AI Processing Disclosure

Effective Date: August 1, 2026 (Draft)

Last Updated: July 26, 2026 — drafts pending legal review

1. Where AI Is Used in Velora

Velora uses third-party AI models to: estimate nutrition from a meal photo (“meal scanning”); generate or assist with workout plans; and provide coaching-style guidance in the Coach feature. Some server-side utility workloads (such as generation and translation tasks that are not user-facing chat) also use a third-party AI routing service. AI outputs are estimates and general information only — see our Medical Disclaimer.

2. What We Send, and to Whom

Meal scanning. A compressed photo of your meal, any notes you add, and your follow-up answers are sent to our AI meal-scan provider to generate a nutrition estimate.

Coach chat. A scoped, minimized context package is sent to our AI coach provider to generate a response — not your full account or full history. This includes things like a short recent nutrition/workout summary and the specific data needed to answer your question; it does not include your full logs or your entire chat history. A small, separate AI model is used first to classify what you’re asking and decide which limited data categories are relevant, before the main coaching response is generated — this routing step is deterministic about what data it’s allowed to request, and the choice is enforced on our servers, not left to the AI to decide for itself.

Our current AI providers:

• SiliconFlow — our primary AI inference provider, used for meal-scan photo analysis and most coach chat responses.
• A frontier AI provider — used only for our top subscription tier, for certain higher-value coaching requests, within a small daily budget; when that budget is used up for the day, coaching continues seamlessly on our standard model.
• OpenRouter — used for certain server-side generation and translation workloads that are not part of live user chat.

We disclose this information to an AI provider only to perform the specific task (generate your meal estimate, your workout, or your coaching response) — never to build our own AI models, never to sell, and never bundled with your identity (your AI-facing requests generally do not include your name, email, or account ID).

3. What Our AI Providers Are Restricted From Doing

Under our arrangements with these providers, each provider is restricted from: (1) selling or sharing the information we send; (2) using the information for any purpose other than performing the requested task for us — including using your data to train, fine-tune, develop, or improve the provider’s or any third party’s AI models; and (3) combining the information with other sources except as needed to perform the service.

4. The Honest Gap We’re Disclosing

We want to be direct about something rather than overclaim it. A stronger, more specific version of this disclosure would say your data is “deleted the moment processing completes and never retained.” We are not making that claim, because we have verified it is not fully supported by our current providers’ own public standard terms:

• For our primary meal-scan and coach provider, the publicly available terms describe a general “retained as long as necessary” standard, not a specific promise to delete your data immediately after each request.
• For the frontier model used on our top tier, no-training is the standard commercial default, which is a meaningfully strong protection — but standard data retention under that provider’s public commercial terms is measured in days to weeks (with faster, contractual “zero data retention” available as an optional arrangement we have not yet signed).

What this means in practice: your data is used only to generate your result, is not used to train these providers’ models, and is not sold — but it may be retained briefly by a provider under their own standard terms (for example, for abuse detection and security) rather than deleted the instant your response is generated. We consider this an important, ongoing gap to close — either by negotiating stronger contractual data-handling terms with these providers, or by adjusting which providers we use — and we are choosing to disclose it honestly now rather than publish language we can’t back up.

5. Meal-Scan Photos Specifically

Meal-scan photos are handled differently from the rest of your data: they are the one category of image that our AI provider must be able to read as an actual picture (not encrypted ciphertext) to do its job. That photo is stored briefly, in a private, access-controlled location, and is deleted from our systems promptly after your scan result is returned — with an automatic fallback cleanup within about 48 hours in the rare case that the normal delete-on-completion path is interrupted. See the Privacy Policy’s “Photos & Images We Process” section.

6. AI Is Not a Substitute for Professional Advice

AI-generated nutrition estimates, workout suggestions, and coaching responses are informational and educational only. They are not medical, nutritional, psychological, or other professional advice, diagnosis, or treatment. See our Medical Disclaimer.

7. Your Choices

You can stop using any AI-powered feature (meal scanning, workout generation, coaching) at any time simply by not using it — Velora’s on-device features (like manual logging, saved-workout playback, and locally stored Apple Health scores) do not require AI at all. Factual questions Coach can answer directly from your own on-device data (like “what did I eat today”) are answered without calling an AI model.

8. Questions

For questions about how AI is used in Velora, contact [email protected].