Tereré · myAdcow OÜ · Last updated: 23 October 2026
Tereré is a nutrition and endurance-coaching application operated by myAdcow OÜ, registry code 16575961, Sepapaja tn 6, 15551 Tallinn, Estonia (the data controller, "we"). Contact: pavlo.chebotkov@adcow.online.
| Data | Purpose | Legal basis (GDPR) |
|---|---|---|
| Account: on iPhone, your Apple ID identifier, e-mail (or Apple's private relay address) and name; on Android, your Google account identifier, the e-mail of your Google account (only if Google reports it as verified) and the name Google shares with us — never your Google password | Sign-in, account management | Art. 6(1)(b) — contract |
| Health & fitness from Apple Health (iPhone) or Health Connect (Android), once you allow it: sleep (duration, stages, bed and wake times), heart-rate variability, resting heart rate, heart rate during sleep and workouts, daily steps, walking and running distance, VO2 max, weight, and your workouts — type, start time, duration, distance, active energy, elevation gained, indoor or outdoor, the device or app that recorded it, and a summary of your heart rate (average, maximum and time spent in each heart-rate band) — plus workouts you log by hand or by voice. The first import reaches back up to 90 days for workouts and 60 nights for sleep. Workouts from Apple Health and Health Connect arrive without a GPS route. On iPhone, some of what the app reads stays on your phone: the daily totals of cycling distance, flights climbed and active energy are used only for the Apple Health summary shown in the app (the distance and energy of each workout do travel with the workout). On Android, the daily distance is Health Connect's total for the day, whatever the activity (see the Health Connect paragraph below). If you connected a Garmin watch earlier: the same kinds of data, the GPS route of each workout, and the daily figures your watch reports — your average stress for the day and how much body battery you charged and spent | Readiness index, training plan, coaching | Art. 9(2)(a) — your explicit consent, given on a separate in-app screen. Apple Health and Health Connect data leave your phone only after you give that consent. Workouts you log by hand rest on the same Art. 9(2)(a) footing — your explicit consent, given when you enter them (see the Profile row below) |
| Nutrition: meal entries you type, say or photograph — the text, the voice note or the photo you send us — photos of product labels you take to add a product, and the products you save to your own library | Calorie/macros estimation, nutrition guidance | Art. 9(2)(a) — explicit consent |
| Your messages to the in-app coach (text or voice) | Generating coach replies. The reply is written by Anthropic (Claude API) for accounts that have accepted the updated AI consent (version 4, October 2026), and by Google (Gemini API) for all other accounts and as a backup if Claude is unavailable. A voice message is first turned into text by Google; Anthropic receives text only. Google also reads your message and the coach's reply to spot plan changes, suggest follow-up questions, keep the coach's notes and translate when needed, and it turns your message into a search key to find relevant summaries of your history (see §4) | Art. 6(1)(b); health content within — Art. 9(2)(a) |
| What you tell the coach about yourself — in the introduction interview, in chat and in answers to its short check-in questions (for example injuries, illness, pain, sleep, alcohol, your schedule) — kept as short notes the coach remembers, plus a log of which suggested questions you tapped (the template and a hash, not the text of the question) | Letting the coach remember you between conversations, and not offering the same suggested question twice in a week. The notes are part of what the AI providers receive when the coach answers: Google, and Anthropic for chat replies (see §4). The “Yesterday evening” marks (below) are not among these notes and never reach an AI provider | Art. 9(2)(a) — explicit consent |
| Profile you enter: age, sex, height, weight and target weight, your sports and goals (with the date of a target event, if you name one), the days you can train, and the pace of weight change you choose | Calorie and macro targets, your first training plan | Art. 6(1)(b) — contract; weight, target weight and workouts you log by hand are health data and rest on Art. 9(2)(a) — your explicit consent, given when you enter them. They are stored on our server even if you decline AI processing; in that case they are used only by our own calculations and are not sent to any AI provider (Google or Anthropic) |
| Connected-service identifiers and access tokens, for accounts that connected them before the App Store release: Garmin (account e-mail and session tokens, for the owner's own account only — the earlier sign-in through our server was switched off on 22 October 2026, the session tokens of every other account were deleted that day, and we never store a Garmin password), intervals.icu (athlete ID and API key), and, for accounts created through our former Telegram bot, the Telegram account ID and username that are still on record (the bot was switched off on 23 October 2026 and sends nothing) | Importing your workouts and wellness data from the services you connected, and sending planned workouts to your Garmin account when you tap “To Garmin” | Art. 6(1)(b) — contract |
| Approximate home and training location: coordinates rounded to about 1 km, the place name, and derived facts about your surroundings (elevation, whether the coast is nearby, parks, trails, pools, gyms) | Fitting your training plan to real terrain, daylight and weather | Art. 6(1)(b) — contract. Where we infer your training place from your workouts, that inference is derived from health data and keeps the Art. 9(2)(a) protection of the data it comes from |
| Weather hints for your place: for the next few days, short classes derived from the forecast — how warm, whether rain is likely, how windy, how strong the sun — an overall character of the week (cold to hot, dry to wet), and sunrise and sunset, which we calculate ourselves. We do not keep Apple's forecast numbers | Moving a session to the cool hours, planning a wet-weather alternative, fitting training to daylight | Art. 6(1)(b) — contract |
| Start points of your recent activities: coordinates rounded to about 1 km and the date — at most the last 60, with no time, no distance and no activity type | Noticing that you have moved and suggesting the area you usually train in | Art. 6(1)(b) — contract; derived from your workout GPS data, so it keeps the Art. 9(2)(a) protection of the data it comes from |
| Street, neighbourhood and park names of the routes you record: the OpenStreetMap names of the places your GPS track passes through, looked up from track points rounded to about 10 m; kept for as long as your account exists | So the coach can speak about the places you actually run or ride through, instead of a nameless line on a map | Art. 6(1)(b) — contract; derived from your workout GPS data, so it keeps the Art. 9(2)(a) protection of the data it comes from |
| Day-state readings from your Garmin watch: an intraday stress index and a body-battery level (one value per 15 minutes across the day) and the calories of the day's workout, together with the raw daily file your watch sends us, which those readings are computed from | Showing you how charged, how tense and how fresh you are during the day, and letting the coach make the evening session easier when you are worn down | Art. 9(2)(a) — your explicit consent. For the intraday readings and the day view we check that consent before we collect or show anything; the raw daily file arrives with the health-data import above, on the same footing as the rest of your health data |
| How you feel, when you tell us: a daytime self-report — a 1–5 tap or a 0–100 slider on the Today screen | Standing in for watch data when you have no watch connected, so the day view and the coach still have something real to work from | Art. 9(2)(a) — your explicit consent, the same one that covers the day-state readings above; we check it before we record a self-report, and without it nothing is stored |
| “Yesterday evening” marks you choose under the Night card on the Today screen: alcohol (with the number of drinks: 1, 2, or 3 or more), cannabis, sex, a late dinner, caffeine after 3 pm, stress, feeling ill, a flight or a time-zone change, sauna or a hot bath, a screen in bed — or “none of these”. One entry per night, which you can change the same day; marking nothing is always possible | Choosing the Night card's tip, and showing you in the night details how your nights with a mark compare with nights without one (averages of heart-rate variability, sleep time, resting pulse and REM sleep over your last 60 nights). Only our own code uses these marks | Art. 9(2)(a) — your explicit consent, given when you tap a mark; before you tap, the screen says what the marks are for and that they never go to AI. These are special-category data (health and sex life). They are stored on our server whether or not you have given AI consent, and are never sent to any AI provider (Google or Anthropic) — not in coach replies, the coach's notes, summaries or search index, nor for translation |
| Your corrections to the night under the Night card: “lay awake from” and “fell asleep later, from” — a time of night you choose, with an optional note. One entry per night, which you can change or remove | Showing the night as it really was — your watch cannot tell lying awake from light sleep — and recalculating your sleep time and readiness index for that day. Only our own code uses these entries | Art. 9(2)(a) — your explicit consent, given when you save a correction; the screen says beforehand that corrections never go to AI. They are stored on our server whether or not you have given AI consent and are never sent to any AI provider (Google or Anthropic). The corrected sleep time, like any sleep figure, is part of the day summary the coach may receive with your AI consent (§4) |
| For each weight entry: the time it was measured and where it came from (typed in the app, Apple Health or Health Connect, or your watch) | Telling a morning weigh-in apart from an evening one, so your calorie target is computed from morning weights, and not asking you to weigh in again when your scale already reported | Art. 9(2)(a) — explicit consent, given when you enter or sync a weight; stored on our server even if you decline AI processing |
| Subscription status: your plan, when it expires, and the Apple product and transaction identifiers of your subscription — sent to us by the app as a transaction signed by Apple (StoreKit) and by Apple's App Store Server Notifications (renewals, cancellations, refunds). On Android, subscriptions are billed by Google Play: the app sends us the purchase token Google gives it, our server checks the purchase directly with Google (Google Play Developer API), and Google tells us about renewals, cancellations and refunds (real-time developer notifications); we keep the product, the plan (monthly or yearly), the order identifier, the state, the expiry date, whether it is a trial or a test purchase, and the purchase token. Before a purchase our server gives the app a coded account reference — a one-way hash of your account number — which Google stores with the purchase, so we can check that it belongs to your account. We never receive your card details, your Apple ID password or your Google password | Unlocking what you paid for, and handling refunds | Art. 6(1)(b) — contract |
| Records of your consent to AI processing: your account number, your decision, its version and time, a digest of the wording you were shown and the IP address it was given from | Proving that you consented (Art. 7(1) GDPR) | Art. 6(1)(c) — our obligation to be able to demonstrate consent |
| Technical data sent with every request: your app language, your time zone, the app version, the platform (iPhone or Android — we keep the last one seen with your profile) and the kind of build (App Store, TestFlight, Google Play or development) | Showing texts in your language, keeping “today” on your local day, switching features that differ between builds | Art. 6(1)(b) — contract |
| Server logs and AI usage counters: your account number, the kind of request, time and, for each AI call, the model and number of tokens | Security, error fixing, cost control | Art. 6(1)(f) — legitimate interest |
Without that consent there is no day state. The day-state readings and the daytime self-report (the 1–5 tap and the 0–100 slider) rest on the same explicit consent as the rest of your health data, and the app checks it before it acts: until you give that consent we do not build the day state and we do not accept a self-report — the day view stays empty, and neither the intraday readings nor a check-in are stored. Withdraw the consent and both stop. The raw daily file your watch sends us is a separate matter and we would rather say so plainly: it keeps arriving as part of the health-data import you switch on when you connect a watch, whatever you decide about the day state. What the consent controls is whether we compute the day state from it, keep the readings and show them to you. That raw file is deleted on the 90-day clock described in §5 and erased when you delete your account.
Health Connect (Android). On Android the app only reads from Health Connect and writes nothing to it. It asks, in Health Connect's own permission dialog, to read these data types: sleep sessions (with sleep stages), heart rate, resting heart rate, heart-rate variability (RMSSD), steps, distance, weight, exercise sessions (your workouts), active calories burned, elevation gained and VO2 max — and, on Android 14 and later, permission to read data older than 30 days, which the first import needs (up to 90 days of workouts and 60 nights of sleep). The app reads Health Connect only while you are using it (when you open it or refresh a screen), not in the background. Your phone turns the raw readings into the same figures listed in the health row above before anything is sent: per night, sleep time and stages, bed and wake times, a night value of heart-rate variability and your median heart rate during sleep; per day, resting heart rate, steps, total distance (all activities together) and weight with the time it was measured; VO2 max; and per workout, its type, start and end, duration, distance, active calories, elevation gained, indoor or outdoor, the app and device that recorded it and a heart-rate summary (average, maximum, time in each band). The second-by-second heart-rate series and GPS routes are not sent. Nothing leaves your phone until you have given the explicit consent described in the health row, and our server refuses health data from an account without it. On our server, Health Connect data is stored and handled exactly like Apple Health data: same purposes, same retention (§5), same export and deletion (§6).
How we limit the use of Health Connect data. We use data read from Health Connect only to provide the features you see in the app — your readiness and Night card, your training plan, workout analysis and the coach. We do not sell it. We do not use it, or pass it to anyone, for advertising, including personalised or interest-based advertising, and we do not transfer it to advertising platforms, data brokers or information resellers. We do not use it to assess creditworthiness or for lending. We transfer it only as needed to run those features — to our hosting provider (AWS) and, with your explicit consent, to the AI providers named in §4 (Google's Gemini API and Anthropic's Claude API), which act as our processors and do not train their models on it — or where needed for security, to comply with the law, or as part of a merger or acquisition. No person at Tereré reads your Health Connect data unless you ask us to and agree (for example in a support request), where it is needed for security (such as investigating abuse), to comply with the law, or in aggregated, anonymised form for internal operations. You can revoke the app's access at any time in Health Connect → App permissions → Tereré; data already on our server stays until you delete it or your account (§6).
Food photos, product labels and voice notes are not stored on our servers. Photos are processed in memory to estimate the meal (or read the label) and immediately discarded; only the resulting text (dish name, ingredients, calories) is kept — the picture itself stays only on your phone (on iPhone the app saves it to your own photo library if you allow that; on Android it does not save it). A voice note you record is sent to our server and transcribed by Google's Gemini API, and our server deletes the audio as soon as the transcription is done; only the text is kept. On iPhone, if you use dictation in the workout-notes field, your speech is recognised by Apple's speech-recognition service instead, under Apple's privacy policy. The text of your chat messages — never photos or audio — also goes to Anthropic, which writes the coach's replies for accounts on the updated AI consent. See §4 for what Google and Anthropic themselves may keep of the requests they receive.
Diagnostics stay on your phone. Crash and performance reports that iOS hands to the app are kept on your device only (at most five, left out of device backups) and reach us only if you share them yourself from Profile. Separately, if you have allowed your iPhone to share analytics with app developers, Apple may send us crash reports of its own; that is an iOS setting controlled by Apple, not by the app. On Android the app collects no crash reports of its own; if you have allowed your phone to share usage and diagnostics data, Google may show us crash reports in Google Play Console.
Data is provided by you or, with your consent, imported from Apple Health (iPhone) or Health Connect (Android) and — for accounts that connected it earlier — from Garmin. Since 22 October 2026 we are a member of the Garmin Connect Developer Program, and signing in to Garmin through our server with your Garmin e-mail and password is switched off for good. Connecting a Garmin account is not available while we move to Garmin's official connection; when it returns, you will sign in on Garmin's own page, choose there what to share, and Garmin will send that data to our server — we will hold only the access tokens Garmin issues to us, never your password. Accounts connected the earlier way (today, the owner's own account) keep receiving data until they are moved to the new connection. You can withdraw Apple Health access at any time in iOS Settings → Health, and Health Connect access in Health Connect → App permissions → Tereré; to disconnect Garmin, remove Tereré in Garmin Connect → Settings → Connected Apps or write to us, and we delete the stored tokens. The calories and macros the app writes to Apple Health (energy, protein, carbohydrates, fat) stay on your device under Apple's control. The Android app writes nothing to Health Connect.
Location. You enter or confirm your home and training place yourself. We never collect your location in the background: the app asks the device for a single position only when you tap for it. That position is rounded to about 1 km before it leaves your phone for our server, and everything we store, and everything our server looks up from it, uses only the rounded value. There is one exception for that tapped position, described next: the app hands it unrounded to your phone's geocoding service (Apple's on iPhone; on Android, the phone's system geocoder, which on most phones is Google's), because a point moved by a kilometre lands in the water on a coast and comes back with no city name at all.
Turning a position into a place name is not done on your phone alone. When you tap “use my location” or type a city, your device asks Apple's geocoding service (on Android, the phone's system geocoder — Google's on most phones) to resolve it, and our server separately asks Nominatim (operated by the OpenStreetMap Foundation) for the settlement name — for the place you entered yourself as well as for any place we infer from your Garmin watch. Apple (on Android, that geocoding service) receives the exact position your device measured, or the city name you typed; for the places you set, our server sends Nominatim and Overpass only the rounded coordinates. Labelling your recorded routes works differently and is described just below. In no case do we send your account, your name, or anything else about you. When you set your places we neither ask for nor store a street address — the place name we keep is a town or a city.
Weather and terrain. On iPhone, the weather forecast comes from Apple WeatherKit and is fetched on your phone, not by our server. Your phone asks WeatherKit about the place you saved — the same coordinates, rounded to about 1 km, that we already hold for it, not your live position — so weather needs no location permission. The phone keeps the forecast only in memory (for at most an hour) and shows it to you. It also passes the numbers to our server, about once a day at most, where they are turned at once into short hints (“warm”, “rain likely”, “windy”); sunrise and sunset are calculated by us. We store, and pass on to the AI, only those hints — not Apple's numbers. The Android app does not fetch a weather forecast. Elevation and relief around a place you set come from public terrain map tiles (AWS Open Data, Mapzen Terrain Tiles): our server downloads the few public tiles covering the area, each roughly 20–40 km across, depending on latitude; the provider sees only the tile numbers and our server's address, nothing about you.
If you have connected a Garmin watch, we may look at where your recent activities started. The first time, we only suggest the area you usually train in and ask you to confirm it. After that, if your activities keep starting more than 20 km away from your saved training place, we update that training place ourselves and show you a banner telling you so, which you can tap to change it. We never move the place you told us you live in, and we keep no history of your previous training areas.
Recorded routes. When you ask for a detailed analysis of a workout that has a GPS track (today, workouts imported from Garmin), our server compares that track with the routes you have recorded before, so the coach can talk about your usual loop instead of a nameless line. Your route history is kept separately for each account, in a route index on our server, and is only ever compared with your own routes. To label them our server asks Nominatim (operated by the OpenStreetMap Foundation) for the street, neighbourhood and park names along the route — at street level, not just the town. A street cannot be found from a point moved by a kilometre, so here, unlike the places you set, our server sends up to ten points of the workout track itself, rounded to about 10 m — the precision a street name needs, and keeps the names that come back with the route. Nothing about you travels with those points — no account identifier, no name. The answers are also kept in a shared lookup cache: it holds public map names against those same rounded coordinates, it is not linked to your account or to any other, and it is purged on a 12-month cycle. This is built from the GPS track your watch already recorded as part of the workout: it is not a separate location request, and it never uses your phone's current position.
| Recipient | Role | Location |
|---|---|---|
| Amazon Web Services | Hosting and database | EU (Frankfurt, Germany) |
| Google (Gemini API) | AI inference: plan building, workout analysis, food recognition (photos and product labels), transcription of the voice notes you record, coach replies (for accounts that have not accepted the updated AI consent, version 4, and as a backup when Claude is unavailable: after a technical error, a switch-off on our side or a usage limit; the context sent is the one listed in the next row), the summaries and search index the coach keeps about you, and — only with your AI consent — translation of texts the app shows you (such as workout and meal names, coach notes and coach replies) into your app language. For every chat message, whichever service writes the reply, Google also reads your message and the coach's reply to spot plan changes, suggest follow-up questions, keep the coach's notes and translate the reply when needed, and it turns your message into a search key to find relevant summaries of your history. When we build or edit your plan, the description of your place and short weather hints derived from the forecast (not the forecast numbers) are part of the prompt we send, and — where we hold your day state — a short summary of it as well: your current and average intraday stress, your body-battery level, how fresh or overloaded your training load says you are, your sleep debt for the week and your 1–5 check-in. The fifteen-minute series behind those figures is never sent. Under Google's EEA API terms (paid tier) Google does not use this data to train its models; it may keep the requests it receives (including photos and audio) for a limited period solely to detect abuse. | Processing may occur outside the EEA, safeguarded by Google's contractual data-protection commitments (incl. standard contractual clauses) |
| Anthropic Ireland, Limited (Claude API) | AI inference for one purpose only: writing the coach's chat replies, for accounts that have accepted the updated AI consent (version 4, October 2026). Anthropic Ireland, Limited (6th Floor, South Bank House, Barrow Street, Dublin 4, D04 TR29, Ireland) acts as our processor; onward processing is carried out by Anthropic, PBC (United States) and its sub-processors, listed at trust.anthropic.com/subprocessors. What it receives with each question: your message and the last 20 messages of the conversation; your first name (if you gave one), sex, sports, goals, diet, weight and calorie and macro targets, and your cycling power threshold (FTP), if known; which data sources you have connected; today's meals and a summary of your food diary; your weight history; your sleep, heart-rate-variability, resting-heart-rate and step figures; your training history, the plan for the coming days and the day's coach signals; a summary of your whole Garmin history, if you connected Garmin; if you connected intervals.icu, its fitness figures and your latest ride; your day-state readings (charge and tension) where we hold them; the names of the places you set (never coordinates); the notes the coach keeps about you and what you told it in the introduction interview; and short excerpts of the summaries of your history that match your question — text only, never photos or audio. Under Anthropic's commercial terms, Anthropic does not train its models on this data. It deletes the requests within 30 days. It keeps them longer only if the law requires it or if they are flagged for a safety review (a possible breach of Anthropic's usage policy): then for up to 2 years, and the safety-classification scores (not the content) for up to 7 years. If Claude is unavailable, the reply is written by Google instead (see the Google row). | Outside the EEA, possibly in several countries including the United States; the requests are stored in the United States for the retention period. Safeguard: standard contractual clauses in Anthropic's data-processing addendum. Anthropic is not certified under the EU–US Data Privacy Framework |
| Apple | Sign in with Apple; App Store billing and subscriptions (we receive signed transactions and App Store Server Notifications); geocoding — turning the coordinates you tap for, or the city name you type, into a place name; the weather forecast for the place you saved (Apple WeatherKit): your phone sends Apple the same rounded coordinates, about 1 km — not your live position; the map picture of a workout's area (Apple Maps), which your phone requests when it draws the workout map; and, if you use dictation, speech recognition. All of this runs over the network against Apple's own services, under Apple's privacy policy. | — |
| Google (Android services) | Sign in with Google (we receive a token signed by Google with your Google account identifier, your verified e-mail and your name); Google Play billing (our server checks your purchases with the Google Play Developer API and receives Google's real-time notifications about your subscription through Google Cloud Pub/Sub); and, on Android, the phone's system geocoder, which on most phones is Google's. Health Connect itself is a store on your phone: reading from it sends nothing to Google. These run against Google's own services, under Google's privacy policy. | — |
| Cloudflare | Network: the app's requests reach our server through Cloudflare (Cloudflare Tunnel), so Cloudflare handles them in transit, including your IP address | Global network |
| Amazon Web Services (AWS Open Data, Terrain Tiles) | Public elevation map tiles around a place you set: our server downloads public tiles, each roughly 20–40 km across depending on latitude; AWS sees only the tile numbers and our server's address, nothing about you. | United States / EU |
| OpenStreetMap Foundation (Nominatim) | Place names: the settlement name for the places you set, and the street, neighbourhood and park names along the routes you record, looked up at street level. For the places you set, only the rounded coordinates are sent, together with your app display language so place names come back in the language you read. For route labelling, up to ten points of the workout track are sent, rounded to about 10 m — the same grid the shared cache is keyed on — with a fixed language preference (Russian, Spanish, English) instead of your app language. Neither request carries an account identifier, your name, or any other data about you. | United Kingdom (operator) |
| FOSSGIS e.V. (Overpass API); private.coffee (backup Overpass server) | Map features near a location (parks, trails, water, pools, gyms). Only the rounded coordinates are sent, with no account identifier, name or other data about you. The backup server is used only when the main one does not answer. | Germany (FOSSGIS); private.coffee does not state where it hosts |
AWS, Google, Anthropic and Cloudflare process data for us under data-processing agreements; the other recipients act under their own terms. Where data goes outside the EEA (Google, Apple, Cloudflare, Amazon in the United States) we rely on the recipient's EU–US Data Privacy Framework certification or on standard contractual clauses. Anthropic processes AI requests outside the EEA, possibly in several countries including the United States; it is not certified under the EU–US Data Privacy Framework, so for Anthropic we rely on the standard contractual clauses in its data-processing addendum. The United Kingdom (OpenStreetMap Foundation) has an EU adequacy decision.
Services you have connected (Garmin, and intervals.icu for a few early accounts) send us the data you chose to share; Garmin also receives the planned workouts you choose to send to your watch. They act as independent controllers under their own privacy policies. Our former Telegram bot was switched off on 23 October 2026: it neither accepts nor sends messages, and nothing is sent to Telegram any more.
We do not sell your data, do not share it with advertisers, and do not use it for advertising. The app contains no advertising or analytics software from third parties. AI processing — including the transcription of voice notes and the translation of texts into your app language — happens only after you explicitly allow it on the in-app consent screen, which names both recipients — Google's Gemini API and Anthropic's Claude API — and says which of them does what; §2 and §4 list every category. The “Yesterday evening” marks (alcohol, cannabis, sex and the other marks under the Night card) are never sent to Google, Anthropic or any other AI service — only our own code uses them. If you use an older version of the app, the coach's chat replies stay with Google. In the new version the coach first asks for the updated consent (version 4, October 2026); if you decline it, AI processing is switched off altogether. You can revoke that consent in Profile at any time, and the app keeps working without AI features.
If you have connected a Garmin account, we receive the data you allowed: daily health figures (sleep, heart rate, heart-rate variability, stress, body battery, steps), your activities with their GPS route, and your training data. We never see or keep your Garmin password.
What the AI does with it. With your explicit AI consent, figures derived from your Garmin data are part of the request our AI providers receive when they build your plan, analyse a workout or write a coach reply: Google (Gemini API) and Anthropic (Claude API), as described above. They act as our processors and do not train their models on it. We do not train any AI model on Garmin data, we do not build summaries, a search index or any other AI knowledge base from it, and we do not sell it or use it for advertising. It is stored by our hosting provider (AWS, Frankfurt) and passes through Cloudflare on its way to the app.
Your choice. Garmin data reaches an AI provider only after you allow AI processing on the in-app consent screen. You can withdraw that consent in Profile at any time; from then on no Garmin data is sent to any AI provider, and the app keeps showing your figures without AI. To stop the data flow altogether, disconnect Garmin (§3). Profile → “Export my data” gives you a copy, and “Delete account” erases everything we hold (§6).
Garmin and the Garmin logo are trademarks of Garmin Ltd. or its subsidiaries; Tereré is not affiliated with or endorsed by Garmin.
Credits: weather data from Apple Weather (other data sources); map data © OpenStreetMap contributors (openstreetmap.org/copyright); terrain data from Mapzen Terrain Tiles (USGS, Copernicus, NOAA and others).
Under the GDPR (and, for users in Argentina, Ley 25.326) you can access, correct, export and delete your data:
You may also ask us to restrict processing, object to it, and withdraw any consent at any time without affecting processing done before (AI consent: Profile; Apple Health: iOS Settings → Health; Health Connect: Health Connect → App permissions → Tereré). Tereré takes no decisions with legal or similarly significant effects on you: the readiness index and training suggestions are automated but only advise.
You may lodge a complaint with the Estonian Data Protection Inspectorate (AKI, aki.ee) or your local authority — in Argentina, the Agencia de Acceso a la Información Pública (AAIP).
Data is stored in the EU with encryption in transit (TLS). AI requests may be processed outside the EEA (see §4). Access is limited to the service itself and protected by authentication tokens.
Tereré is not intended for persons under 16. The app asks for your age when you set up your profile and does not accept an age under 16. We do not knowingly process children's data.
We will post any changes on this page and, for material changes affecting your consent, ask you again in the app.