Draft: pending legal review. The data controller's name, address, and email have not been filled in yet.
Last updated: October 7, 2026 (draft)
This document explains what personal data LiOnly (a translated voice-room service) collects, why, and who it is shared with. It is a draft; it should not be treated as binding until the data controller's identity is finalized and legal review is complete.
Account data: email address, display name, role, and a one-way hash of your password. The password itself is never stored as plain text.
Session data: when you sign in, a token is written to your browser; the database stores only its SHA-256 hash, never the token itself. The token is valid for 30 days.
Transcript data: who spoke or wrote in a room, which language it was translated from and to, the text of what was said or written and its translations, whether the line was spoken or written, and a timestamp. The audio itself is not kept — what remains is the written-down text. Messages typed in a room go into the same transcript, with the same retention period and the same reading rule.
Emoji reactions: relayed live to the people in the room; they are not stored on the server, not counted, and not added to the transcript.
Usage data: which room, which target language, whether speech or a written message was translated, how much translation was produced, and the corresponding cost. This is kept for billing and capacity management.
Audit log: who performed account and room actions (a name or password change, a credit request, moderation such as muting or removing someone from a room) and when. Actions taken by a member whom the room owner appointed as a helper are recorded under that member's in-room identity.
Account and session data are needed to grant room-opening and management rights to the right person.
The transcript exists so that, if there's a dispute about who said or wrote what in a meeting, a person can see their own words; the room owner can read the full transcript, a participant with an account can read only what they themselves said and wrote. A guest who joined without an account cannot read the transcript: their identity lives in their browser, so their own lines can't be told apart reliably. A line that a participant reports with “misunderstood” is visible to the administrator for quality review.
Written messages are translated so that everyone in the room can read them in their own language. Someone who joins later does not see messages written before they joined.
Usage data is kept to measure the service's cost and to run the credit/balance system.
Transcript lines (speech and written messages) are kept for 30 days; lines past that age are deleted automatically.
Account data is kept until the account is deleted.
When an account is deleted: the open rooms you own close right away. Your email address and name are removed from the account, from join records, from the speaker name in transcripts, and from the audit log; audit entries are not removed, the email is replaced with a short marker ("silinmiş:" followed by the first eight characters of the account ID). Your credit request notes are deleted.
What stays: balance movements and purchases are kept for accounting purposes, but are no longer tied to a name or email address. The text of what was said and written in transcripts is deleted when the retention period above ends. An email address that has been blocked from the service stays on the block list to prevent abuse.
The following providers are used to host the service and to produce translations. Each is sent only the data its job requires.
Railway (web application and database): The web application and the database are hosted on Railway; account data, session hashes, transcripts, and usage records are stored in this database. Server region: [TO BE FILLED IN].
Hetzner (translator bot and audio server): The translator bot and the LiveKit audio server that carries the room's audio run on a server rented from Hetzner, located in Germany (Nuremberg). LiveKit is not a separate service provider; it is open-source software we run ourselves on this server.
Cloudflare (domain): Domain name resolution (DNS) goes through Cloudflare. If the Cloudflare proxy is enabled, site traffic also passes through Cloudflare: [TO BE FILLED IN].
Google (Gemini live translation engine): When the default engine is running, the room's audio is sent to Google's Gemini model to produce the translation; audio is processed directly in a single step (speech-to-speech translation).
Soniox (recognition and speech synthesis on the chain engine): When the chain engine is active, speech is sent to Soniox's speech-recognition service to be turned into text, and the translated text is sent to Soniox's text-to-speech service to be read aloud.
Anthropic (translation on the chain engine and translation of written messages): On the chain engine, the recognized speech text is sent to Anthropic's Claude model to produce the translation. Messages typed in a room are sent to the Claude model to be translated into the room's other languages, whichever engine is running; so that the translation is accurate, the request also includes the room's last few messages and the topic and terms entered by the room owner.
Passwords are stored as one-way hashes; plain text is never kept anywhere. The session cookie is marked `httpOnly`, so a script running in the browser cannot read it.
The full list of rights over your personal data under Turkish law (KVKK) is on the KVKK Notice page.
Data controller: [TO BE FILLED IN]. Address: [TO BE FILLED IN]. Email: [TO BE FILLED IN].
Until these fields are filled in, there is no official address to apply to.
If this policy is updated, the date on this page will change.