Audio data policy

Julia is an installable web app (PWA) that lets you write your own book by talking: instead of typing, you dictate your ideas out loud and Julia transcribes them and drafts your manuscript. Voice recordings are therefore the most important material we process, and they deserve a dedicated policy.

This document explains in plain language which recordings we capture, how they travel from your device to the cloud, where they are stored, who processes them for transcription, who can access them, and for how long we keep them. It complements our general Privacy Policy, and both should be read together.

We only record audio when you decide to, by pressing the record button inside the author app, and always with your microphone permission. We process three types of recordings:

• Outline audio: the overall outline of the book you dictate when creating it, to establish the initial structure of your work.
• Idea audio: the loose, unordered recordings where you pour out your material, references, and content. They can be very long.
• Final instructions audio: the nuances, tone, and style guidance you dictate before asking for the draft to be written.

We never activate the microphone in the background or record without your explicit action. You are responsible for the content you choose to dictate; we ask you not to include sensitive data about others without their consent.

Capture uses the browser's native recorder (MediaRecorder) with an optimized codec (opus) at a low bitrate, so files stay lightweight without sacrificing the clarity of your voice for transcription. This reduces the space you consume and speeds up uploads.

• Cloud uploads are resumable: if your connection drops or you close the app, the upload resumes from where it left off and you never lose audio you already recorded.
• A file is marked as backed up only when the upload fully completes, so you never assume a recording is saved when it hasn't reached the cloud.
• All communication between your device and our servers travels encrypted over TLS.

Your recordings are stored in Firebase Storage (Google Cloud), the cloud storage service Julia uses. Each file is organized under a path tied to your user identifier and your book, in the form "books/{uid}/{bookId}/...", where "{uid}" is your unique author identifier.

• The outline audio is stored under "books/{uid}/{bookId}/outline".
• Each recorded idea is stored under "books/{uid}/{bookId}/ideas/...".
• Final instructions are stored tied to the same book, under your own identifier.

This organization guarantees that every recording stays isolated within the space of the author who created it. Access to those files is restricted exclusively to their owner, as detailed in the access and security sections.

To turn your recordings into text, Julia uses the Claude API, the artificial intelligence model by Anthropic. Transcription is a necessary step to organize your material and draft your book.

• Audio is sent to the Claude API exclusively from our server (server-side): your recording is never exposed publicly or shared openly on the internet.
• The API access key is kept in a secure secrets vault and is never delivered to the browser or visible on the client.
• Content is processed only to generate the result you request — the transcript and, later, the draft — and is not used to train third-party artificial intelligence models.
• Each recording stays linked to its text transcript so you can review it, correct it, and use it to generate the manuscript.

Your recordings are yours. Each audio file, together with its transcript, belongs exclusively to the author who created it, and only that author can listen to it or read it from their account.

• No other author on the platform can access your recordings or even discover they exist: per-owner isolation makes it technically impossible.
• Julia staff do not access the content of your recordings routinely or out of curiosity; only authorized personnel could do so exceptionally, when strictly necessary to provide support, comply with a legal obligation, or investigate abuse, always under confidentiality duties.
• Infrastructure providers (Google's Firebase) and transcription providers (Anthropic) process the data as processors, solely to provide the service and subject to their own security obligations.

In short: apart from you, no one — not another author, not unauthorized staff — can freely access your recordings.

We keep your recordings only for as long as you need them to work on your book. You stay in control and can delete them at any time:

• If you delete an idea, its audio recording and associated transcript are deleted, and your consumed space is recalculated.
• If you delete a book, all of its recordings (outline, ideas, and instructions), their transcripts, and their derivatives are deleted in cascade.
• If you delete your account, all your books are irreversibly deleted and, with them, every one of your recordings and transcripts.
• These deletions affect both the documents in the database and the audio files stored in Firebase Storage.

Once deletion completes, the recordings are no longer available and the action cannot be undone. If the law requires it, we keep only the minimal records needed to meet legal obligations — never the content of your recordings.

We apply technical and organizational measures to protect your recordings against unauthorized access, loss, or alteration. Among others:

• Storage security rules (storage.rules) that restrict reading and writing of each file exclusively to its owner, rejecting any attempt from another account.
• Validation of the file type and size of every upload at the rules level, to block disallowed content.
• Encryption of communications over TLS at all times, and API keys kept in a secrets vault, never exposed to the browser.
• Passwordless authentication through a one-time code sent by SMS, with session revocation when appropriate.

No system is infallible; if we detected a security breach affecting your recordings, we would act in accordance with applicable law and inform you when required.

As the owner of your recordings and personal data, you can, at any time:

• Access and listen to your recordings and review their transcripts from your account.
• Correct the transcribed text when it doesn't faithfully reflect what you dictated.
• Delete specific recordings, entire books, or your whole account, with the cascade deletion described above.
• Withdraw microphone consent at any time, without affecting the lawfulness of prior processing.
• Export the result of your work (the manuscript) as a PDF or as the KDP package to take it with you.

The processing of your recordings relies on your consent and on the performance of the service you request from us. You can find more about legal bases, third parties, and your rights in the Privacy policy.

See the details in the Privacy policy.

We may update this Audio data policy for legal or technical reasons, or as the service evolves. When changes are substantial, we will notify you through reasonable means. The last-updated date at the top indicates the current version; continued use of Julia after it takes effect implies that you are aware of it.

If you have questions about how we handle your recordings or want to exercise your rights, you can write to us through the contact page or to the email address below. We handle privacy inquiries as promptly as reasonably possible.

Email: privacidad@juliawrites.app