This English text governs accounts registered outside Korea. However, a local-language version applies where it is more favourable to the user or the mandatory law of the user’s country of residence requires it. Any ambiguity is interpreted in the user’s favour. For accounts registered in Korea, the Korean Privacy Policy governs.
The operator of SudaCall ("the Company") complies with Korea's Personal Information Protection Act and related laws and, for users outside Korea, with the data protection laws of their country that apply to the Company (sections 7 to 7-6), and processes users' personal data as set out below.
Public feed posts: text, photos and videos users upload to the feed (public posts), posting and editing times, author nickname and profile photo, reports and safety review results. Collected when users choose to post or report.
| Category | Items | How collected |
|---|---|---|
| Account (Google) | Google account email, name, profile photo | On Google sign-in |
| Account (Kakao) | Kakao member number, Kakao account email, nickname (gender, birthday, and birth year are used only as sign-up defaults, and only where the user has separately consented) | On Kakao sign-in (items shown on the consent screen) |
| Account (Naver) | Naver user identifier, email, nickname, gender, birthday, birth year | On Naver sign-in (items shown on the consent screen) |
| Account (Apple) | Apple account identifier, email (a relay address created by Apple if you choose to hide your email), name (if you provide it) | On Sign in with Apple in the iPhone app |
| Profile | Nickname, date of birth, gender (optional), profile photo (optional), confirmation that you meet the minimum age (18 for accounts registered outside Korea, 14 for accounts registered in Korea), whether parent mode is on (optional) | Age confirmation and parent mode at sign-up; the rest entered by the user when needed (profile settings, creator registration, and so on) |
| Calls | Call transcript (text), call summary, relationship data (how you are addressed, conversational context), call time and duration, age and gender you mention in a call (used for forms of address and conversational limits when your profile does not have them) | Generated while using the Service |
| Requests during a call (only when you ask) | The type and time (including repeats, days of the week, and the reminder's time zone) of scheduled calls and scheduled reminders (wake-up and medication calls, medicine-time and wake-up reminders, and so on) — never medicine names, doses, or other health information, the search terms and the song found when you ask for a song, the start point and destination names for directions or public transit directions, words you ask us to pass on to your family and confirm, the device state needed to carry out a request (whether notification and alarm permissions are granted, whether a navigation or messaging app is installed, and your phone model if phone guidance is on), and a record of which features we suggested or explained | Generated when you make a request during a call |
| Sleep mode (only if you turn it on) | Your sleep-mode preferences such as sound (including a chosen song), volume, play time, whether you use YouTube Premium, and whether to keep the sleep screen on; when you used sleep mode and how often it was suggested, accepted, or declined; and the length of the part of a call in which you appeared to be asleep | Generated while you use sleep mode |
| Location (only with your separate consent to the collection and use of personal location information) | Your device's location (latitude and longitude) at the moment you ask during a call — used once for what you asked (finding places nearby, weather where you are, the start point for transit directions, or sending to your care contact) and not kept, except that if you ask us to save the place, those coordinates are kept with the saved place; for places you save, the name you give them, the place identifier (Google place ID), your device's coordinates when you saved it, an optional short note you add, and when you saved them (the place name and address Google returns are deleted within 30 days and fetched again when needed); what you consented to and when; and records of the use and provision of location information (who, when, for what purpose, and to whom — never the coordinates) | Collected from your device's location services when you ask during a call; the records are kept automatically |
| Community activity | Your comments and replies and their authors; avatar and comment likes, their targets, the accounts that pressed them, and posting or pressing times | When you post a comment or press a like |
| Voice | Recordings submitted when registering a voice or as a creator, and the consent-statement recording | Recorded or uploaded by the user |
| Payment | Subscription product, payment date and amount (app marketplace order information) | At payment |
| Settlement (creators) | Payout account details, identity verification data | Entered by the user |
| Marketing | Whether marketing consent was given (optional) | Optional consent at sign-up |
| Events and rewards | Invite code and who invited whom, whether the app was installed from an invite link, reward records, and (when an event runs) prize-draw entries, winners, and gift codes issued | When you install from an invite link or enter a code, and while an event runs |
| Collected automatically | Device information, app version, access logs, error logs, service usage records (which screens you open in the app and how long you stay, whether microphone permission was granted, whether you played a voice sample, and which in-call request features were used and whether they worked — never conversation content, anything you type, recipients, song titles, destinations, or location), a device token for notifications, an app installation identifier | Collected automatically during use |
| Check-in | (only if you turn it on) mood, and whether meals, sleep, going out and medication came up, plus topic tags, and how many times sleep mode was used | Derived from call transcripts and sleep-mode usage records |
| Check-in link (care contact) | The label, email address, and emergency text number (optional, Korea only) that the care contact enters | Entered by the care contact when creating the link |
Remembered recipients for texts and calls (on your device only) — When you ask during a call, for example "text my mom", the one phone number you pick yourself on your phone's contact picker is saved only on your device, together with the name you used (for example "mom"). It is never sent to our servers or to AI processing servers, and is used only to fill in the recipient the next time you ask for the same person. SudaCall does not read your whole contact list (it does not use the contacts permission). You can delete them one by one or all at once in Settings > Remembered recipients, and they are also deleted when you sign out or uninstall the app. (Added 2026-09-30)
Information received through social sign-in is used only to identify the member and to prefill the sign-up screen. If you sign in with a different social provider using an email that already has an account, we link it to the existing account rather than creating a new one, so your call history and subscription carry over. We keep a link record showing which provider account belongs to which member; it is deleted when you delete your account.
If you disconnect SudaCall in your Kakao account settings, Kakao notifies the Company. If that account was created with Kakao only, we delete all information, the same as account deletion. If the account can also sign in with Google or Naver, we delete only the Kakao link record and keep the account.
Call audio itself is not stored. Voice during a call is used only for real-time recognition and synthesis; what is stored is the text transcript and the summary.
For requests during a call, SudaCall only opens the screen on your phone. For message, KakaoTalk, and email drafts, placing a call, adding a calendar event, and directions, we open the relevant app screen on your device, and you send, call, or save yourself. For directions we pass the place name you said to your map app, and SudaCall does not use your device's location for this. Alarms and timers are set in your device's clock app (on iPhone, as a SudaCall notification). If you ask for a medicine-time or wake-up reminder as a notification instead of a call, we send a SudaCall notification at that time; we keep only its type, time, and days, and delete it as soon as you say "turn off the reminder" (section 5). If you ask how to get somewhere by public transport, we tell you bus and subway routes by voice; if you do not say where you are starting from, we use your current location as the start point only if you have given the new location consent (section 4-2). The content of a draft or event is created from the conversation and passed to that screen on your device. The contact you pick (name and phone number) and anything stored in your device's contacts or calendar is used only on your device and is never sent to our servers.
Songs play in the official YouTube player on the call screen. When you ask, we either play a short part of a song or play it quietly in the background while the conversation goes on; if you want the whole song, we ask you first, end the call, and open it in the YouTube app on your phone (or on the web if the app is not installed). Section 3 covers the player.
Sleep mode does not decide you are asleep by listening. We do not analyse breathing, snoring, or other sounds; we judge only by how long you have not spoken. Call audio is not stored in sleep mode either. Short words heard after you seem to have fallen asleep (what looks like sleep talk) get no reply, and although they stay in the transcript with a marker, they are not used in call summaries, relationship memory, or check-in reports. If check-in reports are on, your care contact is told only how many times you used sleep mode (section 4-1). You choose the sleep-mode sound. A song you choose plays quietly in the official YouTube player on the call screen during the call; once you fall asleep and the call ends, it keeps playing on the app's sleep screen (dimmed to the lowest brightness, with the player still visible) until the time you set (section 3). If you tell us you use YouTube Premium, we hand the song to the YouTube Music app on your phone after the call, and that app plays it. If you do not use YouTube Premium and want to sleep with the screen off, we play royalty-free audio downloaded from our server (public-domain or CC0 piano pieces and the like) and sounds the app generates (such as noise), and no information is sent to an outside service to play them. We do not use sleep-mode records to judge your sleep or your health.
We check your location only if you have consented separately, and only at the moment you ask during a call. We check it once, only while the app is on screen during a call; we never check it while the app is in the background, and we never track you. Section 4-2 and the Location-Based Service Terms set out the details.
Under Article 22-2 of Korea's Network Act, the Company tells you which device permissions the Service uses, grouped as below. Optional permissions are requested through your device's permission prompt the first time you use the feature that needs them. For location, we also ask for your consent to the collection and use of personal location information, separately from the device permission (section 4-2).
| Type | Permission (device) | Used for | If you do not allow it |
|---|---|---|---|
| Required | Microphone | Voice calls with your AI friend | You cannot make calls |
| Optional | Notifications | Incoming calls, scheduled calls (including wake-up and medication calls), scheduled reminders (medicine time, wake-up time, and so on), after-call messages, announcements | Scheduled and incoming calls, scheduled reminders, and announcements do not arrive (calls you place still work) |
| Optional | Alarms & reminders (exact alarms, Android) | Ringing scheduled calls at the time you set | Calls may ring a few minutes late; a wake-up call you set by voice is set as a clock-app alarm instead |
| Optional | Full-screen notifications (Android 14+) | Showing the call screen over the lock screen | Calls ring as a notification banner only |
| Optional | Display over other apps (Android) | Showing the incoming call screen while you use another app | You tap the notification to open the call screen |
| Optional | Nearby devices (Bluetooth, Android 12+) | Calls through Bluetooth earphones or headsets | Calls use the phone speaker, earpiece, or wired earphones |
| Optional | Camera | Taking a profile photo | Choose from your gallery or use a default avatar |
| Optional | Photos (iPhone photo library) | Choosing a profile photo | Take a photo or use a default avatar |
| Optional | Location (only while using the app) | Once, at the moment you ask during a call — saving a place, finding places nearby, weather where you are, the start point for transit directions, and (with separate consent) sending your current location to your care contact. Background location is never used | Location features are unavailable (say a place name and we will look it up by name). With approximate location only, the nearest place may be wrong |
| Optional | Contacts (where the feature is offered) | Finding a number by name when you ask to call or text someone during a call — on your device only | You pick the contact in the contact picker |
| Automatic | Internet, vibration, keeping the screen on (during calls and on the sleep-mode sleep screen), audio settings (speaker switching, call volume), restoring scheduled calls after a reboot, foreground service (Android — keeps the microphone during a call, plays sleep-mode sounds after the call with the screen off), background audio (iPhone — keeps the call going, plays sleep-mode sounds), set alarm (Android — sets alarms and timers in the clock app) | Running the Service | Normal permissions the operating system grants at install; they do not access information stored on your device |
You can use the Service without allowing optional permissions; only the related feature is limited. You can change them at any time — Android: Settings > Apps > SudaCall > Permissions / iPhone: Settings > SudaCall. You can also check their status in the SudaCall app under Settings > App permissions.
Adding a calendar event does not use any calendar permission. We open your calendar app without one (on Android, its add-event screen; on iPhone, we copy the event details and open the calendar app). We never read what is stored in your device's calendar.
Publishing user feed posts, automated safety checks (including video frame analysis and audio transcription), and handling reports. Public posts are visible to other users in the relevant avatar feed; subscription-only media follows the selected audience.
To the extent needed to provide the Service, processing is entrusted to the companies below and mostly takes place outside Korea (United States, China). The information transferred is the minimum needed to provide the Service (voice stream, conversation text, account identifiers, and location coordinates if you have consented).
| Processor | Entrusted work | Country |
|---|---|---|
| Google (Firebase/Cloud) | Authentication, data storage, server operation, sending notifications (push) and email, storage of user feed posts and media | United States and others |
| Google (Firebase Analytics / Google Ads) | Measuring advertising effectiveness (app installation identifier and the event recording the first call lasting more than one minute — excluding conversation content and account information) | United States |
| LiveKit | Real-time voice call connection (with consent to personal location information, relaying the coordinates your app reads during a call to the call-processing server — never stored) | United States |
| Deepgram | Speech recognition (speech → text) | United States |
| xAI (SpaceXAI LLC) | Speech recognition (speech → text) for Korean-language calls and some calls in other languages (switching to another speech-recognition provider if it fails), and conversation generation (AI model) in some test calls run by the Company | United States and other countries ** |
| Cartesia | Speech synthesis (text → speech), voice model creation | United States |
| Microsoft Corporation (Azure OpenAI) | Primary path for conversation generation (AI model); automated visual safety review of user feed posts and video audio transcription | Outside Korea — the United States and other regions supported by Azure Global Standard. The resource location (Korea Central) does not limit processing to Korea. |
| OpenAI | Conversation generation (AI model), speech recognition (some calls in other languages, and as a fallback path); automated image safety checks of user feed photos and video frames | United States |
| Anthropic | Conversation generation (AI model); call quality analysis and service improvement (reads call records once a day to find and fix quality problems — not used for AI training, kept up to 30 days, then deleted) | United States |
| Google (Gemini) | Conversation generation, summarisation, search (AI model) | United States |
| DeepSeek | Conversation generation (AI model) — fallback path only, used when the primary and other default paths are slow or unavailable (section 3-1) | People's Republic of China |
| Fireworks AI | Conversation generation (AI model hosting) | United States |
| Inworld AI | Speech recognition and synthesis (basic character voices) | United States |
| Fish Audio | Speech synthesis (pro voices) | United States and other countries * |
| ElevenLabs | Speech synthesis (some legacy features) | United States |
| Kakao · NAVER | Social sign-in (member identification) | Republic of Korea |
| NAVER Cloud | In-call search and finding places (search terms, place names), emergency check-in text messages | Republic of Korea |
| TMAP Mobility Co., Ltd. | Calculating public transit routes when both the start and the destination are in Korea (start and destination coordinates — your current location as the start point only with location consent) | Republic of Korea |
| Google (YouTube API Services, Google Maps Platform) | Song and video search (including search during calls) and song playback on the call screen and sleep screen, finding places (search terms, place names), and finding the nearest place's name and address, places nearby, the name of the neighbourhood you are in, and public transit routes when the start or the destination is outside Korea, from location coordinates (with consent) | United States |
The Company applies each provider's opt-out from training where their API data policy offers it, and reviews those policies periodically because they can change.
* Fish Audio states in its privacy policy that it is a US (Delaware) business but does not disclose data centre locations, and says processing may occur in other countries through group companies and subprocessors. The Company is confirming the processing country with that provider and will update this table once confirmed.
** xAI (SpaceXAI LLC) is a US company. It states that its default API does not guarantee a processing region (its US-only processing route does not support voice features), and that API requests and responses are stored encrypted for 30 days to audit for abuse or misuse and then deleted, and are never used for training without explicit permission.
YouTube (Google): SudaCall uses YouTube API Services for song and video search (including search during calls) and for playing songs on the call screen. For search we send only the search terms — never your account details — and we do not access your YouTube or Google account. Songs during a call (a short part of a song, or music playing in the background while you talk) and sleep-mode songs play in the official YouTube player, which is provided directly by Google. While the player is open, Google may collect device information and playback information, may store or read cookies or similar technologies on your device, and may show advertisements in videos. That information is governed by the Google Privacy Policy, and your use of YouTube is governed by the YouTube Terms of Service. We do not receive the information Google collects; we use only the player's state (playing, paused, ended) to run the call. Playback stops when the app leaves the screen or the call ends. The exception is sleep mode: if you fall asleep and the call ends, playback continues on the app's sleep screen (dimmed to the lowest brightness, with the player still visible) until the time you set, and it still stops if the app leaves the screen. A whole song opens in the YouTube app on your phone after the call (or on the web if the app is not installed), and if you use YouTube Premium and ask for it in sleep mode, we hand the song to the YouTube Music app after the call and that app plays it. Your use of the YouTube and YouTube Music apps is governed by the YouTube Terms of Service and the Google Privacy Policy.
Transfer to Microsoft (Azure OpenAI): The transferred data is the conversation text in the table below and the material needed to check user feed posts, sent in real time over an encrypted API during calls and checks. The resource is in Korea Central, but the Global Standard deployment allows processing in the United States and other regions. Microsoft may retain inputs and outputs for up to 30 days for abuse monitoring and then deletes them. On Azure latency, failure or filtering we use OpenAI and the existing fallback paths. Processor privacy contact: Microsoft Corporation (https://aka.ms/privacyresponse).
Transfer to DeepSeek (China): Only when the primary and other default paths for conversation generation are slow or unavailable, the conversation text in the table below may be sent to DeepSeek over an encrypted API and processed on servers in the People's Republic of China. China has a Personal Information Protection Law, but Chinese laws such as the National Intelligence Law may require companies there to cooperate with state authorities, including by providing data. As with other overseas transfers, you can refuse this transfer only by stopping use of the Service or deleting your account (table below).
To provide the AI call service, the Company transfers personal data overseas as set out below. Every processor in the table in section 3 other than the Korean companies (Kakao, NAVER, NAVER Cloud, TMAP Mobility) is an overseas transfer. The recipient and country differ by the type of information. If you live outside Korea, your personal data is also transferred to the Republic of Korea, where the Company is located and operates the Service.
| Type | Recipient (country) | Details |
|---|---|---|
| Conversation content (text) — full utterances and relationship records (nickname, form of address, age range, gender, interests, and other conversational context), and the names and addresses of places saved or found during a call, and notes you add to them; public transit route results (stop and line names, travel time); and (in calls with phone guidance on) your phone model | Microsoft (Azure OpenAI, US and other overseas regions) — primary path for call response generation / OpenAI (US) — other default generation and fallback on Azure latency, failure or filtering / DeepSeek (China) · Fireworks AI (US) — fallback path on latency or outage | Sent in real time during a call so the AI can compose a reply. Calls handled on the Azure path use Microsoft Global Standard regions (the US and other overseas regions). Other default calls use the United States; if a fallback path is used during an outage or latency spike, data may be sent to servers located in China. |
| Voice — the user's voice (speech recognition), character voice synthesis and cloning, call connection | Inworld AI · Deepgram · OpenAI · Cartesia · LiveKit (all US), xAI (US and other countries) — speech recognition for Korean-language calls and others, Fish Audio (pro voices) | Real-time recognition, synthesis, and call connection are handled by US providers. Speech recognition for Korean-language calls and some calls in other languages is handled first by xAI (switching to Deepgram or another provider if it fails), and xAI does not guarantee that processing stays in the United States (see the footnote to the table in section 3). In some test calls run by the Company, xAI also generates the conversation. Pro voices created by creators are synthesised through Fish Audio and sent to that provider's overseas servers (see the footnote to the table in section 3). |
| Additional processing — call summaries and relationship memory, quality assessment, content review, news briefings, call quality analysis and service improvement | OpenAI · Google · Anthropic (all US) | Conversation text is summarised and analysed after the call ends. |
| Search terms — song and video search terms (including search during calls), place names, and the neighbourhood name (district and city — never coordinates) when you ask about the weather where you are | Google (US) — YouTube API Services, Google Maps Platform, Gemini | Only the search term is sent, when you ask or when the conversation needs a search. Account details are never sent with it. Place search in Korea uses NAVER Cloud (Republic of Korea). A weather search uses only the neighbourhood name; weather searches in Korea use NAVER Cloud (Republic of Korea) or Google (US). |
| Location — your device's coordinates at the moment you ask during a call (with consent to personal location information) | Google (US) — Google Maps Platform | Sent once, at that moment, to find the nearest place's name and address, places nearby, and the name of the neighbourhood you are in (weather). Coordinates used as the start point for transit directions go to TMAP Mobility (Republic of Korea) when both ends are in Korea, and to Google when either end is outside Korea. The coordinates your app reads are relayed to the call-processing server through the call-connection provider LiveKit (US), which only relays them and does not keep them. Account details are never sent with it, and the Company does not keep the coordinates sent to Google (they are kept only with a saved place, if you asked to save it — section 4-2). Without consent nothing is sent, and the call service works as usual. |
| User feed posts (public posts) — uploaded text, photos, video frames and audio or transcription needed for checking | OpenAI (United States) — image moderation; Microsoft (Azure OpenAI) — visual safety review and transcription; Google (Firebase/Cloud) — storage (stored in the Seoul, Korea region) | Encrypted API transfer on upload, automatic review or report processing. Image review receives every photo and bounded video samples. Transcription receives video audio. Users can decline these transfers by not uploading media, or delete posts or their account; existing call features follow the existing policy. |
| Advertising measurement data — app installation identifier and the event recording the first call lasting more than one minute, plus device and app-version information and basic events such as first launch sent automatically by the Firebase Analytics SDK | Google (Firebase Analytics / Google Ads) (United States) | To count which users who installed the app after seeing an ad went on to continue a call. No conversation content, account information or email addresses are sent. We do not collect advertising identifiers; advertising personalisation signals are disabled, and this data is not used for personalised advertising. Retention follows the Google Analytics data-retention setting (set by the Company to 2 months). |
| Item | Details |
|---|---|
| Time and method of transfer | Sent via API over encrypted communication (TLS) at the time of use, such as during a call |
| Retention and use period | Per each provider's API data policy (for example, xAI keeps requests and responses for 30 days to audit for abuse, then deletes them; the Company applies each provider's opt-out from training and reviews policy changes periodically) |
| How to refuse | If you do not want overseas transfer, you can refuse by stopping use of the Service or deleting your account |
| Effect of refusal | Because AI conversation is generated through overseas providers, refusing transfer means the call service cannot be used |
Call content (transcripts, summaries, relationship data) is visible only to you and is not disclosed to other users, including creators. The Company accesses call records only to the minimum extent needed to improve service quality and verify safety, under internal controls, and cooperates with state authorities only upon a lawful official request, within the procedures and scope set by law.
Only if you turn it on yourself, SudaCall sends a periodic check-in summary drawn from your calls to a person you choose (your "care contact"). If you do not turn it on, nothing is sent and nothing else about the service changes.
These summaries are generated automatically by AI from call transcripts. They are not a medical diagnosis or assessment and must not be used as a basis for health decisions. SudaCall is not a healthcare provider (in the United States, it is not a HIPAA covered entity), and this feature is not medical care.
The Company processes personal location information as a location-based service provider under Korea's Act on the Protection, Use, etc. of Location Information (the "Location Information Act"). The details are governed by the Location-Based Service Terms.
User feed posts and media: kept until the user deletes them or the account is permanently deleted (7 days after the deletion request). Media rejected by safety review is deleted after 7 days. Deleted posts are withdrawn from public display immediately; storage deletion failures are retried. Where law requires retention, the legally required records are segregated and retained only for that period. Provider-side retention follows the verified processor terms in section 3-1.
You may at any time request access to, correction of, deletion of, or suspension of processing of your personal data. You can do this yourself in the app under Settings (Account), or request it at the contact details below. You can delete call records directly in the app, and delete relationship data with "Reset memory". Your rights over personal location information (withdrawing consent, pausing, access, correction) are governed by section 4-2 and the Location-Based Service Terms. If you live in the United States, Japan, Taiwan, Brazil or elsewhere in Latin America, sections 7-2 to 7-6 set out further rights and how to exercise them.
Age is checked at sign-up by confirming that you meet the minimum age, and also by date of birth if you enter one. If the user is below the minimum age, sign-up is refused and the information entered is not stored.
This section applies to users and creators in the United States and is provided to meet state biometric privacy laws, including the Illinois Biometric Information Privacy Act (BIPA), the Texas Capture or Use of Biometric Identifier Act (CUBI), and Washington's biometric statute.
Depending on your state (for example California under the CCPA/CPRA, and comparable laws in Colorado, Connecticut, Texas, Virginia and others) you may have the right to know what personal data we hold and why, to get a copy, to correct it, to delete it, and to limit the use of sensitive personal data.
This section applies to users in Japan and gives the information required by Japan's Act on the Protection of Personal Information (APPI).
Recipient safeguards: the following summary supplements sections 3 and 3-1. We apply each provider’s opt-out from AI training where offered. The stated retention periods remain unchanged; an unspecified period is not a promise of immediate deletion.
OECD principles: we assess recipient safeguards against the OECD’s eight privacy principles: collection limitation, data quality, purpose specification, use limitation, security safeguards, openness, individual participation and accountability. Minimising transferred data and limiting purposes (sections 1–3), training opt-outs and retention controls (sections 3 and 5), encryption and access controls (section 8), disclosure in this policy, rights requests (sections 6 and 7-3) and contractual commitments under each provider’s business terms are the corresponding safeguards.
Countries that cannot be fixed in advance: Azure Global Standard can route processing among supported global regions; Korea Central is the resource location and does not fix the processing country. xAI’s default voice API does not guarantee a region, and Fish Audio does not disclose data-centre locations and may use overseas group companies or subprocessors. This routing and lack of disclosure prevent us from identifying every processing country in advance. See the provider information and the Japanese PPC’s foreign-system information; further details are available from the section 9 contact.
United States government access: US laws and intelligence frameworks, including the CLOUD Act, FISA and Executive Order 12333, may permit government access to data under the applicable procedures and safeguards.
| Recipient | Training and retention safeguards |
|---|---|
| Google (Firebase/Cloud) | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| LiveKit | Call connection and relay; relayed location coordinates are not stored. Other retention follows section 3-1. |
| Deepgram | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| xAI (SpaceXAI LLC) | No training without explicit permission; encrypted API requests and responses retained for 30 days for abuse monitoring, then deleted (section 3). |
| Cartesia | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Microsoft Corporation (Azure OpenAI) | Not used to train foundation models; abuse monitoring retains inputs and outputs for up to 30 days, then deletes them (section 3-1). |
| OpenAI | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Anthropic | Call quality analysis is not used for AI training; retained for up to 30 days, then deleted (section 3). |
| Google (Gemini) | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| DeepSeek | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Fireworks AI | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Inworld AI | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Fish Audio | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| ElevenLabs | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Kakao · NAVER | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| NAVER Cloud | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| TMAP Mobility Co., Ltd. | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
| Google (YouTube API Services, Google Maps Platform) | Training opt-out applied where offered; retention follows the API policy described in section 3-1. |
This section applies to users in Taiwan and gives the notice required by Article 8 of Taiwan's Personal Data Protection Act (PDPA).
This section applies to users in Brazil under the General Data Protection Law (Lei Geral de Proteção de Dados Pessoais — LGPD).
This section applies to users in Latin American countries other than Brazil (for example Mexico, Argentina, Colombia, Chile and Peru).
Data protection officer: Lee Heekyu (Head of Strategic Planning, Strategic Planning Office) · sudacall@creamhouse.net · +82-2-508-2712
Location information manager: Lee Heekyu (Head of Strategic Planning, Strategic Planning Office) · sudacall@creamhouse.net · +82-2-508-2712
Business: CreamHouse Co., Ltd. (Representative Choi Woojin, business registration number 220-86-89511) · 19 Yangpyeong-ro 21ga-gil, Rooms 208–209, Yeongdeungpo-gu, Seoul, Republic of Korea
Reports and consultation on privacy infringement (Korea): Privacy Infringement Report Centre (privacy.kisa.or.kr, 118), Personal Information Dispute Mediation Committee (kopico.go.kr, 1833-6972)
We respond to enquiries within 5 days of receipt.
If this policy changes, we announce it in the app 7 days before it takes effect (30 days before, for material changes).
Revised 2026-09-09 — with the introduction of social sign-in (Kakao, Naver), Kakao and Naver account information was added to the items collected, and the handling of linking and unlinking sign-in methods was made explicit. The default processing path for conversation generation also changed, so the overseas transfer details were corrected to match reality (default path: United States). This revised policy takes effect on 16 September 2026.
Revised 2026-09-19 — account deletion now has a 7-day hold (sign in again within it to restore your account), and section 5 now sets out the purpose, items, and period of the minimal record kept for 1 year after deletion to prevent abuse. Preventing abuse was added to the purposes in section 2. This revised policy takes effect on 19 September 2026.
Revised 2026-09-29 — service usage records (which screens you open in the app and how long you stay, whether microphone permission was granted, whether you played a voice sample — never conversation content or anything you type) were added to the items collected automatically, analysing how the app is used was added to the purposes in section 2, and section 5 now sets their retention (1 year from collection, destroyed without delay when you delete your account). In addition, sections 1, 2, and 5 now set out the items, purposes, and retention for the features you can ask for during a call (scheduled calls, alarms, songs, directions, message drafts, placing calls, adding calendar events, and passing words on to your family) and for sleep mode, and a new section 1-1 lists app permissions. Sections 3 and 3-1 now name the outside services used for song and video search, song playback on the call screen, and finding places (YouTube API Services, Google Maps Platform, NAVER Cloud), and check-in reports (section 4-1) now include how many times sleep mode was used and passing on words you ask us to pass on and confirm. A new section 4-2 covers personal location information processed with separate consent (saving places, finding places nearby, and sending your current location to your care contact), and section 9 now names the location information manager. We also brought the description of features already in service — Sign in with Apple, the sign-up age check, invites and events, and sending notifications and check-in text messages — in line with how they work. This revised policy takes effect on 29 September 2026.
Revised 2026-10-01 — scheduled reminders (such as medicine-time and wake-up reminders), public transit directions, and guidance on using your phone for your phone model were added to the things you can ask for during a call. Sections 1, 1-1, 2, and 5 now set out the items, purposes, and retention of scheduled-reminder settings. TMAP Mobility (Republic of Korea), which calculates transit routes, was added to sections 3 and 3-1, and your phone model and transit route results were added to what is sent to conversation-generation providers. Weather where you are and the start point for transit directions were added to the purposes of personal location information (section 4-2); these two are provided only to users who consent anew to them. This revised policy takes effect on 1 October 2026, the day it was announced. The previous version is here.
Revised 2026-10-04 — xAI (SpaceXAI LLC) was added to the providers that handle speech recognition (transcription) during calls. xAI now handles speech recognition first for Korean-language calls and some calls in other languages, and the existing providers (such as Deepgram) take over if it fails. In some test calls run by the Company, xAI also generates the conversation. Sections 3 and 3-1 were updated accordingly with the processor and the overseas transfer details (item transferred: call audio; country: United States and others; retention: 30 days under xAI's policy), and OpenAI's entrusted work was brought in line with how it actually works (speech recognition for some calls in other languages and as a fallback path). The items of personal data collected and the purposes of processing have not changed. This revised policy takes effect on 4 October 2026, the day it was announced. The previous version is here.
Revised 2026-10-07 — Anthropic's role was extended to work that improves call quality. Once a day an AI reads call records (conversation text) to find and fix places where the conversation went wrong; Anthropic does not use this content to train AI and keeps it for up to 30 days before deleting it. The outsourced tasks in section 3 and the overseas transfer details (additional processing) in section 3-1 were updated accordingly. The items collected and the purposes of processing did not change. This revised policy takes effect on 7 October 2026.
Revised 2026-10-10 — Section 5 now states that the record of your consent to auto-renewal ("after the trial, you are charged monthly; cancel any time"), taken when you start a paid subscription or free trial (account identifier, version of the wording, product, time of consent, device type, app version), is part of the contract and payment records kept as required by law, and is kept for 5 years even after you delete your account. This consent record is the only new item collected; the purposes of processing (performing the contract and handling disputes) have not changed. This revised policy takes effect on 10 October 2026. The previous version is here.
Revised 2026-10-16 — Added what we collect from text, photos and videos users post to avatar feeds (public posts), the purposes (posting, automated safety checks, report handling), and retention and deletion. Added Microsoft (Azure OpenAI) as a processor and overseas transfer recipient for the primary path of conversation generation and for feed post checks.
Revised 2026-10-18 — This policy now covers every account registered outside Korea, not only the United States: the minimum age outside Korea is 18 (sections 1, 2, 4-2 and 7); the United States provisions (sections 4-1, 7, 7-1 and 7-2) apply if you live in the United States; new sections 7-3 to 7-6 give the notices for Japan, Taiwan, Brazil and the rest of Latin America; section 3-1 now states that data of users outside Korea is also transferred to Korea and explains the DeepSeek (China) fallback path; and English is the governing text for accounts registered outside Korea. The items collected, purposes, processors and retention periods have not changed. This revised policy takes effect on 18 October 2026, the day it was announced.
2026-10-19 — Clarified local-language safeguards, the response deadline, deletion when an overseas account holder is found to be under 18, foreign-recipient safeguards for Japan and international-transfer grounds for Brazil. Brazilian account call content is excluded from quality analysis until separate consent. This revision takes effect on 19 October 2026, the day of announcement. Update notice.
2026-10-28 — Added advertising measurement to the purposes in section 2 and disclosed Google (Firebase Analytics / Google Ads) in sections 3 and 3-1, including the data transferred, purpose and retention. An app installation identifier, the event recording the first call lasting more than one minute, and basic SDK information and events are used to count continued calls after ad-driven installs. No conversation content, account information or email addresses are sent. Advertising identifiers are not collected, advertising personalisation signals are disabled, and this data is not used for personalised advertising. The Company sets Google Analytics data retention to 2 months. This revision takes effect on 28 October 2026. Update notice · Previous version.