Privacy Policy
Last updated 9 September 2026
Kaguya ("the app") transcribes and translates spoken conversation. This page states exactly what the app sends, where it sends it, and what is kept.
In short
· Recorded audio is sent to an external processing service so that it can be transcribed.
· Transcripts, translations and the audio itself are kept only on your device. We do not store them on our servers.
· Features that improve recognition can be turned off in settings. Nothing is stored for everyone by default (see 4).
· We do not collect your name, address, phone number or email address.
1. Who we are
The provider of the app ("we") handles information within the scope set out in this policy. Kaguya-talk Operations (the "operator") handles information as described here. Contact details appear at the end of this page.
2. Audio and text
To transcribe speech and to produce translations and summaries, data is sent to an external processing service (OpenAI, L.L.C., United States). Audio and text take different routes.
| Recorded audio | Sent directly from your device to the processing service. It does not pass through our servers. We never receive or store it. |
|---|---|
| Text (translations, summaries, study-mode phrases) |
Transcribed text is sent to the processing service via our server (Cloud Functions), so that usage and cost can be counted. The content of what was said is not stored while passing through. We record only the amount processed (token counts) and the estimated cost. |
| Where | OpenAI, L.L.C. (United States) |
| Purpose | Transcription, translation, summarisation, study-mode phrases |
Handling at the destination is governed by that service's own API data policy. We make no representation as to whether or for how long it retains the data.
3. What stays on your device
The history of transcripts, translations, summaries and recorded audio is stored only on your device. We do not store this history on our servers.
When a translation or summary is produced, that text passes through our server (see section 2), but it only passes through and is not stored.
- You can delete any recording at any time
- You can delete the audio alone and keep the text
- Removing the app removes all of it
4. What we store to improve recognition
When the same word is misrecognised repeatedly, the app can learn to recognise it correctly. Storage is kept strictly separate by purpose, and each can be turned off in settings. Nothing is stored for all users by default.
| Scope | What is stored |
|---|---|
| You (on by default) |
The misheard word, the correct word, up to 12 surrounding words from the same
utterance, language, domain and occurrence count. Used only for your own recognition. Never shared with other users. Retained for one year. You can delete all of it from settings at any time. |
| Your organisation (Business) |
Used only within the same organisation. Never used to improve dictionaries or models for other organisations or for Kaguya as a whole. Retained for one year. |
| Quality programme (off by default) |
Only if you opt in, the misheard word and the correct word alone
are used to improve Kaguya's overall accuracy. Surrounding words, full utterances, audio and summaries are not included. The record is not linked to you. Retained for two years. |
Every record carries the following.
- Which scope it belongs to (you / your organisation / quality programme)
- When it was recorded
- What it is for
- Its retention deadline
- Where the correction came from (your edit / a glossary rule / a re-check)
These records are stored against the anonymous user identifier described in section 5. In the "You" and "Organisation" scopes the link is necessary so that corrections apply to your own recognition. In the "Help improve accuracy" scope, only the pair of misheard word and correct word is taken, and it is not linked to a user.
Never stored
Full utterances, recorded audio and summaries are not stored on our servers in any scope.
Surrounding words in the "You" scope are fragments of what was said. Where the subject matter is sensitive, turn the feature off in settings or do not record.
5. Usage records
To enforce plan limits and prevent abuse, we store the following.
| Stored | An identifier distinguishing users, monthly minutes used, number of sessions, estimated processing cost, plan |
|---|---|
| Not stored | Name, address, phone number, email address, location |
This identifier is an anonymous user identifier issued automatically by Firebase Authentication (provided by Google LLC) using anonymous sign-in. No name or email address is required; the identifier alone is enough to start using the app. It does not, by itself, identify a person.
If you delete and reinstall the app, a new identifier is issued. In that case your previous usage minutes and the learning data in section 4 are not carried over.
6. Microphone
The app uses the microphone. While recording, Android shows a notification and iOS shows its recording indicator. The microphone is not used when not recording.
7. Third parties
Other than sending data to the processing service described above, we do not provide the information we hold to third parties, including for advertising or analytics.
8. Retention and deletion
Records on your device remain until you delete them. To request deletion of information stored on our servers, please contact us.
9. Before you record
Please satisfy yourself that recording is appropriate for the people present. In some countries and regions, consent is required before recording.
10. Changes
Changes to this policy are published on this page. Significant changes are also announced in the app.
11. Contact
Operator: Kaguya-talk Operations
Email: info@kaguya-talk.com