Privacy Policy

1. Information on your device

Plans, steps, deadlines, packing lists, history, packing preferences, selected places, onboarding answers, acceptance records and preparation memories are stored on your device. No earlie account is required. Apple backups may include this data if enabled.

AI automatically adds, edits, merges, reorganizes and removes memory using your plan input and existing memory, then replaces the full text only when you confirm a plan. In Settings, you can review and reset this text memory and any pending update. Places, packing removal history and buffer settings are stored separately. You cannot add, edit or delete individual items. Reset the memory if its contents are inappropriate. Changing buffer time in the plan editor updates the default for the next new plan. Saving locally does not itself retrain an AI model.

2. Information sent for AI processing

Using earlie requires agreement to this policy and the Terms. Acceptance text and links appear below the final onboarding button. There is no mode for using the app without agreement or an in-app consent-only withdrawal control. If you do not agree, stop using and delete the app.

New AI plans and AI edits send your request or edit (including speech converted to text), date, time zone, language, saved or pending memory, and items previously removed from similar plans through Cloudflare to OpenAI to generate and review steps, packing items and timing. AI edits also send the current plan, including manual edits and previous changes. AI may return a full replacement for the memory. It is instructed not to remember identifying information, one-off plans, guesses or instructions embedded in user input; this is not a guarantee of perfect detection.

Destination selection may send city/local-area context, candidate names, addresses, distances and travel estimates. Exact current GPS coordinates are not sent directly to OpenAI, but an entered address or candidate may reveal location. Reusing saved templates or history normally does not regenerate them with AI. This is not a separate mode that removes the requirement to agree.

3. Safety checks and AI provider handling

Before AI processing, relevant input, memory and candidate text are sent to OpenAI Moderation to check for harmful, sexual or other disallowed content. If input is flagged or the check is unavailable, plan generation does not proceed. AI-generated replacement memory is checked again by Moderation and for instruction-override patterns before saving. If that output is flagged, malformed or cannot be checked, only the memory update is discarded and the plan is still shown. Checks can make mistakes. Do not enter passwords, payment information, detailed medical information or other people's personal information without permission.

Generation and review use the Responses API with store: false. This does not mean zero retention: OpenAI may normally retain inputs and outputs for abuse monitoring for up to 30 days, with legal or safety exceptions. Supported models may retain temporary prompt caches for up to 24 hours. Moderation normally has no application-state or abuse-monitoring-log retention; consult provider terms as well.

Our OpenAI project currently has input/output sharing enabled. Eligible shared content may be used by OpenAI for service improvement and model training. The 30-day period is for abuse-monitoring logs, not a retention limit for all information shared for improvement or training. Stopping use or uninstalling does not automatically undo earlier transmissions or training use.

4. Maps, speech and notifications

With location permission, Apple location and map services process current location, destinations, search terms, travel mode and arrival times. Location access is optional; you can enter an estimated travel duration without choosing an exact destination.

Voice input uses Apple speech recognition, on-device when supported or otherwise processed by Apple. We do not send recorded audio to our API or OpenAI. Recognized text used for AI processing is handled as described above.

Notifications and Live Activities can reveal plan and step content to people who see your Lock Screen. Notifications are scheduled locally; we do not register push tokens with our own server. Manage location, microphone, speech and notification permissions in iPhone Settings.

5. Usage analytics, diagnostics and support

For reliability, troubleshooting, usage/cost monitoring and abuse prevention, Cloudflare and Langfuse process request time, latency, success/failure, token counts, model, app version and a random per-request identifier. For answer-quality review and troubleshooting, Langfuse stores validated plan input, memory used for processing, the generated candidate, review findings and corrections, and the final response under the same request. Plan text and place names appearing in input/output are included. API-key patterns and email addresses are mechanically redacted, but this is not guaranteed anonymization. Recorded audio, map candidate lists and exact current-location coordinates are not included in these content logs. Content is not sent to Cloudflare console logs or custom Firebase events. Records are not linked using a persistent advertising or user identifier.

The app uses Google Firebase Analytics and Crashlytics to understand usage and investigate problems. These services process screen views, onboarding completion, preparation starts/completions, completion source (app or Live Activity) and outcome, distribution environment and test-use classification, step counts, usage sessions, app/device/OS information and SDK identifiers for app installations and related measurement. Crash and speech-input diagnostics include stack traces, permission and audio-route states, processing stages and error codes. Our custom events and diagnostic logs exclude plan or memory text, transcripts, recorded audio and destinations. Completion events are relayed through Cloudflare to Google Analytics and retried after communication failures. For retries and duplicate prevention, Cloudflare retains content-free completion records and SDK identifiers for approximately up to 10 days after receipt. We use an SDK configuration without advertising identifier (IDFA) collection and disable ad personalization.

Cloudflare may process IP addresses and connection information for delivery and security. This website has no advertising, tracking pixels or analytics scripts.

Website support is by email (aigahaku+earlie@gmail.com); the app's Settings opens a Google Form. App name, language and app version are prefilled and sent to Google when the form opens. You enter contact details and your message and submit them yourself. These are processed for support and investigation. Plans and memories are not automatically attached.

6. Providers and international processing

We use Cloudflare for hosting, delivery and security; OpenAI for generation, review and moderation; Langfuse for performance diagnostics and input/output quality review; Apple for platform, map, speech and notification services; and Google for Firebase usage analytics, crash diagnostics and support forms. Data may be processed in the United States, Japan or other countries outside your residence, under provider contractual protections and this policy. We do not sell personal information or use it for cross-app advertising tracking.

OpenAI API data controls · Cloudflare · Langfuse · Apple · Google

7. Stopping use, deletion and correction

If you do not agree, stop using earlie and choose Delete App in iPhone Settings. Offload App retains data. There is no in-app button for deleting all app data or withdrawing consent only. You can review and reset the full memory in Settings. See Stop using & data.

We have no per-account plan-history database, but Langfuse stores the input/output quality logs described above. No fixed-day automatic deletion is currently configured for these logs; the data access window is not a deletion deadline. Diagnostics are retained as needed for troubleshooting, security and operation; support records as needed for the request and legal obligations. Contact us about access, correction or deletion, including earlier test records. We respond as required by law and technically possible.

Uninstalling does not directly delete Apple backups, support records or data already held by providers. This guidance does not limit applicable statutory rights.

8. Children, health and changes

earlie is not primarily directed at children and is not a medical device for diagnosis or treatment. Even for medical appointments or packing medication, avoid unnecessary health or identifying information. Contact us if a child's information has been provided inappropriately.

Material changes will be announced here or in the app; permission will be requested again where legally required. For privacy, access, correction or deletion requests: aigahaku+earlie@gmail.com. Operator: Kazuya Saito / Sato Apps.