixed battery of typical questions per language. It stores the exact fragment where the mention appears, so the judgment is traceable.
What it refuses to do
- It's read-only: it doesn't write to any user or content table.
- It has no access to therapy data or any personal information — it doesn't need it.
Beyond each agent's own limits, all of them share two global rules: no identifiable therapy data ever leaves toward third parties, and on clinical urgency signals (chest pain, fainting, cyanosis) the conversation stops and refers to emergency services.