TrainLoose Privacy Policy

Updated October 24, 2026

Your choices

TrainLoose has two optional app-level switches: Apple Health and AI Trainer. Apple Health covers workout upload, automatic sync and workout measurements, including available heart-rate summaries; Apple's single Health dialog asks for the corresponding device reads. Health summaries reach DeepSeek only when both switches are on. AI Trainer asks for DeepSeek consent once, not for every message; a new disclosure or a withdrawal on another device may require confirmation again. Existing installs confirm the combined choices once if their current grants do not already match; previous grants are never silently expanded. Background workout check-ins and workout notifications are off and are not asked at GA. Heart-rate zones, LTHR and personalization remain optional settings, not additional permissions.

Your account and saved data

Sign in with Apple identifies your account. TrainLoose stores your Apple account identifier and encrypted provider credentials, first-party login sessions, and the product data you choose to save: plans, goals, planning context, review decisions and bounded visible conversations. Your Apple password is not uploaded. Optional Health sharing is described below.

Sign-in security / 登录安全

To protect sign-in and prevent abuse, TrainLoose creates a keyed fingerprint (HMAC) from the requester's network address and an app authentication scope. Rate-limit entries store that fingerprint, operation, window, count, expiry and update time; they do not contain raw IP addresses or account/session IDs. These records are pseudonymous, not anonymous. The rate window is ten minutes; entries become eligible for cleanup one hour after it ends. Cleanup currently happens in bounded batches on subsequent authentication requests, so expiry is not a guaranteed physical-deletion deadline and entries may remain longer when authentication is idle. Encrypted database backups may include them under the separate seven-day backup policy. The records are used for sign-in security, not advertising, cross-app advertising tracking or training personalization, and are not included in Trainer requests to DeepSeek. These security checks are separate from optional Apple Health and AI Trainer data sharing.

为保护登录并防止滥用,TrainLoose 根据请求方网络地址和应用认证范围生成带密钥的指纹(HMAC)。限流记录保存该指纹、操作类型、时间窗口、次数、到期及更新时间,不在这些记录中保存原始 IP 地址或账号/会话标识;这些是假名化数据,不是匿名数据。限流窗口为十分钟,窗口结束一小时后具备清理条件。当前仅在后续认证请求发生时按有界批次清理,因此逻辑过期不等于已物理删除,在无认证请求期间可能保留更久。加密数据库备份也可能包含这些记录,遵循独立的七天备份策略。这类记录只用于登录安全,不用于广告、跨 App 广告追踪或训练画像,也不随 Trainer 请求发送给 DeepSeek。这些安全检查与可选的 Apple Health 和 AI Trainer 数据分享相互独立。

Optional cloud AI

Cloud AI is optional and Apple login is not consent. With current permission, TrainLoose sends your submitted text, bounded recent visible conversation, relevant plans, goals, standing planning context and user-written Trainer instructions to DeepSeek for new initial planning, Trainer and plan continuation. Submitted text can contain sensitive information and is not automatically filtered out. Current feelings and intent come first. Only explicit Review applies proposed changes to existing plans or goals. Saving instructions or reviewing goals does not itself call AI; only you edit instructions. You may withdraw in settings; already admitted work may finish, but later model requests need current permission. Withdrawal does not delete saved data or recall provider copies. Existing plans, Review and manual history entry/reads remain available without AI. Ordinary Trainer can read bounded saved manual workout evidence across up to 90 local dates: date, activity, known duration or unknown, coarse source, confidence, incomplete coverage and current numeric session RPE (including zero and decimals). Private feedback, notes, exact workout clocks and source identifiers are excluded unless you explicitly submit text in chat. If asked to record a completed workout, Trainer can save it directly; Keep or Adjust on a separate plan proposal does not undo that save. Nothing writes back to Apple Health. Saving or syncing does not initiate AI; background workout check-ins are off in this GA release. The Apple Health switch covers upload, automatic sync and device measurements together. Health summaries reach DeepSeek only when both Apple Health and AI Trainer are on; the authorized history window still bounds the combined read. Eligible evidence includes workout average/maximum heart rate, up to 24 relative summary intervals, distance and active energy, never raw sensor samples or routes. Analyze this workout explicitly requests analysis of one current saved record, not an automatic sync report or workout/plan change. It uses ordinary Trainer budget/recovery and does not update learned memory. Status checks do not call AI; stale source or permission hides old results. The learned training profile remains an optional personalization setting, not an additional permission prompt. Relevant unexpired observations can then reach DeepSeek. Ordinary Trainer may propose a small dated memory supported by your explicit statements or eligible workout patterns; it is saved only after a successful reply and source/privacy checks, with no extra background model call. Pause stops both learning and profile use; clear removes content without pausing future learning. This is not medical assessment, personality inference or compliance scoring. Health-derived memory requires current task-specific Health permission and is excluded from initial/renewal input. Background workout check-ins and workout notifications are off and are not offered in this GA release. Sync does not automatically call DeepSeek or change a plan. TrainLoose deterministically estimates training load from qualified active heart-rate zone time first, otherwise explicit RPE and known duration, or unknown. Original RPE and RPE times minutes (AU) remain separate, never added to these estimates. Qualified HR needs an explicitly configured sport LTHR, six increasing boundaries for seven zones, matching active duration and at least 90% observed active coverage. Pauses, conflicting overlaps and gaps over 30 seconds are excluded; missing time is not scaled up. Age, maximum HR and coarse averages do not create a qualified threshold. Ordinary Trainer and explicit selected analysis may receive current estimate, method, coverage and limits; Health needs linked Health-AI consent. Raw samples and zone configuration are not sent. Ordinary Trainer can also read application-calculated daily CTL/ATL/opening TSB recorded contributions within the permitted window, not total fitness, recovery or diagnosis. Unrecorded exercise and initial history remain unknown; decay does not prove rest and positive TSB does not authorize harder training. These are not official TrainingPeaks TSS or guaranteed multisport equivalents. No diary, threshold or complete history is required. This does not expand background inputs, analyze every sync or change plans. If you separately enter optional About you details (reported age with its confirmation date, sex, height in cm or weight in kg), new user-requested initial planning, Trainer and plan continuation tasks may send a minimized snapshot to DeepSeek under this current AI permission. Saving or clearing these fields does not require AI or Apple Health permission. Missing values and prefer-not-to-say sex remain unknown. An age not confirmed within 366 calendar dates is marked stale and its number is not sent; age never advances automatically. Details are not inferred from Health, messages or learned memory and do not authorize background demographic analysis. Your current feelings and intent take priority. These details do not enable BMI, calorie targets, ideal weight, diagnosis or automatic heart-rate thresholds. Existing AI grants do not authorize this expanded input; a new explicit confirmation is required.

Retention and deletion

Saved account data remains while the account is active, with bounded conversational context, not a permanent full transcript. Explicit account deletion removes online data and sessions. Encrypted rolling backups expire after seven days; erasures are reapplied before restored service. Separate minimal account/source identity and revision fences prevent resurrection; these are linkable private metadata, not anonymous data, and do not contain removed values, raw HealthKit IDs, messages, workout facts or plans. Provider processing follows linked DeepSeek Open Platform terms; no zero retention, no training or provider erasure is promised. Contact shuhengzhao@qq.com for access/deletion questions. Complete export and account deletion cover manual/Health workouts, optional feedback, RPE, statistics, saved zone/configuration evidence, source-bound analyses, check-in questions/choices, instructions, goals and readable memory. Historical save receipts support recovery but never recreate deleted records. Missing data is unknown, not zero. Instructions remain until replaced, cleared or account deletion. Instruction/goal receipt text expires after seven days; content-free account recovery fences remain. Learned experience/workout patterns expire within 42 days, other supported preferences within 180 days, without background refresh. Forget/clear prevents restoration by old tasks or evidence; disaster restore clears stale memory and input copies. Source correction/deletion or RPE clearing scrubs affected derived questions. Restore disables automatic/background choices and clears questions; switching analysis off alone does not erase retained text. Source/RPE/statistics/zone or AI/Health permission changes invalidate and scrub affected selected-analysis results. Execution evidence clears on completion/Stop; ordinary chat and receipts retain an acknowledgement, not analysis text. Regrant and old backups cannot revive invalidated results; export is historical, not authority for current display. Estimates and daily curves are recomputed, not a second stored workout history. Saved records/estimates remain readable without AI. Local settings alone do not erase cloud data or rewrite pending uploads. Controls cannot recall exported or provider copies. Saved details are independent account settings in complete export and account deletion. Unfinished model tasks retain a private first-admission copy for at most 24 hours; ordinary edits apply to new tasks, not retries. Clearing any field, choosing not to disclose sex, clearing all details or changing AI permission invalidates affected task copies before further model access. Completion, Stop, expiry and disaster restore remove these private copies. A separate restore journal retains only account/request identity, revision, removed-field names and time, not personal values, to prevent resurrection. Clearing settings cannot recall data already sent to the provider, retained conversation text or exported copies.

Access and control

In You → Privacy you can withdraw AI permission or export your saved account product data. Account deletion removes online account product records and sessions; deletion protection is applied before restored service. Copies you save or share outside the app remain under your control. Local Health access is managed on your device separately and does not grant cloud upload permission.

We do not write message bodies, Health payloads, credentials or model reasoning to application request logs. Limited operational status and deletion safeguards are not a conversation transcript or a full data export.

DeepSeek Open Platform terms · Contact: shuhengzhao@qq.com

Web-only: Apple Health sync identifiers / 仅网页补充:Apple Health 同步标识

Apple Health uploads also include a random sync-stream identifier stored for this TrainLoose account on this app installation, together with sequence and request identifiers, so the server can deduplicate and recover ordered uploads; it is not a hardware or advertising identifier.

Apple Health 上传还会包含为本应用安装中的当前 TrainLoose 账号保存的随机同步流标识,以及顺序号和请求标识,用于服务器去重和恢复有序上传;它不是硬件标识或广告标识。

Web-only: Account operations / 仅网页补充:账号运行记录

To operate and protect the service and recover requests, TrainLoose saves account-linked operational metadata such as request type and time, request and review status, call counts and estimated reserved resource usage; these are not purchase records or a daily chat allowance.

为运行和保护服务及恢复请求,TrainLoose 会保存关联账号的运行元数据,例如请求类型和时间、请求与审核状态、调用次数及预留资源用量估算;这些不是购买记录,也不表示启用了每日聊天额度。

Web-only: Personalized training suggestions / 仅网页补充:个性化训练建议

With the current AI permission and, for Health-derived information, the current Apple Health sharing permission, we use your saved goals, Instructions, eligible preferences and any optional details or permitted workout summaries to customize training suggestions; your current feelings and intent still take priority.

在当前 AI 授权范围内,并在使用 Health 衍生信息时同时满足当前 Apple Health 分享授权,我们会使用已保存的目标、Instructions、适用偏好,以及你选择提供的个人资料或获准使用的运动摘要,定制训练建议;你的当下感受与意愿仍优先。

Web-only: Trainer failure diagnostics / 仅网页补充:Trainer 失败诊断

If a Trainer request fails, we may retain a bounded diagnostic record containing a support reference, failure category and stage, model and tool counts, and closed tool-result codes; the reference is derived from account and request identifiers, so this is linkable metadata, not anonymous data, and it contains no chat, workout, Health or provider-response content. Operational logs rotate by size; the seven-day backup policy is not a seven-day log-retention promise.

Trainer 请求失败时,我们可能保留有界的诊断记录,包括支持查询标识、失败类别与阶段、模型与工具计数,以及限定的工具结果代码。该查询标识由账号和请求标识派生,因此这些是可关联的元数据,不是匿名数据;记录不包含聊天、运动、Health 或模型服务商回复正文。运行日志按大小轮换,七天备份政策不表示日志也承诺七天删除。

Web-only: App Store subscriptions / 仅网页补充:App Store 订阅

Apple processes TrainLoose Plus purchases, renewals and payment details. The app checks verified subscription status on your device with StoreKit and provides an opaque account-derived identifier to Apple as purchase metadata. In this release, purchase transactions are not uploaded to TrainLoose servers, and TrainLoose does not receive your card details. Buying Plus does not grant AI or Apple Health permission. Permission changes, export and account deletion do not require Plus. Deleting your TrainLoose account does not cancel your Apple subscription; manage cancellation, restoration and refund requests through Apple.

Apple 处理 TrainLoose Plus 的购买、续订和付款信息。应用通过 StoreKit 在设备上检查已验证的订阅状态,并向 Apple 提供由账号派生的不透明标识作为购买元数据。本版本不向 TrainLoose 服务器上传购买交易,TrainLoose 不接收银行卡详情。购买 Plus 不代表同意 AI 或 Apple Health 分享;修改权限、导出和删除账号不要求 Plus。删除 TrainLoose 账号不会取消 Apple 订阅;取消、恢复购买和退款请求通过 Apple 管理。

Account deletion

Use Delete account in the app to request deletion of your TrainLoose account. A confirmed successful deletion removes online account data and sessions; the app then clears local account data and signs out. If the deletion request refuses account access, the app clears local account data and signs out without claiming deletion succeeded. For any other unconfirmed deletion result, including a server error, timeout, offline request or invalid receipt, it changes nothing on your device and lets you try again. If a later account check refuses access, the app clears local account data and signs out; that alone is not proof that a new server-side deletion succeeded. Account deletion does not delete source workouts from Apple Health. The seven-day encrypted-backup policy and minimal private deletion safeguards described above still apply. Copies already exported or sent to DeepSeek cannot be recalled by this control.

可在应用内使用 Delete account 请求删除 TrainLoose 账号。服务器确认成功后,在线账号数据与会话被移除,应用清除本地账号数据并退出登录。删除请求若被拒绝账号访问,应用会清除本地账号数据并退出,但不声称删除成功。其他删除结果无法确认时(如服务器错误、超时、离线或无效回执),设备上的数据保持不变,并可主动重试。后续账号检查若拒绝访问,也会清除本地账号数据并退出,但这本身不是新的服务器删除已成功的证明。账号删除不删除 Apple Health 中的来源运动。上述七天加密备份及最小私有防复活元数据政策不变;已导出或已发送给 DeepSeek 的副本不能由此追溯撤回。

Optional Health sharing

The optional Apple Health switch controls cloud sharing. Apple Health read access is still managed by iOS and never grants cloud upload on its own. If you allow upload, TrainLoose can sync a foreground 28-day workout slice to your account: activity type, start/end timestamps, known duration or unknown, timezone and its basis, hashed source identity, and request/stream metadata for deduplication and recovery. Hashes are linkable private identifiers, not anonymous totals. Raw HealthKit UUIDs/anchors, source app/device names, heart-rate/power/route/sample series stay on your device. Turning on Apple Health requests the corresponding device reads in one Apple dialog and grants workout upload, automatic sync and measurement reads together; the available statistics covered by this notice are described below. When you turn on Apple Health under its current disclosure and allow the corresponding reads in Apple's single Health dialog, TrainLoose may upload workout-associated average/maximum heart rate, at most 24 relative-time summary intervals, distance in meters and active energy in kilocalories when available. Raw measurements, GPS, power/sample series, device/source names and HealthKit identifiers/anchors remain on device. Optional statistics use the same bounded ordered sync and deletion recovery, separately revisioned from workout facts. Missing readings never become zero; the explicit new-version zone clearing rule is described below, and other known statistics retain their protection. Sync does not require RPE or start a selected-workout analysis. The existing history window is unchanged; automatic upload is included in the Apple Health switch. With Apple Health on under its current disclosure and the corresponding reads allowed in Apple's single Health dialog, TrainLoose may save seven active heart-rate zone durations, observed active duration, aggregation bounds and your optional explicit LTHR/six zone boundaries with the associated workout. Raw heart-rate samples stay on device. Local zone settings are optional and account-scoped; saving them alone sends nothing. Missing configuration or unusable evidence stays unknown, never guessed or zero. A new versioned upload with null zones can clear previously saved zone evidence; an unreadable device store or failed query does not become a clearing upload. The same sync stream, source identity and history window are retained; automatic upload follows the Apple Health switch. Pending uploads keep their original bytes and settings. Old grants do not authorize this expanded collection; a new disclosure and explicit confirmation are required.

Health summaries may be used by AI only when both Apple Health and AI Trainer are on. No separate Health-in-Trainer prompt is required. For a Trainer message you choose to send, DeepSeek may receive requested date-level workout type, known duration or unknown, mapping confidence and incomplete coverage within the current 28-day slice. Exact workout clocks, sync/source/account identities, raw samples, private free-text feedback and cumulative older profiles are excluded from this read; current numeric RPE is the separately disclosed exception. Your stated feeling and intent remain primary. Permission changes do not invoke the model or apply a plan; existing-plan changes still require your review. DeepSeek processing is governed by its linked Open Platform terms; TrainLoose does not promise provider zero retention or no training. Processing may be outside your country. Do not share information you do not want the provider to process. If you separately enable training memory, eligible multi-day workout summaries may support dated observations during a Trainer task with current Health permission. Only relevant unexpired observations supported by currently eligible sources may be included. Pause stops learning and model use; withdrawing Health AI permission invalidates dependent memory, and regrant alone does not restore it. This adds no Health data types or history days. Initial planning and renewal do not receive Health-derived memory. Background workout check-ins and workout notifications are off and are not offered in this GA release. Sync does not initiate AI or change plans. Your feelings and intent come first. With the current AI permission, an ordinary Trainer turn you choose to send may read current saved numeric session RPE, including zero and decimals. Apple Health evidence additionally requires current linked Health-AI permission. That evidence may include workout-associated average/maximum heart rate, up to 24 relative-time heart-rate summary intervals, distance and active energy. These are limited statistics, not raw samples, exact workout clocks, routes, source identifiers or private free-text feedback. Reported RPE times known minutes remains AU, not TSS. The separate qualified-HR estimate and daily reference contributions are described below, not medical advice. Missing evidence stays unknown. Analyze this workout is a separate explicit request about one current saved record, not an automatic report on sync and not a plan or workout change. It uses the ordinary Trainer request budget and recovery. It does not update optional training memory. An old result is hidden when its source or permission is no longer current; checking status does not send another model request. Old permissions do not authorize these new fields; review the new AI and, for Health, linked Health notices. Estimated training load is calculated by TrainLoose, not by the language model. Qualified workout-associated active heart-rate zone time takes precedence for this numerical estimate; otherwise it uses your explicitly reported session RPE and known duration, or remains unknown. The original RPE and RPE times minutes (AU) remain separate. Qualified heart-rate evidence needs your optional, explicitly configured sport threshold (LTHR) and six increasing boundaries for seven zones, matching active duration and at least 90% observed active-time coverage. Pauses, conflicting overlaps and gaps over 30 seconds are excluded; missing time is not scaled up. No threshold is inferred from age, maximum heart rate or coarse averages. Ordinary Trainer and explicit selected-workout analysis may receive the current estimate, method, coverage and limitations under current AI consent; Health evidence also requires current linked Health-AI consent. Raw samples and zone configuration are not sent to DeepSeek. Ordinary Trainer may additionally read application-calculated daily estimated CTL, ATL and opening TSB based only on recorded training in its permitted history window. These are partial contributions, not total fitness, recovery or a medical assessment. Unrecorded exercise and initial history stay unknown; decay does not prove rest, and positive TSB is not permission to train harder. No diary, configured threshold or complete history is required for chat, planning or sync. Your feeling and intent still come first. Estimates are not official TrainingPeaks TSS or guaranteed equivalents across sports; AU is not added to them. This release does not expand background check-in inputs, automatically analyze each sync or change an accepted plan.

Saved workouts remain in your active account beyond the 28-day sync/read window until deleted. Confirmed source deletions remove online canonical workout facts and private feedback, but retain minimal account/source/canonical identifiers, revision, deletion and sync receipt metadata to prevent resurrection. External account/source deletion markers are retained separately from seven-day encrypted backups and replayed before restored service. They contain no workout facts. Account deletion removes online account data and sessions. Saved conversations and copies already downloaded or sent to DeepSeek are separate from workout deletion; no deletion from provider copies is promised. Health data is not used for advertising, marketing profiles or sale. Privacy questions: shuhengzhao@qq.com Reported RPE and retained check-in questions/choices are saved account data included in the complete export and account deletion. Correcting or deleting source evidence, or clearing its RPE, removes affected derived question text. Minimal private identity/revision and recovery fences prevent erased reports returning through old requests or backups; they contain no erased report or question text. Restore disables automatic/background choices and clears derived questions. Turning analysis off does not erase retained questions, conversations or copies already exported or sent to the provider. Saved workout statistics and source-bound analysis associations are included in the complete account export and account deletion. Source/RPE/statistics changes, deletion and AI/Health permission changes invalidate and scrub affected selected-analysis results. Temporary execution evidence is cleared on completion or Stop. Ordinary chat history and receipts retain only an acknowledgement, not the selected-analysis text. Regrant and restored backups cannot revive invalidated results. A complete export is historical data, not permission to redisplay a result as current. Copies already exported or sent to DeepSeek are separate; no provider erasure is promised. Saved zone summaries and the configuration recorded with them are included in the complete account export and account deletion and follow existing source deletion and backup-erasure protection. Estimates and daily reference curves are recomputed from current records, not stored as a second workout history. The owner may read already saved records and their estimates without AI consent. Changing local settings alone does not erase cloud data or rewrite an old pending upload. Source/RPE/zone changes or permission changes invalidate affected selected-analysis results.

Turning off Apple Health stops new cloud upload, automatic sync, measurement reads and Health-informed AI use. Turning off AI Trainer stops new AI requests without changing the Apple Health upload choice. Local stopping is immediate, but remote withdrawal is confirmed only by the server. Already admitted provider work may finish and stay charged, but cannot publish a new Health-informed reply after withdrawal. Regrant does not restart obsolete requests or restore deleted workouts. Withdrawal does not erase stored workouts, plans, conversations or provider copies. Local Health access remains a separate device setting. Core Calendar, review and ordinary text Trainer remain available without Health sharing.

Manual workout history

Calendar forms and explicit Trainer recording save completed workouts to your TrainLoose account, separately from planned sessions. Date and activity are required; duration and feedback are optional. Neither entry method writes into Apple Health. Form entry does not require AI permission. Using saved manual evidence in Trainer requires the current cloud AI notice above; Apple Health sharing remains a separate choice. Account export includes saved workout data. Account deletion removes online workouts and private feedback.

Workout corrections and deletion / 运动修改与删除

Where available, Calendar and Saved history let you edit or delete a selected manual or Trainer-saved workout without a plan, cloud AI or Health permission. Native deletion asks for confirmation. An explicit Trainer instruction can also correct or delete a selected saved workout directly; this still requires ordinary cloud AI permission for the new message, but adds no new model-data scope or automatic regrant. Workout changes are separate from plan Review: Keep or Adjust cannot undo them. Apple Health records remain source-owned; edit or delete them in the source app, then sync. TrainLoose does not write to Apple Health or silently hide its records. A lost reply can be checked without another model call or replaying the change, including after permission withdrawal.

Deleting a manual or Trainer-saved workout removes its online facts and private feedback. Minimal opaque account/workout/capture/request identifiers, revision, deletion and recovery metadata remain separately from rolling backups to prevent restored or delayed requests from reviving it; these are linkable private metadata, not anonymous data, and contain no workout facts or feedback. Workout deletion does not erase saved conversations, exported copies or content already sent to DeepSeek. Ordinary edits remain backup-covered account data; a backup created before a correction cannot guarantee that correction after disaster recovery.

功能开放后,可以在 Calendar 或 Saved history 中编辑、删除选中的手动或 Trainer 保存的运动,不要求计划、云端 AI 或 Health 权限;直接删除会先确认。明确发给 Trainer 的指令也可以直接修改或删除选中的记录,新消息仍须普通 AI 同意,但不增加模型读取范围或自动重新授权。运动修改与计划审核独立,Keep / Adjust 不能撤销它。Apple Health 记录由来源管理,请在来源应用中修改或删除后再同步;TrainLoose 不写入 Apple Health,也不会静默隐藏记录。回复丢失时可以查询原请求,撤回权限后仍可恢复已完成回执,不重复修改或调用模型。

删除手动或 Trainer 保存的运动会移除在线运动事实和私有反馈。另行保留最小的不透明账号、运动、补录与请求标识,以及修订、删除和恢复元数据,防止备份恢复或迟到请求复活记录;这些不是匿名数据,不含运动事实或反馈,保留时间独立于滚动备份。删除运动不会删除已保存聊天、已导出副本或已经发送给 DeepSeek 的内容。普通编辑属于备份覆盖的账号数据;灾难恢复使用早于修改的备份时,不能保证保留该次修改。

Goals, instructions and optional training memory / 个性化

For new initial planning, Trainer and plan-renewal requests, your saved Main goal, training goals and user-written Trainer instructions are automatically included as bounded context for DeepSeek. Current feelings and explicit intent take precedence. Saving instructions or reviewing goals itself does not call AI. Only you can edit instructions; changing goals still needs explicit Review and does not regenerate plans. If you separately enable the learned training profile, relevant unexpired observations may also be sent to DeepSeek. During an ordinary Trainer reply, the model may propose a small dated memory from your explicit statements or eligible saved workout patterns; TrainLoose saves it only after the reply succeeds and source and privacy checks pass. Learning is not an extra background AI call. Pause stops both learning and profile use by AI; clear removes learned content without pausing future learning. This is not medical assessment, personality inference or a compliance score. Health-derived memory additionally requires separate current Health permission for that Trainer task; initial planning and renewal do not receive Health-derived memory. An existing AI grant does not automatically grant this expanded scope.

Current instructions remain until replaced, cleared or the account is deleted. Instruction and goal-operation receipt text expires after seven days; minimal account-scoped recovery fences remain. Temporary task input copies expire within 24 hours and are removed on completion or applicable privacy changes. Learned observations expire within 42 days for recent experience/workout patterns or 180 days for other supported preferences, without background refresh. Forget/clear removes dependent memory and prevents old tasks or old evidence from restoring it. Content-free private deletion fences survive backups; restoration conservatively clears stale memory and copied inputs. Current instructions, goals and readable memory are included in complete account export and deletion. Controls cannot erase copies already exported or sent to the provider.

Main goal 和 Trainer instructions 是你主动保存的账号设置。新规划、Trainer 对话和接续计划会在当前 AI 同意范围内自动使用它们,当下感受和明确意愿优先。保存描述不调用模型,Agent 不能改写描述;目标修改须经审核确认,不自动重做计划。

长期训练画像需要单独启用,可查看依据、纠正、忘记、清空或暂停。暂停保留可读内容,但停止学习和向模型提供画像;清空不会暂停以后的学习。旧聊天、旧运动和迟到请求不会自动学回已忘记内容。Health 衍生画像仍须本轮有效的独立 Health-AI 同意;不扩大同步天数或数据类型,初次规划和接续计划不会使用 Health 衍生画像。旧 AI 授权不会自动扩大。

描述、目标和当前可读画像纳入账号导出及删除。描述与目标操作回执正文最长保留七天,请求内临时副本最长保留 24 小时并在完成或相应隐私变更后清除;画像观察有 42 天或 180 天有效期。为防止旧请求或备份复活已忘记内容,会保留不含正文的私有恢复与删除元数据;已发送给服务商或另行导出的副本不能由这些控制追溯删除。隐私联系:shuhengzhao@qq.com。

Workout effort, automatic sync and check-ins / 运动感受与主动问询

New RPE-capable manual forms require a user-entered whole-session effort value from 0 to 10, including decimals and explicit zero. Apple Health sync never waits for RPE; you can add, correct or clear a separate TrainLoose annotation later, without editing Health-owned facts or writing back to Apple Health. Legacy saves/recovery and the current Trainer save contract are unchanged. Reported effort × known minutes is an AU subtotal, not TSS, CTL, ATL, TSB, recovery or a diagnosis. Missing reports/duration/coverage stay unknown. Reported RPE/AU are retained separately; the new numerical estimate may use qualified heart-rate zones as described below.

Automatic workout upload is included in the Apple Health switch. It uses the same limited workout fields, history window and recovery stream. iOS may defer background work; no instant or continuous delivery is promised. Workout notifications are off and are not requested in this GA release; this does not block permitted sync. No APNs or coaching push is enabled.

Background workout check-ins and workout notifications are off and are not offered in this GA release. Sync does not initiate AI or change plans. Your feelings and intent come first.

Reported RPE and retained check-in questions/choices are saved account data included in the complete export and account deletion. Correcting or deleting source evidence, or clearing its RPE, removes affected derived question text. Minimal private identity/revision and recovery fences prevent erased reports returning through old requests or backups; they contain no erased report or question text. Restore disables automatic/background choices and clears derived questions. Turning analysis off does not erase retained questions, conversations or copies already exported or sent to the provider.

Conversation history redisplays only the currently retained tail: up to eight complete exchanges and 16 KiB per thread, not a permanent archive. Signing out clears private device display. Opening old text never resumes a thread, sends a message or accepts a proposal. Home rhythm reads saved owner records for up to 90 dates; this does not enlarge the Health collection or model-sharing window or imply that missing days were rest days.

新版手动补录要求用户填写 0–10 的整次运动 RPE,支持小数和明确的零;Health 先同步,之后可单独补填、纠正或清除 RPE,不修改 Health 来源事实。旧请求恢复及当前 Trainer 补录版本不变。RPE×已知分钟仅为已记录部分的 AU 小计,不是 TSS、CTL、ATL、TSB 或恢复诊断;缺失不当作零;原始 RPE 独立保留,新增数值估算可优先使用合格心率分区。

Apple Health 一个开关统一控制运动上传、自动同步和运动测量读取;AI Trainer 单独控制向 DeepSeek 发送信息。后台 AI 问询和运动通知在本次 GA 中关闭,不询问授权。自动同步沿用原字段、天数和恢复记录,iOS 可能延迟执行。本次 GA 不产生 Workout ready 运动通知,也不启用 APNs 或问询推送;通知关闭不妨碍已授权同步。

后台问询在本次 GA 中关闭;同步不自动调用 DeepSeek,也不自动修改计划。你的感受与意愿优先。普通 Trainer 读取当前 RPE 需要本次有效的 AI Trainer 同意;Health 来源同时需要 Apple Health 开启。旧授权不自动扩大,但不逐条消息询问权限。

完整导出与账号删除包含 RPE、自动同步选择和保留的问询。来源纠正、删除或 RPE 清除会移除关联问询正文;防复活元数据不含已清除值或正文。灾难恢复会关闭自动及后台权限、清除派生问题;撤回本身不追溯删除对话、外部导出或服务商副本。对话历史仅显示每线程最多八轮/16 KiB 的保留片段;退出登录清除本地展示。Home 最多查看 90 天已保存记录,不扩大 Health 采集或 AI 分享范围,空白日期不等于休息。隐私联系:shuhengzhao@qq.com。

Current workout evidence and explicit analysis / 运动证据与单次分析

With the current AI permission, an ordinary Trainer turn you choose to send may read current saved numeric session RPE, including zero and decimals. Apple Health evidence additionally requires current linked Health-AI permission. That evidence may include workout-associated average/maximum heart rate, up to 24 relative-time heart-rate summary intervals, distance and active energy. These are limited statistics, not raw samples, exact workout clocks, routes, source identifiers or private free-text feedback. Reported RPE times known minutes remains AU, not TSS. The separate qualified-HR estimate and daily reference contributions are described below, not medical advice. Missing evidence stays unknown. Analyze this workout is a separate explicit request about one current saved record, not an automatic report on sync and not a plan or workout change. It uses the ordinary Trainer request budget and recovery. It does not update optional training memory. An old result is hidden when its source or permission is no longer current; checking status does not send another model request. Old permissions do not authorize these new fields; review the new AI and, for Health, linked Health notices.

When you turn on Apple Health under its current disclosure and allow the corresponding reads in Apple's single Health dialog, TrainLoose may upload workout-associated average/maximum heart rate, at most 24 relative-time summary intervals, distance in meters and active energy in kilocalories when available. Raw measurements, GPS, power/sample series, device/source names and HealthKit identifiers/anchors remain on device. Optional statistics use the same bounded ordered sync and deletion recovery, separately revisioned from workout facts. Missing readings never become zero; the explicit new-version zone clearing rule is described below, and other known statistics retain their protection. Sync does not require RPE or start a selected-workout analysis. The existing history window is unchanged; automatic upload is included in the Apple Health switch.

Saved workout statistics and source-bound analysis associations are included in the complete account export and account deletion. Source/RPE/statistics changes, deletion and AI/Health permission changes invalidate and scrub affected selected-analysis results. Temporary execution evidence is cleared on completion or Stop. Ordinary chat history and receipts retain only an acknowledgement, not the selected-analysis text. Regrant and restored backups cannot revive invalidated results. A complete export is historical data, not permission to redisplay a result as current. Copies already exported or sent to DeepSeek are separate; no provider erasure is promised.

新版普通 Trainer 对话在你主动发送且当前 AI 同意有效时,可读取已保存的明确数字 RPE,包括零和小数;Health 来源还需要新版关联 Health-AI 同意。可用证据包括运动关联的平均/最高心率、最多 24 段相对时间心率汇总、距离和活动能量,不发送原始样本、GPS、来源标识、精确运动时刻或私密自由文本。RPE×已知分钟仍是独立 AU,不是 TSS;新增心率估算和每日参考贡献见下文,不是医疗判断。缺失仍是未知,感受与意愿优先。

启用 Apple Health、确认其当前统一说明并在 Apple 的同一系统对话框中允许对应读取后,可补传上述有限统计;基础同步不等待心率或 RPE。原同步天数与数据范围不扩大;自动同步纳入 Apple Health 开关,后台问询在本次 GA 中关闭。“分析这次运动”由你单独发起,只分析当前选定记录,不因同步自动运行,不改计划或运动,不自动写入训练画像。旧授权不会自动扩大;手动记录及当前已保存内容的查看不要求 AI 同意。

来源、RPE、统计、权限改变或删除后,关联单次分析正文失效并清除;重新同意或备份恢复不能复活旧结果。请求临时证据在完成或停止等待时清除;普通聊天只保留完成提示。完整导出含保留的统计和分析记录,但不是当前显示权限。已发送服务商或另行导出的副本不承诺追溯删除。账号删除覆盖这些新数据。隐私联系:shuhengzhao@qq.com。

Estimated recorded training load / 已记录训练的估算负荷

Estimated training load is calculated by TrainLoose, not by the language model. Qualified workout-associated active heart-rate zone time takes precedence for this numerical estimate; otherwise it uses your explicitly reported session RPE and known duration, or remains unknown. The original RPE and RPE times minutes (AU) remain separate. Qualified heart-rate evidence needs your optional, explicitly configured sport threshold (LTHR) and six increasing boundaries for seven zones, matching active duration and at least 90% observed active-time coverage. Pauses, conflicting overlaps and gaps over 30 seconds are excluded; missing time is not scaled up. No threshold is inferred from age, maximum heart rate or coarse averages. Ordinary Trainer and explicit selected-workout analysis may receive the current estimate, method, coverage and limitations under current AI consent; Health evidence also requires current linked Health-AI consent. Raw samples and zone configuration are not sent to DeepSeek. Ordinary Trainer may additionally read application-calculated daily estimated CTL, ATL and opening TSB based only on recorded training in its permitted history window. These are partial contributions, not total fitness, recovery or a medical assessment. Unrecorded exercise and initial history stay unknown; decay does not prove rest, and positive TSB is not permission to train harder. No diary, configured threshold or complete history is required for chat, planning or sync. Your feeling and intent still come first. Estimates are not official TrainingPeaks TSS or guaranteed equivalents across sports; AU is not added to them. This release does not expand background check-in inputs, automatically analyze each sync or change an accepted plan.

With Apple Health on under its current disclosure and the corresponding reads allowed in Apple's single Health dialog, TrainLoose may save seven active heart-rate zone durations, observed active duration, aggregation bounds and your optional explicit LTHR/six zone boundaries with the associated workout. Raw heart-rate samples stay on device. Local zone settings are optional and account-scoped; saving them alone sends nothing. Missing configuration or unusable evidence stays unknown, never guessed or zero. A new versioned upload with null zones can clear previously saved zone evidence; an unreadable device store or failed query does not become a clearing upload. The same sync stream, source identity and history window are retained; automatic upload follows the Apple Health switch. Pending uploads keep their original bytes and settings. Old grants do not authorize this expanded collection; a new disclosure and explicit confirmation are required.

Saved zone summaries and the configuration recorded with them are included in the complete account export and account deletion and follow existing source deletion and backup-erasure protection. Estimates and daily reference curves are recomputed from current records, not stored as a second workout history. The owner may read already saved records and their estimates without AI consent. Changing local settings alone does not erase cloud data or rewrite an old pending upload. Source/RPE/zone changes or permission changes invalidate affected selected-analysis results.

训练负荷由 TrainLoose 确定性计算,不由模型计算。合格的有效活动心率分区优先;否则使用用户明确填写的整次 RPE 与已知时长,或保持未知。原始 RPE 和 RPE×分钟的 AU 独立保留,不与估算值相加。分区需要你自愿为某项运动明确配置 LTHR 和六个递增边界;不从年龄、最高心率或粗均值猜测。活动时长须匹配,观测覆盖至少 90%;暂停、歧义重叠和超过 30 秒的空隙不补齐,也不按比例放大。没有阈值或完整历史不影响聊天、规划或同步。

启用 Apple Health、确认其当前统一说明并在 Apple 的同一系统对话框中允许对应读取后,可将七区间时长、有效活动时长、聚合边界和对应的可选显式阈值/分区配置保存到同一账号运动记录。原始样本留在设备端。配置仅保存在本机、按账号隔离,保存配置本身不上传。旧授权不自动扩大;历史天数不变,自动同步纳入 Apple Health 开关。旧待确认上传保留原字节。新版空分区可清除已存分区,但读取失败不当成清除。

当前 AI 同意有效时,普通 Trainer 或主动单次分析可读取当前估算值、方法、覆盖与局限;Health 来源还需当前关联 Health-AI 同意,不向 DeepSeek 发送原始样本或阈值/分区配置。普通 Trainer 还可读取许可历史内的每日估算 CTL/ATL/TSB;后台问询输入不因此扩大。它们仅是所示范围内已记录训练的部分贡献,不代表总负荷、体能或恢复。CTL/ATL 使用 42/7 天递推,TSB 是昨日收盘差;缺失运动、未评分日期和初始状态仍未知。衰减不证明休息,正 TSB 不代表适合加量,不是官方 TrainingPeaks 指标或医疗判断。感受与意愿优先。

完整账号导出包含保存的分区与配置,来源删除、账号删除及备份擦除保护覆盖它们;估算和每日曲线按当前记录重算,不另存运动历史。查看自己的已存记录和估算不要求 AI 同意。只修改本地配置不等于删除云端数据。来源、RPE、分区或权限变化使关联单次分析失效;旧结果不能通过重新授权或备份恢复复活。隐私联系:shuhengzhao@qq.com。

Optional About you details / 可选个人资料

If you separately enter optional About you details (reported age with its confirmation date, sex, height in cm or weight in kg), new user-requested initial planning, Trainer and plan continuation tasks may send a minimized snapshot to DeepSeek under this current AI permission. Saving or clearing these fields does not require AI or Apple Health permission. Missing values and prefer-not-to-say sex remain unknown. An age not confirmed within 366 calendar dates is marked stale and its number is not sent; age never advances automatically. Details are not inferred from Health, messages or learned memory and do not authorize background demographic analysis. Your current feelings and intent take priority. These details do not enable BMI, calorie targets, ideal weight, diagnosis or automatic heart-rate thresholds. Existing AI grants do not authorize this expanded input; a new explicit confirmation is required.

Saved details are independent account settings in complete export and account deletion. Unfinished model tasks retain a private first-admission copy for at most 24 hours; ordinary edits apply to new tasks, not retries. Clearing any field, choosing not to disclose sex, clearing all details or changing AI permission invalidates affected task copies before further model access. Completion, Stop, expiry and disaster restore remove these private copies. A separate restore journal retains only account/request identity, revision, removed-field names and time, not personal values, to prevent resurrection. Clearing settings cannot recall data already sent to the provider, retained conversation text or exported copies.

About you 中的年龄及确认日期、性别、身高(厘米)、体重(千克)均可选,不填也能聊天和安排训练;保存、修改或清空不要求 AI 或 Apple Health 授权。重新明确同意新版 AI 告知后,新发起的首次规划、Trainer 与计划接续请求可向 DeepSeek 发送最小资料快照。旧同意不会自动扩大。缺失及性别“不愿透露”按未知处理;年龄满 366 个公历日期未确认则标记过期,不发送年龄数值,也不会自动加一。不从 Health、聊天或记忆推断这些资料,不因此开启后台画像分析、BMI、热量目标、理想体重、诊断或自动心率阈值;当下感受和意愿优先。

资料独立保存并纳入完整导出与账号删除;未完成任务的私有快照最长保留 24 小时。普通修改仅影响新任务,重试保留首次快照;清除字段、不愿透露性别、清空资料或变更 AI 授权会阻止相关旧快照继续发送。完成、停止、过期及灾难恢复移除任务快照。恢复保护日志仅保存账号/请求标识、版本、清除的字段名称和时间,不含字段值。清空不追溯撤回已发送给服务商、已保留对话或已导出的副本。隐私联系:shuhengzhao@qq.com。