Ore uses service providers to operate the app and public website. This page reflects the service categories, AI routing, configured model families, and nutrition providers known to this server as of 2026-10-13. Providers may change as Ore's infrastructure, routing, or features change.
Configured model-family labels describe the requested model; a label does not by itself identify the inference provider that receives a dynamically routed request.
Depending on the requested feature, processed content may include meal text, a voice transcript, selected photos, or prior draft or logged-meal items when the user requests a revision. While a meal draft is being made, the same meal text and selected photos are also sent through OpenRouter to a separate fast model that returns a provisional meal name; that request is limited to inference endpoints that do not use the content for training and do not retain it. For voice logging, the voice recording is sent through OpenRouter to speech-to-text models that turn it into text; those requests are limited to inference endpoints that do not use the content for training and do not retain it, and Ore does not store the recording. Saved-food recognition sends meal text plus the names, aliases and ingredient names in the user’s saved-food library to TypeSafe Jev through OpenRouter; saved nutrition totals, diary history and account identifiers are not part of that decision request. Meal-revision interpretation sends the requested change, current item names and portions, and targeted clarification context to Jev, without account identifiers or Health information. For a consented notification decision, TypeSafe Jev receives computed logging and app-activity aggregates through OpenRouter; no meal text, photos, Apple Health data or account identifiers are included. For a morning Insight, processed content includes the nutrition goal and daily calorie, protein, carbohydrate, fat, and fiber targets; up to the last 28 days of confirmed meals with titles, ingredients, portions, and per-meal nutrition; dietary restrictions entered in onboarding; the headlines and suggested actions of recent Insights; and, when Apple Health is connected, the daily step total and derived Activity Fuel calorie credit for those days — never raw HealthKit samples. The model writes the Insight text, and Ore checks it against fixed content rules before display. For optional Health-informed program creation, separate expanded permission allows structured profile answers and compact summaries from up to 28 completed days to be sent to Google Gemini through OpenRouter. Summaries cover weight and weight-measurement availability, steps, active calories and recording coverage, and workout types, durations and counts. Gemini assesses the complete record and generates one provisional maintenance estimate, calorie target, macro targets and short explanations in a single request. Software validates the returned structure, arithmetic and supported storage limits before presenting it for review. There is no separate AI writer call. Raw HealthKit samples, source identifiers, account identifiers, body-fat estimates and time zone are excluded. Declining uses a standard non-AI estimate, which may have different targets. Existing programs change only after review and acceptance. Ore does not intentionally attach the user's Ore account name, email address, Apple account identifier, or Apple payment records to AI requests. User-provided content may itself contain personal or health information.
See the Privacy Policy and Washington Consumer Health Data Privacy Policy and Nevada Consumer Health Data Privacy Policy for the data categories, purposes, choices, and rights that apply. Questions: support@oreapp.fit.