AI Disclosure

The Vaely App includes features powered by artificial intelligence. This page is the transparency disclosure required by Article 50 of the European Union Artificial Intelligence Act (Regulation (EU) 2024/1689), the assessment of exemption from high-risk classification under Article 6(3) of the same Act, and the related disclosures that follow from honest practice rather than legal obligation.

1. The AI systems used in Vaely

The App employs the following AI components, all running on the device:

• Apple Foundation Models. A 3-billion-parameter on-device language model provided by Apple as part of iOS 26 and later. Used for natural-language extraction (turning a spoken or typed sentence into structured fields), for weekly summaries, for pattern observations, and for the conversational query interface.
• Apple NaturalLanguage / NLEmbedding. Apple's framework for generating sentence embeddings (512-dimensional vectors). Available on every iPhone since iOS 14, fully on-device. Used to find related past entries when the user queries their own data.
• Local retrieval over a SQLite database with sqlite-vec. Vector similarity search across the user's own entries, performed entirely on the device.

No third-party AI service, no remote inference endpoint, no cloud-based language model is invoked by the App. Apple performs the inference on the device's Neural Engine; no request reaches Apple's servers as part of running the model.

2. Roles under the EU AI Act

Under Article 3 of the EU AI Act, the relevant roles for Vaely are:

• Provider: Apple Inc. provides the Foundation Models system. Apple bears the provider obligations under the Act.
• Deployer: Vaely deploys the AI system within the App. Vaely bears the deployer obligations under the Act, including the transparency obligation under Article 50.
• Affected person: You, the user. The Act gives you specific rights when interacting with an AI system.

3. Transparency disclosure under Article 50

Article 50(1) requires that natural persons interacting with an AI system be informed of that interaction unless it is obvious from the context. We make this disclosure here and in the Privacy Policy:

Article 50(2) covers AI-generated synthetic content (audio, image, video, or text). The Vaely App does not generate synthetic media. AI-produced text in the App is always summarization of user-provided content, never independently generated text presented as factual claim. Where the App displays AI-summarized text in a doctor PDF or briefing, the document headers identify the content as a summary of the patient's journal, not as independent generation.

Article 50(3) covers emotion-recognition systems and biometric categorization systems. The App does not use emotion recognition or biometric categorization. The App's "mood" and "energy" fields, where you record your own self-described state, are user-provided values — they are not inferred by AI.

Article 50(4) covers deep fakes. The App does not generate or manipulate deep fakes.

4. Article 6(3) exemption from high-risk classification

Annex III of the EU AI Act enumerates high-risk AI systems. Vaely's AI features do not fall within any Annex III category: not biometric identification, not critical infrastructure, not education or vocational training, not employment management, not access to essential public or private services, not law enforcement, not migration or border control, not administration of justice or democratic processes.

Where an AI system would otherwise fall within Annex III, Article 6(3) provides an exemption when the system performs any of the following:

1. A narrow procedural task.
2. Improvement of the result of a previously completed human activity.
3. Detection of decision-making patterns or deviations from prior decision-making patterns, where the AI is not meant to replace or influence the previously completed human assessment without proper human review.
4. Preparatory tasks for an assessment relevant for Annex III purposes.

Vaely's Article 6(3) exemption assessment

Vaely's AI features satisfy multiple of the Article 6(3) categories simultaneously:

• Narrow procedural task. The natural-language extraction performs a narrow, well-defined transformation: take a sentence, produce a structured set of fields. The model is not asked to perform open-ended reasoning, generate independent claims, or produce content beyond the user's input.
• Improvement of a previously completed human activity. Every AI output in Vaely begins with a journal entry the user has already written or spoken. The AI structures, summarizes, retrieves, or formats that prior human-completed material. The AI does not initiate, replace, or precede the human activity.
• Detection without replacement. Pattern observations identify recurrences in the user's own logged data ("your journal shows you log more joint pain on days following sleep under five hours"). They are presented as descriptions of past entries, not as predictions, not as risk scores, not as recommendations. The user remains the decision-maker; observations support their reflection rather than replacing their judgment.

On these grounds, Vaely's AI features fall within the Article 6(3) exemption. Vaely is not, and Vaely will not become without explicit re-classification by the operator and public update of this disclosure, a high-risk AI system within the meaning of the EU AI Act.

5. Self-imposed limits

Beyond the legal floor, Vaely imposes the following limits on its AI features as a matter of design:

• The system prompt that governs Apple Foundation Models within Vaely instructs the model to frame outputs as summaries of user entries ("your journal shows..."), never as independent analyses ("analysis indicates...").
• The system prompt forbids quantified risk percentages, prospective predictions, dosage recommendations, and clinical terminology that would imply a diagnostic capability the App does not have.
• The confirmation step is mandatory for every natural-language entry. The App does not bypass review even for repeat-pattern entries.
• Safety-keyword detection that may show a "contact emergency services" prompt is rule-based, not AI-generated, to avoid the model deciding that an emergency keyword should be ignored.
• The AI does not write to your journal without your confirmation. The AI does not delete from your journal. The AI does not modify entries you have already saved.

6. Risks and how we mitigate them

Risk	Mitigation
Misextraction of medication name or dose	Confirmation step is mandatory; user reviews every extraction before save; original speech is preserved as a free-text field for cross-reference.
Hallucinated symptom or detail	Same confirmation step; system prompt instructs the model to extract only what was said; out-of-scope content is shown as free-text rather than being structured incorrectly.
Pattern observation drawing a misleading correlation from a small number of entries	Observations are conservative by design; minimum entry counts apply before patterns are surfaced; observations are framed as descriptive of past entries, never as predictive; the user can dismiss any observation.
Model interpreting user input as a request for a diagnosis or treatment recommendation	System prompt forbids diagnostic and prescriptive output; the App's UX does not provide an interface for that mode of interaction.
Dependence on the AI replacing the user's own judgment	Confirmation step keeps the user in the loop on every entry; doctor PDFs include the user's own words alongside the AI's structured summary; the App's anti-guilt design avoids creating a sense of obligation toward AI outputs.

7. Your rights when interacting with an AI system

Where applicable EU law gives you the right to:

• Be informed that you are interacting with an AI system. (This page is that disclosure.)
• Opt out of AI-enabled features. The App's onboarding offers manual structured input as an alternative; you can enable manual-only mode at any time in Settings.
• Receive a meaningful explanation of how an AI output that affects you was produced. For Vaely's outputs, the explanation is summarized in this document and detailed in the on-device AI feature page.
• Request that we update this disclosure if the AI system changes. We commit to updating this page within 30 days of any material change to the AI components, the system prompt, or the bounds described above.

8. Future changes

If Apple updates Foundation Models in a way that changes the AI capabilities meaningfully, we will update this disclosure. If we change Vaely's system prompt, the bounds described in Section 5, or the AI components used, we will update this disclosure with at least 14 days' notice on the website. If we ever introduce an AI feature that would push Vaely outside the Article 6(3) exemption, we will (a) update this disclosure before launch, (b) seek explicit user opt-in, and (c) follow the high-risk AI obligations of the EU AI Act in full.

9. Contact

For any question about Vaely's AI systems, this disclosure, or your rights as an affected person under the EU AI Act, write to tezaapps@gmail.com with the subject line "AI Disclosure".

---

Effective 6 May 2026. Version 1.0.0. This is the first version of this disclosure.