Qlemi Voice-first writing

Privacy

Qlemi holds unpublished manuscripts. This page says what happens to them, including the parts that are not flattering.

Last updated 6 September 2026.

The short version. Your manuscript is stored on a server in Sweden that only I run. It is not sold, not shared, and not used to train anything. But it does not stay on that machine: reading a paragraph aloud sends its text to OpenAI, and some features send text to Anthropic. If that is a problem for your book, it is better to know now than later.

Who is responsible

Qlemi is run by Reda Ekengren, a sole trader (enskild firma) in Sweden, who is the data controller for everything described here.

Questions, requests and complaints: [email protected].

What Qlemi stores

What is sent to AI providers

This is the part that matters for an unpublished book, so it is spelled out rather than summarised. Two companies receive text, both outside the EU.

FeatureGoes toWhat is sentWhen
Reading aloudOpenAI The text of the paragraph being read Automatic on every play. Listen through a book and the whole book has been sent.
DictationOpenAI The audio you recorded, plus character names from your codex as a spelling hint Automatic when you dictate
Cleanup of dictationAnthropic The transcript of what you said Automatic after each dictation
Import structure detectionAnthropic Chapter titles, paragraph counts, and the first 240 characters of each chapter Automatic when you import a manuscript
Building the codex from the bookAnthropic The manuscript, in slices On request only when you press Scan manuscript
Consistency checkAnthropic The chapter's paragraphs and your codex entries On request only when you press Check codex

Both providers state that data sent through their APIs is not used to train their models by default. That is their commitment, not mine, and I cannot promise more than they do. Their current terms are the authority: OpenAI and Anthropic.

What I can promise is the part I control: your manuscripts are never sold, never shared with anyone else, never published, and never used to train anything of mine.

If you would rather not send your book anywhere

Reading aloud is the core of the tool and it cannot work without OpenAI, so there is no version of Qlemi that keeps your prose entirely on one machine. If that is the requirement, Qlemi is the wrong tool, and it is better to say so than to bury it.

Where your data lives

On a self-hosted server in Sweden that I run and maintain. The database holds the manuscripts; generated audio sits on a disk volume on the same machine. Nothing is stored with a third party.

Backups, described exactly as they are: the database is dumped nightly, kept for 30 days on the server and for a year on a second machine in the same home. There is no point-in-time recovery, so in the worst case up to a day of work could be lost. There is no copy outside the building yet, so a fire would take both. That is planned, not done.

How long things are kept

WhatHow long
Dictation recordingsOne hour. They are held in memory for the transcription job and never written to disk.
Uploaded import filesOne hour, the same way.
Email verification codes10 minutes, stored hashed.
Undo copies of deleted itemsUp to 7 days.
Sign-in session cookie30 days.
Generated audioKept indefinitely. It is stored under a hash of the text, so it can outlive the paragraph it was made from.
Paragraph revision historyKept for as long as the book is. Qlemi does not destroy prose you have written.
Manuscripts, codex, notesUntil you delete them.

Your rights

Under the GDPR you can ask for a copy of your data, have it corrected, have it deleted, take it elsewhere, or object to how it is handled. Email [email protected] and you will get an answer within 30 days.

Two of these you can already do yourself: any book can be exported to docx or epub at any time, and books, chapters, paragraphs and codex entries can be deleted from inside the app.

Deleting your account is not yet a button. Email me and I will delete the account and everything in it by hand. Building this properly is on the list; until it is, this is how it works, and pretending otherwise would be worse.

Email

Qlemi sends transactional email only: verification codes, password resets, and the occasional note that a job finished. These go through Lettermint, an email provider, which receives your address and the message. No marketing, no newsletter.

Cookies

One cookie, holding your sign-in session. It is httpOnly, Secure and SameSite=strict, which means it cannot be read by scripts and is not sent to other sites. There is no tracking, no analytics and no advertising anywhere in Qlemi.

Payment

Qlemi takes no payments today and stores no card or billing details. Paid credits for AI usage are planned. This page will be updated before a single payment is taken, and card details will be handled by the payment provider and never by Qlemi.

Children

Qlemi is not intended for anyone under 16 and accounts should not be created for them.

Security

Traffic is encrypted with TLS. Passwords are hashed with Argon2. Every query is scoped to the signed-in account, so one writer cannot reach another's book. The server is not shared with anyone else and the database is not reachable from the internet.

Qlemi is in active development and run by one person. It is careful where it counts, and it is not audited, not certified, and does not pretend to be.

If you find a security problem, email [email protected]. Reporting one in good faith will never be held against you.

Complaints

If you think your data has been handled wrongly, tell me first at [email protected]. You also have the right to complain to the Swedish Authority for Privacy Protection (IMY) at imy.se.

Changes

If this page changes in a way that affects what happens to your manuscript, you will get an email about it. The date at the top says when it was last revised.