Privacy Policy

Last updated 20 October 2026

Hanashi (“we”, “us”) runs a service for paid video conversations with real Japanese people, with live captions and translation. People may use it for language practice, shared interests or social conversation. This policy explains what we collect, why, who helps us process it, how long we keep it and what you can ask us to do. It is written to meet Hong Kong’s Personal Data (Privacy) Ordinance, the EU and UK GDPR, Japan’s APPI, and US state privacy laws such as the CCPA.

1. Who can use Hanashi

During beta, learners must be at least 16 and Senseis must be at least 18. We ask your age during onboarding and prevent profiles below the applicable minimum from being completed. If you believe someone below the minimum has an account, contact us so we can review and remove it.

2. What we collect

• Account: your name, email address and Google account identifier when you sign in with Google, plus your sign-in sessions.
• Welcome promotion: claiming welcome Points requires SMS verification through Twilio Verify. Your full phone number is sent to Twilio to deliver the code; Hanashi stores a keyed hash, the last four digits and verification/audit records, not your full number or SMS code. The phone-use hash is retained permanently to prevent a second claim, including after account removal. Signup and paid purchases do not require phone verification.
• Profile: display name, photo, age, languages, level, interests and bio. Other members can see your profile, but not your age or email. A learner's optional last name, conversation preferences and goals stay private. Your chosen display name need not be your legal name. Sensei profile photos and optional 10–20 second hello videos are published after founder photo approval. Social accounts stay private to the Sensei and founder/admin for verification, sourcing and possible partnerships; learners never receive their handles or links. Replacing media removes its previous stored version.
• Conversation records: who talked with whom, when, for how long, Points used, historical conversation packages, payments, ratings, feedback and reports.
• Required operational transcripts: finalized original speech and translations from both registered participants, speaker identity, timestamps, languages and session ID are retained for 30 days from the call’s start. Authorized Hanashi personnel may review them for safety investigations, dispute resolution and translation-quality review. One unchecked policy acknowledgment is required during initial setup or before an existing user’s next call; only material policy changes require acknowledgment again. No earlier conversations are added. This policy does not record audio or video.
• Words you choose to save: finalized source text, translations and recap cards saved only after both participants affirmatively enable written conversation memory. Saved conversation memory is unavailable to under-18 and unknown-age accounts; eligible learners can still call without optional Learning Memory. Required operational transcripts are separate and apply to registered calls after acknowledgment. No audio or video is saved by transcript permission.
• Sensei welcome practice: your scenario questions, written answers, AI feedback and evaluation status are saved with your account for you and the founder to review and improve the exercise. The existing language provider reads the submitted answers. This practice gives coaching, with no numerical grade or automatic rejection. Avoid real names and other people’s private details in your answers.
• Technical data: IP address, device and browser type, connection quality, error and latency measurements, used for security, abuse prevention and reliability.

Private founder interviews: The founder chooses a locked room mode. “No recording” rooms do not record or save transcripts. “Recording required” rooms require each participant’s explicit pre-join agreement to video recording and saved transcription, with the notice version and timestamp logged. You may decline and ask for an unrecorded interview without automatic disqualification. After both people agree and connect, a persistent indicator shows capture status. Material is stored privately in Cloudflare for interview review and technical troubleshooting, accessible only through founder authentication, and expires 30 days after room creation; scheduled cleanup deletes it. Withdrawing consent ends the recorded room and deletes retained review material. Conversation contents are not written to operational logs or published. Founder downloads are separate copies and cannot be deleted automatically by Hanashi. Provider processing and database recovery copies follow their own retention terms; expiry does not promise instant erasure from backups. Recorded rooms remain unavailable while storage or provider privacy clearance is disabled.

We record a call only for the people who allow it. Hanashi asks each person once whether their side of calls may be recorded so clips can be made afterwards: “clips OK”, “ask me first” or “don’t record me”. The recording is made on the sensei’s device and includes only the people who allowed it, plus the call’s captions and any moments marked during it. Both people in the call can open it for 60 days to make clips; during the pilot the founder may also review it to improve Hanashi. It is never published by us. You can change your answer at any time in your settings. Otherwise, speech is sent to our speech providers as it happens so captions and translations can appear, and is not kept as audio by us under the operational transcript policy. Required written transcripts are captured even when visible subtitles are hidden; muting the microphone stops capture of that person’s speech. Optional written memory is saved only when both eligible participants enable it. Private assistance uses the providers needed to deliver your request.

Clips are made on your own device from that recording and are not uploaded to us. A clip of only yourself needs no one’s permission. A clip that shows the other person is only possible if they allowed recording, and if they chose “ask me first” it waits for their approval.

Optional language-use progress saves only numerical word counts and associated utterance identifiers and revisions for up to 90 days when a learner chooses “Remember my word counts”. It does not save audio, captions or AI answer text. These counts are estimates, not a fluency score. Turn the option off to pause saving, or use “Forget” in the call to delete saved counts and disable saving.

3. Why we use it (legal bases)

• To provide the service you asked for: accounts, matching, calls, captions, translation, private assistance and recaps (performance of a contract).
• To keep people safe: age checks, blocking, reports, moderation and fraud prevention (legitimate interests and legal obligations).
• To keep the service working and improve it: diagnostics and aggregated usage (legitimate interests).
• Conversation memory is off by default and needs explicit consent from both participants. We remember each preference for future calls. Either person can turn it off in the call or settings: saving stops and all retained conversations they participated in, including completed calls, and their learning derivatives are deleted. Account settings let each participant read, export or delete a saved conversation. We record the permission version, purpose, date and withdrawal history. Saved text is used only for recap, learning feedback, useful phrases and continuity with the same Sensei. Live captions work with memory off.

We do not sell your personal data or share it for cross-context behavioural advertising, and we do not use your conversations to train AI models.

Required transcript acknowledgment records notice of a condition of calling, separately from optional Learning Memory consent. Learning Memory withdrawal or deletion does not erase operational transcripts. Only authenticated authorized administrators can review these records, and each successful access is audited without transcript contents. Speech recognition or translation can fail; records may be incomplete or inaccurate.

4. Camera and microphone

Hanashi needs your microphone, and your camera if you turn it on, to connect a call. Your browser asks for permission, and you can withdraw it at any time in your browser settings. Calls connect directly between the two people where possible, or through Cloudflare’s relay when a direct route is not available.

5. Who processes your data

We use these service providers only to run Hanashi, under contracts that limit how they may use the data:

• Cloudflare: Hosting, the database, call signalling and the relay that connects calls
• Google: Sign-in, and Gemini for AI replies, translation and learning help
• Microsoft Azure: Speech recognition, translation and AI voices
• ElevenLabs: Live speech-to-text for captions
• Alibaba Cloud (Qwen): Backup translation
• OpenAI: Backup language processing and transcription
• Resend: Sign-in and service emails, when enabled
• Twilio Verify: SMS verification only for claiming welcome Points, when enabled
• Stripe: Card payments for Hanashi Points packs and historical purchase refunds

We may also disclose data where the law requires it, to protect someone’s safety, or as part of a merger or sale of the service, in which case this policy continues to apply.

6. International transfers

We are based in Hong Kong, and our providers process data in several countries, including the United States, the European Union, Japan and Singapore. Where the law requires it, we rely on safeguards such as the European Commission’s Standard Contractual Clauses.

7. How long we keep it

Required operational transcripts become inaccessible exactly 30 days from call start, including incomplete, interrupted and failed calls. Automatic cleanup removes expired text on the existing 15-minute schedule without user activity. Reports do not extend this lifetime. Policy acknowledgments and content-free administrative access audits are separate records; managed database recovery copies and provider retention follow their own policies. This operational policy does not change optional Learning Memory or private-interview retention.

• For new conversations started after both adults explicitly accept the current saving policy, written transcripts, learning memory and saved recap cards stay until either participant deletes them or withdraws permission. Older conversations keep their original 30-day expiry from the call request; reads stop at expiry and scheduled deletion removes live text. Managed backups can retain earlier data until their platform retention expires.
• Bug reports: description and steps expire after 90 days, optional screenshots after 30 days. Minimal review/reward audit remains for financial integrity. Do not submit credentials or private conversations.
• Exploratory careers submissions, including optional PDF resumes: private to founder review in our Cloudflare database, deleted after 90 days or sooner using your private removal receipt.
• Permission history and moderation corrections document choices and safety decisions. Founder name corrections require an audit and reason, and the affected user receives an in-app notice.
• Pilot research recordings (only if you agreed): 60 days, then deleted.
• Sign-in sessions: 30 days, or until you sign out.
• Account, profile and conversation records: while your account is open. After you delete your account we remove or anonymise them within 30 days, except where we must keep something to resolve a safety report, prevent abuse or meet a legal obligation.

8. Cookies and local storage

We use essential sign-in cookies. A Sensei-sharing link can set a seven-day first-arrival cookie to preserve that Sensei attribution through signup; it does not grant a discount unless an approved program is active. We use your browser’s local storage to remember preferences such as interface language and sound settings. We do not use advertising or third-party analytics cookies.

9. Your rights

Depending on where you live, you can ask to access, correct, delete or export your personal data, object to or restrict certain processing, and withdraw consent at any time. California residents can also request the categories of data we hold and will not be discriminated against for exercising these rights. To make a request, email isaac@hanashi.chat. We answer within 30 days (40 days where Hong Kong law allows it) and may need to confirm your identity first. You can also complain to your local data protection authority, which in Hong Kong is the Office of the Privacy Commissioner for Personal Data.

10. Security

We use encryption in transit, Cloudflare D1 encryption at rest, access controls and short-lived credentials for our providers. No service is perfectly secure; if a breach affects you, we will tell you and the relevant authorities as the law requires.

11. Changes and contact

If we change this policy materially, we will tell you in the app before the change takes effect. This policy is governed by the laws of the Hong Kong Special Administrative Region. Questions: isaac@hanashi.chat.