Effective date: 2026-10-13
Narrivela is operated by Ian Chen, an individual publisher based in China. Contact yingcong.ian.chen@gmail.com for privacy questions or requests. This policy covers the Narrivela website, web studio and Narrivela tools used through an AI client. It applies to the features actually available to you; a preview or sign-in setup page does not open speech generation or allocate free service credit.
We process account information to sign you in and restrict access to your tasks. Auth0 manages registration, password authentication, email verification and login security. Narrivela verifies a signed access token and uses a stable, hashed account identifier to associate tasks and usage records with your account. This identifier is pseudonymous, not completely anonymous. Narrivela's task and usage databases do not store your password or access token. The studio keeps its access token in the current page's memory, with a short-lived login transaction in browser session storage. It does not save your script in persistent browser storage.
When you request speech, we process the script, your selected voice and language, task progress and generated audio. A script selected from a text or JSONL file is read for the task. When a connected client supplies a file for a requested tool, Narrivela downloads the selected file from that client's file service. Export features may also process the visual material or video you supply. We send the speech text and the necessary voice and language parameters to Alibaba Cloud's Qwen-Audio 3.0 Plus service in Beijing. Public speech generation uses system voices and does not accept delivery-style instructions. Narrivela processes and stores task files on an operator-managed Mac server physically located in China to generate, resume, assemble and deliver the requested results.
We use task and usage records to provide the service, enforce the free allowance and shared budget, prevent abuse and investigate errors. Monetary usage records contain the account hash, request identifier, billing quantity, estimated cost, time, period and pricing version. Character-limit records also contain reservation status, submission fingerprints and task identifiers. These records do not contain the script or audio. We keep this history while operating the one-time allowance programme, because expiry or deletion of audio must not restore an already used allowance. These records have no automatic task-based expiry; you may ask us to review their retention or deletion separately.
Cloudflare provides website hosting and the public connection to the application, and processes the network and request information needed to provide those services. Auth0 uses a US tenant. Resend sends account verification and password recovery emails through the North Virginia, US sending region, and processes recipient addresses, email contents including account-action links, and delivery records. Support messages sent to our Gmail address are handled through Google's email service and kept for handling your request and any necessary follow-up. These service regions describe the current configuration, not a guarantee that every copy, operational record or supplier support activity stays in one country. Data may be processed outside your country. The service providers' own retention and processing rules apply to the information they hold.
If you use Narrivela through ChatGPT, Codex or another client, that client separately handles your conversation, selected files and tool results under its own policies and account settings. Narrivela cannot erase the client's conversation history or copies you download or share.
Completed speech tasks and their working files normally remain available for approximately 24 hours. Follow the expiry time displayed for your result. Unfinished tasks are subject to cleanup after seven days from creation; failed or cancelled tasks may expire sooner. A separately created HTML or video export has its own 24-hour period from creation. Download links are short-lived access links and can expire before the file does. Anyone holding a valid link may access its result, so keep the link private and download your results while available. Expired files are removed when cleanup runs; work in progress, a stopped server or an interrupted cleanup can delay physical removal.
Internal caches can contain audio and script fragments. Cache files become eligible for cleanup 24 hours after their last update; cache cleanup pauses while speech work or exports are active. Deleting a task removes its working files and associated published exports after active work stops, but does not immediately erase shared cache copies, usage history, logs or backups. Local log archives are retained for 14 days after archiving and valid backup snapshots for seven days after creation, with expiry enforced during maintenance. Current logs are rotated during maintenance rather than deleted by each entry's age. A failed maintenance operation or an unrecognised or damaged archive may need manual review before it can be removed. Backups may include task files, caches and usage records and have restricted operator access.
Narrivela uses account access checks, short-lived signed result links and restricted local file access to protect service data. The checked application code does not intentionally log scripts or login tokens in normal task error messages; infrastructure and unexpected errors can still create diagnostic records. The site does not include advertising or third-party analytics scripts in the current application code. Authentication and infrastructure providers may use cookies or similar technologies for their services. We do not use your scripts or generated audio to train models or publish them as public examples without your permission. Supplier data handling is described in their disclosures; a local 24-hour result period does not delete their separate records.
You may request access, correction, an export of your available Narrivela data, account closure or deletion by emailing yingcong.ian.chen@gmail.com, preferably from your registration address. Include the relevant task identifier or describe the request. Do not send a password, API key, access token or verification/reset link. We will verify that the request relates to your account before handling it. Account deletion is handled manually rather than through a self-service account deletion button. We will explain what we removed, what remains in independent caches, backups or provider systems, and any records retained for allowance administration, an unresolved request or an applicable obligation. You can also request review or deletion of support correspondence. Deleting a task is not an account-deletion request.
We will publish changes with an updated effective date. Material changes to data use, providers or retention will be announced on the website before taking effect; we may also use an account service email where necessary. Privacy questions and concerns can be sent to the contact address above.