Vidleaf

Privacy Policy

Effective 2026-11-07Version 2026-11-07 This English text is the governing version. A Chinese translation is provided for convenience; where the two differ, this English text prevails.

Reader-video associations in this version: Public reader associations cover YouTube, Bilibili, Bilibili International, Douyin and Xiaohongshu videos. When a signed-in account reopens audio or video they transcribed from a local upload, the server saves a same-page association only after confirming that the current page and upload belong to that exact account. Administrators see only “Local upload”; no filename, reader link or transcript is shown or generated. The existing button-recording and account statistics controls apply, and historical pageviews are not backfilled.

Three things first:

  1. We do not collect your phone number, real name or identity documents. Only one registration route stores your email address — "continue with email". Registering with Google does not store an email address, and neither does registering with an invitation token.
  2. Reading a video that has already been processed needs no account. Browsing that way leaves only traffic statistics, which you can switch off, and one cookie holding a random number and nothing about you. When you arrive through a link with campaign labels, the server also adds one to that campaign's hourly count of opens, which holds nothing about you. Default-on reading punctuation also processes long subtitle passages as you read; see §4. Its control is separate from traffic statistics.
  3. When you use transcription or speaker detection, your audio is sent to third-party processors. Alibaba Cloud in mainland China provides the base transcript, default speaker detection and short-lived private object storage; optional GPT correction and verification sends relevant clips to OpenAI in the United States. pyannoteAI in Europe is used only when the explicitly configured legacy speaker fallback is enabled. The same applies to audio you upload.
  4. Traffic statistics and optional account usage statistics are retained long-term without automatic age-based deletion. The retention and deletion of other service data are explained in §6.

On first use with no previous choice, campaign attribution, submission links, account activity statistics, Google Analytics and configured Google Ads measurement are enabled by default and can be disabled separately. Defaults do not override previous refusals, deletion, expired or unverifiable choices. DNT/GPC and the global statistics opt-out still apply; past activity is not backfilled. Agreement checkboxes for signup, audio processing, purchases and refunds start selected and can be unchecked. Consent or an operation is submitted only after the final button is clicked; opening a page does not register, transcribe, pay or refund.

1. What we collect

Optional Google Analytics (separate choice)

This feature is on by default only for first use without a saved choice. Previous choices, including refusals and expired grants, are preserved. Without permission, after refusal, or with DNT/GPC active, we do not load GA4. Withdraw anytime in Privacy & statistics settings. Withdrawal reloads the page to stop running code and clears this browser's GA4 cookies. Requests already sent cannot be recalled, and data already received by Google is not automatically deleted. Declining does not affect the service.

With permission, Google LLC receives page categories, browser language, interface language, referring websites, controlled campaign labels, browser and device information, and observations of read submissions, validation failures, reading results, visible content, searches, exports, copies and server-confirmed sign-ups to measure visits, sessions and conversions. The network connection also exposes your IP address to Google. Page addresses exclude queries, reading content and video identifiers; titles use fixed page categories, and referrers retain only the website origin. Campaign values come from a fixed allowlist; other values become other. We do not send emails, account names, searches, transcripts, notes, attempt or task identifiers or source video links, associate site account IDs, or enable Google Signals or personalized advertising. Page views are sent manually with Enhanced Measurement disabled. The separate Google Ads choice is described below.

For production orders created after this permission, a purchase event is sent only when our server returns confirmed payment and fulfilled credits. It includes the confirmed amount, currency and an irreversible deduplication identifier derived from the random order ID. Original order IDs, payment credentials and account details are not sent. This browser stores at most 256 deduplication identifiers, cleared on withdrawal or permission expiry. Historical orders, sandbox payments, unverified payments and refunded states are not backfilled as purchases; refunds are not automatically sent to Google. Receiving exported content does not prove a file was saved. Refusal, blocked scripts, not returning to the site or network failure can cause missing events. GA4 is not a complete business or financial ledger.

The earlier page-only permission does not automatically authorize these business events: a new choice is required, while earlier refusals remain effective. Actions before permission are not backfilled, and missing sources stay unknown. Site statistics and Google Analytics have different permission scopes, so their user, visit and event totals need not match.

This browser stores your separate choice for 90 days. With permission, Google's _ga and _ga_5NRF64W0FZ cookies are configured for 90 days and may refresh on visits. This is not the retention period for Google reports, which follows the GA4 property's retention settings. Activity before permission is not backfilled. Google may process data in other countries; see the Google Privacy Policy. This optional processing starts only with a separate choice and does not expand the uses covered by earlier choices.

Category What it is
Account Username, display name, an irreversible password hash, sign-up time, active flag, plan
Email address (email route only) Your email address, used to sign in and to receive the verification code at registration. The other two registration routes store no email address.
Third-party sign-in identifier (Google only) The identifier Google issues for you on this site. It contains no email address and no picture, and it is irreversible — we cannot derive your email address from it.
Verification-code records (email route only) The email address, sha256(address + code), the number of attempts made, whether it was used, and its expiry. The code itself is never written to the database.
Quota grants Kind, amount, source and time of sign-up bonuses and top-ups
Your content Video links you submit or audio files you upload (when you choose a video, only the audio track extracted on your device is uploaded; the video picture never leaves your device); subtitles retrieved or transcribed; translations; summaries; speaker labels; labels you write; your searches. Original uploads and processed results have different retention, described in §§3 and 6
Jobs Status, time, duration and failure reason for each submission
Usage and cost Type, duration, count and provider-reported cost of each third-party call, used by the operator for cost accounting and quota metering
Sign-in records The time and method of each sign-in (password / Google / email sign-up), success or failure, IP address and approximate country, region and city. We query a local IP database, without sending your IP to a geolocation provider. We do not collect precise location or browser details. A failure is recorded only for an existing account; mistyped identifiers are not stored. These security records are visible to you and administrators, retained long-term without automatic age-based deletion, and cleared when the account is deleted. Older records remain uncollected; unavailable locations show as unknown. VPN or proxy exit locations may differ from your actual location. Geolocation data: DB-IP, licensed under CC BY 4.0.
Traffic statistics A pseudonymous browser identifier, page category, referring host name and active time (see §7); new visits also record public IP addresses and supported button actions by default, unless already declined; each can be turned off separately
Campaign arrival counts How many times pages were opened from links with controlled campaign labels, summed per hour and normalized label combination; no cookie, IP address, visitor or visit ID (see “Campaign arrival counts” below in this section)
Purchase funnel counts and blocked-quote diagnostics How many times signed-in users opened a processing confirmation or the plans page, or submitted a task using purchased credits, summed per hour and fixed category; ordinary counts contain no account, video or amount. A paid-source quote blocked by another limit also records the account, feature, fixed reason and time, without video, amount or order ID (see “Purchase funnel counts” below in this section)
IP and email address for rate limiting Rate-limit counters keep only the first 32 hex characters of a SHA-256 digest of the IP address or email in Redis. These counter keys expire after 120 seconds and are not written to the database. They are separate from the sign-in security records above.
Notify-me address Only when you type it into the “Top-up pack · in the works” box: the email address and the interface language. Used only to send one message when that feature opens; not linked to an account, not used for anything else; email us to have it removed
Collection counters while signed out A hash of your IP (IPv6 grouped to /64) as the key for "how many new videos you can still process today". It expires after 26 hours and is never written to the database.

Campaign arrival counts

When a page of this site is opened from an address with controlled campaign labels (utm_source, utm_medium, utm_campaign, utm_content), the server adds one to the current hour's count for the normalized label combination (only the fixed values; anything else is counted as other). Browser navigations are counted separately from other requests such as link previews, crawlers and prefetch. This count is kept for every visitor by default. It sets or reads no cookie, writes nothing to browser storage, and stores no IP address, visitor or visit ID, full address or other query parameter. It shows only how many times each campaign link was opened, cannot be tied to a person, and is not linked to traffic statistics or to the optional campaign attribution records below. Requests with a DNT or GPC signal are not counted; because the counts hold nothing about anyone, they cannot be looked up or deleted per person. The hourly counts are kept long-term without automatic age-based deletion.

Purchase funnel counts

When a signed-in user opens a processing confirmation (to review the estimated use), opens the “Plans & purchase” page, or submits a task that uses purchased credits or a single pass, the server adds one to the current hour's count, so we can tell whether the free allowances are enough and how many people need to buy more. The counts are summed only by fixed categories: the step, the processing feature (transcription, summary, translation, speaker detection), a reason or outcome category (for example free allowance used up, asset longer than 120 minutes, purchased balance insufficient, first-order price shown), and whether the account is one of this site's internal accounts (administrators and test accounts). Ordinary counts record no account, video, transcript, amount or order ID, and are not linked to traffic statistics, campaign statistics or the optional account activity statistics. If a purchased-source quote is blocked by a duration limit, paused processing, unavailable paid processing, daily capacity, an unsupported speaker-detection source, an expired uploaded original or another fixed server limit, an administrator diagnostic separately records the account, feature, fixed reason and time. It contains no video, transcript, amount or order ID and is not sent to Google or other third-party analytics. So that an account counts only once per day, the server hashes the username one way with a random value created for that day; the hash and the random value are kept in the server cache for 36 hours, after which the counts cannot be tied to anyone. The hourly counts and blocked-quote diagnostics are kept long-term without automatic age-based deletion. Ordinary counts contain no identity and cannot be looked up or deleted per person; deleting an account removes diagnostics still linked to it. The administrator dashboard also sums the number and amount of orders, payments and refunds by time, showing totals only, not accounts.

Optional campaign attribution

When the campaign and conversion option is enabled, our statistics script reads controlled campaign labels (UTM source, medium, campaign and content) and add them to your visit record. First-touch and current 30-minute session labels are kept in sessionStorage for this tab. We do not store full links or arbitrary query parameters, or put these labels in the statistics cookie. Collection stops when statistics are off or DNT/GPC declines it. Normalized campaign records are retained with visits long-term without automatic age-based deletion. Local browser storage also keeps random consent-state identifiers to enforce withdrawal and synchronize tabs; they contain no account information or link content.

After you accept the current policy and both campaign and account activity statistics are enabled, the server may associate the current visit ID and campaign with authorized activity events to summarize reading, search, export and registration by channel. These channel summaries do not show accounts, raw cookies, IP addresses, transcript or note text, or full links. This association is not created when the option is off.

To diagnose browser security policies, we separately keep daily counts of policy categories and controlled source categories for up to 8 days. We do not retain page addresses, script samples, accounts or query parameters from these reports.

When campaign attribution is enabled, the visit may record whether registration succeeded, without an account identifier, to measure signup conversion. Linking a visit to account activity also requires account statistics to be enabled. First-use defaults preserve all previous choices.

When traffic statistics are allowed, about 10% of page loads are sampled for LCP, INP and CLS. Only histogram counts grouped by frontend version and device category are retained, for up to 8 days. No page address, account, metric ID or page content is sent. Reports stop when traffic statistics are off or DNT/GPC declines collection. Fewer than 20 samples are not treated as evidence that a target is met.

Optional video submissions linked to visits

Campaign statistics and “Link my video submissions to my visits (optional)” control the association of future submissions with the current visit. Both start enabled on first use without a previous choice and can be disabled separately. The administrator can view recorded page categories, submission times, video titles and links, and current processing results in order under pseudonymous visitor and visit IDs, to find problems in submitting and reading. Signed-in accounts must also accept the current policy and allow optional account activity statistics. This earlier submission-linking option does not use the anonymous task-access credential as a tracking identifier. This option does not record typed content, pointer movements or IP addresses; visit IP records have their own control below. Local uploads show no file name, link or transcript content.

These links are retained long-term without automatic age-based deletion and are enabled on first use without previous choices. Older campaign choices are preserved. Past submissions are not backfilled and links are never inferred from IP addresses or timing. Disabling this option or all traffic statistics deletes this browser's submission links; DNT/GPC stops recording. Disabling or deleting account activity statistics also clears the account's submission links. Necessary task records remain separate and are not erased by deleting optional links. Visitor IDs distinguish browsers, not real identities; clearing the statistics cookie or using another browser creates a new ID.

Signed-in browsing and task links

While traffic statistics are enabled, new signed-in pageviews are linked to the server-verified account by default. No campaign statistics or separate checkbox is required; existing sign-in sessions are included. Administrators can view page categories, visit times, recorded active time, and task types, submission times and current results explicitly linked to the observed page, to diagnose browsing and processing problems. This does not enable campaign statistics or optional account activity statistics. Visit IP addresses, button clicks and action results keep their separate controls.

The server verifies identity for each new page event. Signing in, switching accounts or signing out starts a new page record; earlier records are never assigned to a later account using the same browser. Historical browsing and task links are not backfilled, and identity is never inferred from IP addresses, nearby times or shared videos. Missing evidence remains unknown or unlinked. Raw cookies, sign-in credentials, typed content and local filenames are not displayed. Administrator and known internal test visits are excluded from account page history.

Links are kept long-term without automatic age-based deletion. Turning off all traffic statistics or sending DNT/GPC stops collection and removes this browser’s links; deleting an account removes its links. The account statistics deletion control can still remove existing account links; later pageviews follow the current traffic statistics setting. Turning off campaign statistics does not stop this default recording. Necessary task, usage and sign-in security records retain their own rules and are not erased by deleting browsing links.

Guest tasks use the separate guest activity setting below and verified task credentials. G codes identify task-access credentials and V codes identify browser visits; neither proves a natural person’s identity or backfills historical links.

Guest tasks, IPs and clicks together

From 6 October 2026, new visits with basic traffic statistics enabled link future guest tasks to the exact observed page and visit using a server-verified, pseudonymous reference to this browser’s guest task credential. The administrator can view tasks, results, visit IPs, approximate regions and supported button clicks together under the same G code. IP and click recording retain their own switches. Neither IP nor timing is used to guess ownership. Raw credentials and cookies are never displayed.

Guest linking is on by default for new visits, independent of campaign statistics and signed-in account browsing links. Earlier refusals remain off. In Traffic statistics, turn off “Link guest tasks to visits” to stop future guest links and delete this browser’s existing guest links; necessary task records and independently enabled IP and click records remain, without guest association. Turning off all traffic statistics or sending DNT/GPC stops collection and removes these browser links. The links are kept long-term without automatic age-based deletion, historical browsing and tasks are not backfilled, and signed-in account links follow the basic traffic statistics rules above.

From 15 October 2026, basic button recording also controls public-video context on reader pages. After a public YouTube or Bilibili video loads in subtitles, summary or search, its video identifier and caption language may be saved for that exact pageview. Administrators can see its title and open the reader. Local upload names, input and transcript text are excluded; video context is not sent to third-party analytics. Turning off button recording deletes this browser’s video associations. Turning off all traffic statistics or sending DNT/GPC stops collection. Older views without a saved identifier show “Video not recorded”; nearby tasks or times are never used to guess the video. Video-association diagnostics also save the fixed collector and reader protocol versions reported for that view and a controlled result category, such as not reported, context error, unsupported source or missing video metadata. They contain no device fingerprint, full URL, filename or raw error text. The same button-recording choice controls these diagnostics and deletes them with video associations. Missing historical diagnostics remain unknown and do not establish that an old page was used.

IP and time diagnostic matching (updated 7 October 2026)

To diagnose account tasks separated from browsing records, administrators can automatically view saved pages and button clicks with the same IP within 10 minutes before or after a successful sign-in, including historical records that are still stored. Candidates share the account’s visit list with confirmed activity, and each inferred record is labelled “Inferred association”, with original visit IDs, sign-in times and account task IDs in the same window. Original ownership and confirmed account visit/click totals are unchanged. A shared IP does not establish a person's identity. Multiple accounts or candidate visits are marked ambiguous; visits explicitly owned by another account are excluded.

This diagnostic uses existing sign-in IPs and browsing records only. Missing data is not collected retrospectively and inferred links are not stored. IPs are page-opening snapshots, not locations measured at each click. Turning off “Record visit IP” or “Link guest tasks to visits” excludes this browser from matching. Disabling or deleting optional account activity statistics excludes the account. Turning off click recording deletes existing clicks and removes them from candidates. Turning off all traffic statistics or withdrawing through DNT/GPC also excludes the browser. The optional-link sections' no-backfill rules continue to apply to confirmed identities and stored links; this separately labelled historical candidate view replaces earlier statements that no IP/time comparison is performed at all. Sign-in IPs are also used for this diagnostic; retention periods are unchanged.

Optional account activity statistics

Separate from traffic records, optional account activity statistics use a server-generated HMAC account reference, a minimal video reference, event and operation IDs, controlled action categories, timestamps and a server-assigned subject category. These references are pseudonymous and can link activity within an account; they are not anonymous. No raw video URL, search text, transcript or notes are collected. Authorized visits and account activity may be associated by visit ID only while account activity statistics are enabled.

Purpose: understand whether reading, quotation delivery and later reuse work. Export delivery means a browser download was initiated or text was copied; it does not prove a file was saved or used. Collection begins only after agreement to the current policy and an enabled account analytics preference. On first use after accepting the current policy, only a missing preference is initialized to enabled; a saved refusal or deletion is preserved. Turning off traffic statistics also disables account activity collection. DNT and GPC are respected.

Detailed optional events are deleted under the deletion controls below. You may disable collection or delete these events from the statistics controls. Required usage and billing records are separate and are not erased by this optional analytics control. There is no waiting period for this initial rollout. First-use defaults do not override saved choices, and past activity is not backfilled.

Button and action-result statistics

New visits record later activations of supported buttons by default as part of basic traffic statistics, to help find problems in reading, translation, summaries, search and export. You can turn this off separately in Traffic statistics at the page footer. Campaign and account identity linking are not required. Existing explicit refusals remain off, and past clicks are not backfilled. Default recording is not recorded as your consent to campaign statistics.

Button-click records contain only fixed button and page categories, interface language, device category, server receipt time, and pseudonymous browser and visit identifiers. They do not contain input values, user content inside buttons, full URLs, query parameters, transcripts or filenames. A click means a button was activated, not that an operation succeeded or its result was saved. Administrators can see rankings and individual records. Signed-in account names follow basic traffic statistics by default and require server verification of the current page identity; earlier browser activity is not assigned to the last signed-in account. Visit IP records have the separate control below.

The same action-recording choice also controls submitted subtitle-search content. An input snapshot is recorded only when you click Search or submit with Enter, never while typing without submitting. Searches with matches, no matches, failures and browser validation refusals can be recorded for administrators to inspect alongside results from the exact same action. Each snapshot retains up to the first 2,048 characters, its original character count and whether it was truncated; links within submitted text are included in that snapshot. Only administrators can read these records. Search content is not sent to Google Analytics or advertising platforms. Missing results remain unknown, not assumed failures, and historical missing inputs are not backfilled. Search records use the same retention, disabling and deletion rules as button and action-result records below.

From 8 October 2026, the same recording control also covers separate browser observations of reading, search, export and copy results: readable or empty captions, matches or no matches, fixed failure categories, export content received, a download handed to the browser, and a completed clipboard write. Each observation contains a fixed result category, its page association and server receipt time; a random per-action identifier is hashed for duplicate detection. Failure diagnostics may include a fixed error code, stage, HTTP status and request ID. From 16 October 2026, read submissions rejected with HTTP 400/422 may also record the recognizable HTTP(S) input link from that request for administrator troubleshooting. Credentials, fragments and query parameters other than valid video IDs or Bilibili part numbers are removed. From 21 October 2026, attempted read submissions rejected by browser validation or HTTP 400/422 also retain a sanitized snapshot of the submitted input, even when no link can be extracted. This includes submitted words, share text, whitespace and malformed addresses, up to 2,048 sanitized Unicode characters, with original character count, truncation and redaction flags. URL credentials, queries, fragments and explicitly labeled credentials are redacted before collection and checked again by the server; invisible controls are escaped. Only the exact click or Enter submission is captured, never unsubmitted typing or subsequent edits. Only administrators can read this content; it is not sent to Google Analytics or advertising platforms. This does not automatically collect transcripts, local filenames or account identifiers. Links follow the same recording and deletion controls as the failure records. These observations are separate from button clicks and server task states. Receiving export content does not prove a download, and handing a download to the browser does not confirm a file was saved to disk. Results are browser-reported and may be missing; missing records do not mean failure. No historical result records are backfilled. Historical input links are not backfilled.

Click and action-result records are kept long-term without automatic age-based deletion. Turning off button and action-result recording or all traffic statistics stops collection and deletes this browser's earlier click and result records. DNT/GPC also stops collection and clears them. Existing explicit refusals continue to apply to both. Disabling campaign statistics does not disable this independent recording. Turning off all traffic statistics removes the corresponding username links; click and result records can also be disabled separately. Deleting an account removes click and result records still explicitly linked to it.

Reading-flow measurements use the same button and action-result control: a random attempt identifier (stored as a hash), retry links, fixed stages and reasons, prompt exposure/continuation/closure, browser-observed elapsed time, page visibility and instrumentation version. These distinguish input, submission, missing subtitles, transcription, displayed content and restored results. This tab may keep continuation metadata in sessionStorage, without input content or account names, for resuming an attempt within 24 hours; the browser removes it when the tab closes. Server completion uses only an existing, permitted and verified task association; unavailable associations remain unknown. Returns use pseudonymous browser records, without inferring cross-device identity. These new records contain no input text, transcripts or local filenames, and old stages are not reconstructed. Turning off this control or all statistics stops and erases these browser records; DNT/GPC also applies. Server records follow the long-term retention and deletion rules above.

Visit IP records

New visits record the public IP received by the server for new pageviews by default as part of basic traffic statistics. You can turn this off separately in Traffic statistics at the page footer; campaign and account identity linking are not required. Existing explicit refusals remain off, and past visits are not backfilled. Only the server extracts the IP from a trusted network entry point; the browser sends no IP field. Unreliable or non-public addresses appear as unknown, and internal proxy addresses are not stored. When administrators view records, the current local IP database estimates the approximate country, region and city for each IP. We do not send IP addresses to a third-party lookup service, collect precise location, or use an IP address alone to confirm an account identity or rewrite visit ownership. Geolocation data: DB-IP, licensed under CC BY 4.0.

Administrators can see the IP for that pageview alongside button records unless those are disabled. Signed-in account names are linked by default under traffic statistics after server verification of the current page identity. These records can be viewed together and are linkable personal data, not anonymous data. Shared networks, VPNs and proxies can give several people the same IP; an IP does not prove identity or precise location. These records help diagnose problems in visiting and using the site and are kept long-term without automatic age-based deletion.

Turning off IP recording or all traffic statistics stops collection and clears this browser's earlier visit IP records. DNT/GPC also stops collection and clears them. Disabling campaign statistics does not disable independent IP recording. After an account change, each new page uses the current verified session without relabelling earlier pages. Visit IP records continue to follow the independent IP control. Deleting an account clears visit IP records still explicitly linked to it. Necessary sign-in security IP records remain separate and are not controlled by this traffic statistics choice.

People estimates: When visit IP records, button recording and guest association are enabled, reports may use already recorded IPs, action times, device categories, entry pages and outcomes for the same public video (including its specific part) to estimate whether sequential retries under different browser identifiers belong to one visitor. A shared IP alone does not merge visitors. Account conflicts, concurrent activity, competing candidates or insufficient evidence keep them separate. Results are labeled estimates; original browser counts, task access and visit ownership remain unchanged. No permanent identity link or additional device fingerprint is created. Disabling the relevant records, DNT/GPC or erasure removes cleared evidence from this estimate.

2. What we do not collect

Payment, entitlement and refund records

When paid products open, we store account-linked order IDs, product and accepted-policy snapshots, the displayed policy language, amounts and currencies, provider payment and refund IDs, payment times, and entitlement reservations and delivery records. A single pass also retains its bound asset identifier and deadlines. These records support fulfillment, payment reconciliation and after-sales service, independently of optional analytics preferences. Waffo checkout and its payment processors handle information needed for payment; Vidleaf does not receive or store full card details. The payment channel receives store, product and random order references, not your transcripts, audio or notes.

Phone number, real name, identity documents, payment card details, precise location, raw User-Agent, contacts, device identifiers. Device class (mobile / desktop) for traffic statistics is decided in your browser and the raw information is not sent back.

Your email address needs saying separately, because it depends on how you registered:

How you registered Do we store your email address
With email Yes. It is your sign-in identifier
With Google No. We hold only the irreversible identifier Google issues
With an invitation token No. That route has no email field

In other words, you can use every feature without ever giving us your email address: register with Google.

3. What is yours alone, and what is shared across the site

This one directly affects what you can delete, so it gets its own section.

Yours alone (invisible to other users): account details, the labels you write, your job list, your usage records, and transcripts, translations and summaries produced from your local audio uploads. Labels are stored per account; different users' labels on the same video cannot see each other.

Shared cache (only for public Bilibili and YouTube videos, keyed per video, not per user): video metadata, subtitle tracks, translations, summaries. Local uploads are keyed by account; two accounts uploading the same file cannot read each other's result. Removing an item from “My tasks” hides the history entry but does not delete a saved upload transcript. Contact support under §8 to request deletion of a processed result.

Three consequences, all stated:

  1. Others will use results you generated. Another user opening the same video reads the existing subtitles, translation and summary rather than paying to generate them again. This is the main reason we can keep costs down.
  2. You cannot delete these unilaterally. They no longer correspond to your account alone. To remove them, use the takedown route in Terms §12 — which applies to the whole video and therefore affects other users too.
  3. The interface does not show who submitted a video, and other users cannot see your account. But the database records which job first fetched a video, and that job traces back to an account; the operator can also see per-account call records for cost accounting and per-account sign-in records. The operator can also see, in the admin dashboard, each submitted video link with its title, duration and processing result, and whether it was submitted while signed in (a submission made without signing in is normally shown as “Guest”; a pseudonymous visitor ID is also shown after separate permission for the optional linking described below), to investigate failures, handle complaints and understand how the site is used; for a local upload only “Local upload”, its duration and the result are shown, never the file name. From 9 October 2026 (Beijing time), a submission made while signed in also shows the account's username. So the accurate statement is: anonymous to other users, not anonymous to the operator.

Shared does not mean public. A common pattern among similar products is to publish each video's AI summary on a sign-in-free page that search engines index, reusing the original video's title. Vidleaf does not:

Common pattern Vidleaf
Reuse across users Yes Yes
Result page public Sign-in-free URL No separate public result page; saved public-video results can be opened through a reader link under guest rules
Indexed by search engines Yes No
Open to AI crawlers for training Commonly yes No
Cache reuse disclosed Usually not This section
Takedown route for rights holders Often none Terms §12

4. Third-party processors

To do what you ask, the following providers receive the corresponding content. Requests carry no account name and no internal user identifier.

AI punctuation for reading is on by default. As you read long passages without sentence breaks, Vidleaf sends small portions of existing subtitle text to OpenAI's gpt-6-luna text model to add punctuation and paragraph breaks for display. No audio is sent for this feature. Vidleaf pays for it without using your free or purchased account allowances. The original text and timestamps are preserved, and reading results are cached for the same subtitle version. Local-upload results retain their existing account access restrictions. Uncheck “AI punctuation” in the transcript reader to stop further requests and return to the original; requests already sent cannot be withdrawn. Search, copy and exports do not use this reading layer, and existing translation choices are unchanged. This control is separate from traffic statistics and audio-processing consent.

Processor Purpose Receives Destination Contact
Alibaba Cloud Base speech transcription; short-lived private object storage Original audio, transcoded MP3 segments or copied m4a/AAC segments are first written to private Alibaba Cloud object storage, then sent to Qwen Filetrans Mainland China (Beijing) aliyun.com/policy/privacy
Alibaba Cloud Speaker detection; short-lived private object storage The complete audio file is first written to private Alibaba Cloud object storage, then sent to Qwen Filetrans for sentence-level anonymous speaker IDs; no real-name or cross-video identity database is created Mainland China (Beijing) aliyun.com/policy/privacy
OpenAI GPT correction and timing verification; summaries, translation and AI punctuation Audio segments for GPT text correction and subtitle-interval verification; subtitle text in segments for summaries, translation and punctuation United States and other regions openai.com/privacy · OpenAI API data controls
pyannoteAI Legacy speaker-detection fallback Only when the legacy fallback is explicitly enabled: the complete audio file (uploaded to their object storage, then submitted as a job) Europe pyannote.ai/privacy-policy · retention at docs.pyannote.ai/data-retention
Bilibili Reading subtitle lists Video link plus a server-held account session (see Terms §8) Mainland China bilibili.com/blackboard/privacy-pc.html
YouTube Reading subtitles The video link only United States policies.google.com/privacy
Mailtrap (Railsware Products Studio LLC, California) Sending the registration verification code Your email address, and the subject and body of that message — the body contains the six-digit code United States mailtrap.io/privacy
Google (Google LLC) Google sign-in When you press "Continue with Google", your browser talks to Google directly: Google sees your IP address, your browser information and the fact that you are on our sign-in panel. Our server only fetches Google's public signing keys, and that request carries nothing about you United States policies.google.com/privacy

Only registration and sign-in involve the last two. Reading subtitles, transcription, translation and summaries never go near them.

Three things worth stating about the Google row:

  1. The script is loaded on demand. Google's sign-in button script (accounts.google.com/gsi/client) is fetched only when you open the sign-in panel and Google sign-in is actually enabled. This does not enable the separately optional advertising measurement described in section 1.
  2. It will not sign you in by itself. Automatic account selection is turned off when the button is initialised, so there is no "already signed in with some Google account" — you have to press it.
  3. We take one identifier and nothing else. What Google returns to us is a signed token; from it we read only the per-site user identifier and a display name. We neither read nor store the email address or the picture.

About the Mailtrap row: it receives your address only if you choose "continue with email", and only that one verification message. It does not receive your password, your content or your usage. How long it keeps that message is in §5.

This is the current default configuration. If you use BYOK, content goes to the provider you configure, governed by your agreement with them.

Exercising your rights with these providers: contact us first (§12) and we will pass on a request to access, correct or delete; or contact them directly using the details above. We will tell you the outcome.

Changing or adding a processor is announced on the site 10 days in advance. If you do not accept a new processor, stop using the corresponding feature or close your account. The immediate security-response exception in section 11 applies to necessary anti-abuse verification; audio-provider changes still follow the advance notice rule.

Third-party audio processing, stated plainly: Alibaba Cloud receives base-transcription segments in mainland China, optional GPT processing may send clips to OpenAI. Default speaker detection also uses Alibaba Cloud Qwen in mainland China; pyannoteAI is used only when the explicitly configured legacy fallback is enabled. If you do not want a particular piece of audio sent to these processors, do not use those features; reading existing subtitles sends no audio. For uploaded audio there is no "subtitles only" option, so confirming transcription necessarily sends it to the processors listed above.

Email security verification — Cloudflare Turnstile

When you request a signup code or a password reset, your browser connects to Cloudflare, Inc. to distinguish people from bots. Cloudflare receives browser security signals including your IP address, TLS fingerprint, User-Agent, site key and page origin; processing may occur outside your country or region. Cloudflare also uses signals to improve bot detection, as described in its Turnstile Privacy Addendum.

Vidleaf loads verification only when you request one of these email actions, after the signup terms step where applicable. Ordinary browsing and password sign-in do not load it. Our server sends the verification token and site secret to Cloudflare; it does not include your email, password or a separate IP field. We do not store challenge tokens or provider responses. Cloudflare retention follows its own policies. This necessary email-security check is separate from optional analytics; cancelling it sends no email. The affected form links to the verification notice.

5. How long the providers keep it

We do not know, and we cannot control it. That depends on each provider's own policy and is outside this service. You can read their privacy policies directly. One point deserves separate mention:

Alibaba Cloud receives audio and writes it to private object storage. Base transcription uses audio segments; default speaker detection uses the complete audio file. Objects use short-lived signed access, encryption and private ACLs; we request deletion after success, failure or cancellation, while unknown results and faults may delay cleanup. Qwen performs base transcription and sentence-level speaker detection in mainland China. Vidleaf does not use OpenRouter or Whisper for new base transcription.

OpenAI receives subtitle text and no additional audio for these text features. OpenAI states that API inputs and outputs are not used to train its models by default. OpenAI's data controls also describe limited retention for abuse monitoring and applicable exceptions; see the linked controls for the current period and scope. Requests to it carry no account name.

Only when the explicitly configured legacy fallback is enabled, pyannoteAI receives the complete audio file. For this one we went and read their documentation, so we can give you the actual numbers rather than "it depends on them":

What they hold How long
The uploaded audio file Deleted automatically within 48 hours, whether or not a job used it
Job results (the speaker segments) 24 hours after the job completes
Audio on the processing servers Deleted as soon as processing ends

The source is docs.pyannote.ai/data-retention, which says "automatically deleted within 48 hours".

We have no way to call a "delete" on their side, because they do not offer one — their design is to expire it on a timer rather than have callers do it. An earlier version of this text said we could not delete it and called that the most worthwhile thing to improve; that was inferred from our own request allowlist without reading their documentation, and it was inaccurate. This is the corrected version.

One more of the same kind, also stated as it is: Mailtrap keeps the verification message, body included, in its sending log, and that body contains the six-digit code. How long depends on its billing tier — per its public pricing page, 3 days on the free tier, 5 on Basic, 15 on Business, 30 on Enterprise.

The practical risk here is smaller than it first looks, and the reason is worth stating: the code itself expires in 10 minutes, so the code sitting in that log is already dead long before anyone could use it. What the log does retain is the fact that this address received a registration message from us at a given time. We state it here anyway, rather than leaving you to go and read a pricing page.

6. How long we keep it

Data Retention After you ask us to delete
Traffic statistics Long-term, without automatic age-based deletion Turning collection off stops new records
Button click records Long-term; only new clicks while enabled are recorded; withdrawal and deletion remain available Disabling button recording or all traffic statistics deletes this browser's click records. Account deletion removes records still explicitly linked to that account
Campaign arrival counts Summed per hour; kept long-term without automatic expiry They hold nothing about anyone, so they cannot be deleted per person; requests with DNT / GPC are not counted
Purchase funnel counts and blocked-quote diagnostics Ordinary counts are summed per hour and kept long-term without automatic expiry; the hashes and random values used to count an account once a day are deleted after 36 hours. Account-level blocked-quote diagnostics are kept long-term Ordinary counts hold nothing about anyone; deleting an account removes its linked blocked-quote diagnostics
Sign-in records Long-term; cleared when the account is deleted Deleted when the account is closed
Login sessions 30 days with "remember me", otherwise 12 hours, deleted on expiry Deleted on sign-out
Rate-limit IP hash 120 seconds —
Cached results Redis hot cache: 7 days; saved database results are not automatically deleted when that cache expires —
Job state 24 hours —
Temporary audio on our servers Deleted as soon as processing ends; it exists only in a temporary directory —
Audio files you upload Kept for at most around six hours after upload, whether or not it was submitted or transcribed: a successfully transcribed original stays available during that window for speaker detection (which needs the complete audio) and is then cleared by a scan every minute; an original whose transcription failed or was cancelled is deleted at once, and failures are logged and retried. Uploading the same file again restarts the window from that upload; a file already transcribed is not charged again. An outage or storage fault may delay cleanup; a recovery scan follows. The original lives only in a temporary server directory, not in the database or backups. Processed text is stored separately; see §3 —
Your email address (email route) For as long as your account exists Deleted on closure; a copy may persist in offsite backups for up to about four weeks (see below)
Google sign-in identifier For as long as your account exists As above
Verification-code records They expire after 10 minutes; expired rows are deleted the next time anyone requests a code —
Quota grants For as long as your account exists Deleted on closure
Collection counters while signed out (IP hash) 26 hours —
Your labels, job records, usage records For as long as your account exists Deleted on closure; a copy may persist in offsite backups for up to about four weeks (see below)
Subtitles, translations, summaries (shared cache) For as long as your account exists; we set no deletion date See §3: not deletable unilaterally; use Terms §12

The last row, stated plainly: Vidleaf does not run periodic deletion of business data. Video metadata, subtitles, translations, summaries, labels, job records and call and cost records are kept for as long as your account exists. We would rather not give you a deadline we cannot meet. To have it deleted, use §8.

About "up to about four weeks in offsite backups": the offsite backup keeps, independently, the latest snapshot of each of the newest seven calendar days and the latest snapshot of each of the newest four calendar weeks (deploy/backup/README.md). It deliberately accepts no delete command from the main site — otherwise compromising the main site would take the backups with it. The cost of that choice is exactly this: after we delete your data from the live database, it still exists in those snapshots until newer ones push them out. The ceiling is therefore the age of the oldest weekly snapshot, about four weeks.

Why the verification-code row is worded so awkwardly: the cleanup happens as part of issuing a code, not on a timer of its own. So the accurate statement is "expires after 10 minutes and is deleted the next time anyone requests a code", not "deleted after 10 minutes". Through a stretch with no registrations at all, the last expired row sits there. Writing "deleted after 10 minutes" would be a falsehood — and it is exactly the kind of falsehood that ends up in a privacy policy.

7. Cookies

The following three categories are our own cookies and are not shared with third parties. Separately permitted Google Analytics and Google Ads cookies are described in the optional measurement notices in section 1.

Cookie Purpose Attributes Lifetime
Session Keeps you signed in HttpOnly, Secure over HTTPS, SameSite=Lax 30 days / 12 hours
Traffic statistics Distinguishes browsers with a pseudonymous identifier HttpOnly, Secure, SameSite=Lax 90 days
Signed-out access A random number whose only purpose is to let a job you submitted while signed out be read back by you HttpOnly, Secure over HTTPS, SameSite=Lax 30 days

8. Your rights, and how to exercise them today

Stated as it is, without promising what we cannot do:

What you want Possible today How
Change your password Yes My account → change password (an account created with Google has no password, so this does not apply)
Reset a forgotten password yourself Yes, for email accounts Sign-in page → "Forgot password"; we send that address a reset link. Not for Google or invitation accounts: a Google account has no password, and for an invitation account we hold no email address. Email support@vidleaf.app in those cases
Change the email address on your account Manual As above
Unlink Google and switch to a password Manual As above. An account created with Google has no password, so one has to be set first
Remove a video from "My tasks" Yes But this hides, it does not delete — the data stays and simply stops being shown to you
Turn off traffic statistics Yes Footer → "Traffic statistics"
Export your subtitles, translations, summaries and labels Yes "Export all" on the reading page
Delete all data for a video Manual Email support@vidleaf.app
Close your account and delete everything Manual As above
See what we hold about you Manual As above

What we commit to: deletion and closure requests are completed within 15 working days of receipt — your account, labels, job records and usage records deleted, and anything that cannot be deleted anonymised. The public-video shared cache in §3 does not correspond to your account alone; use Terms §12 for its takedown. Local-upload results are account-isolated and included in a manual deletion request. Hiding an item in “My tasks” does not delete the result.

A self-service route is being built. Until it ships, the email route above is the official one and we will not delay or refuse because there is no button.

9. Security

10. Minors

Vidleaf is not directed to children under 14, and we do not knowingly collect their information. If you are a guardian and find that a child has registered, contact us and we will delete it.

11. Changes to this Policy

Version 2026-11-07: Default speaker detection moves from pyannoteAI in Europe to Alibaba Cloud Qwen Filetrans in mainland China (Beijing). The model returns sentence-level anonymous speaker IDs, which are applied directly to speaker labels; no real-name or cross-video identity database is created. The complete audio file is written to private Alibaba Cloud object storage and deleted after success, failure or cancellation, with unknown results and faults possibly delaying cleanup. pyannoteAI remains only for the explicitly configured legacy fallback. The audio-processing notice moves to 2026-11-07 and requires fresh confirmation before new transcription or speaker detection.

Version 2026-11-05.1: All base speech transcription moves from OpenRouter whisper-large-v3 to Alibaba Cloud Qwen Filetrans. Audio segments use short-lived private Alibaba Cloud object storage and are deleted after the task. OpenAI remains only for optional GPT correction and timing verification, while pyannoteAI remains the speaker-detection processor. The audio-processing notice is updated and asks for confirmation again before new transcription or speaker detection.
Version 2026-11-04.1:Summaries, translation and reading AI punctuation now use OpenAI gpt-6-luna Fast instead of DeepSeek. The same subtitle or text-to-process content is sent without an account name, and free or purchased allowances are unchanged. Audio recipients and the separate audio-export consent are unchanged.
Version 2026-10-22.1:Purchase-funnel counts now include account-level paid-quote block diagnostics. Only when a signed-in user selects purchased S/L credits or a single pass and the server cannot start because of a duration limit, paused processing, unavailable paid processing, daily capacity, an unsupported speaker-detection source, an expired uploaded original or another fixed server limit may administrators view the account, feature, reason and time. Records exclude video, transcript, amount and order ID and are not sent to third-party analytics. Earlier anonymous “blocked for another reason” counts cannot be reconstructed. Deleting an account removes linked diagnostics.
Version 2026-10-21.1:On first use with no previous choice, campaign attribution, submission links, account activity statistics, Google Analytics and configured Google Ads measurement are enabled by default and can be disabled separately. Defaults do not override previous refusals, deletion, expired or unverifiable choices. DNT/GPC and the global statistics opt-out still apply; past activity is not backfilled. Agreement checkboxes for signup, audio processing, purchases and refunds start selected and can be unchecked. Consent or an operation is submitted only after the final button is clicked; opening a page does not register, transcribe, pay or refund.
Version 2026-10-21:Signed-in browsing and task links now follow traffic statistics by default, including existing sessions, without campaign statistics or a separate checkbox. Analytics opt-out, DNT/GPC, account deletion and per-event server identity checks remain; historical browsing is not backfilled.
Version 2026-10-19: Email signup and password reset now use Cloudflare Turnstile before sending mail to prevent automated abuse. Reset notices are no longer emailed for nonexistent, disabled, deleted or passwordless accounts; the page uses a generic response. This update discloses Cloudflare browser verification and allows security measures necessary to address ongoing abuse to take effect upon publication, with notice in the affected flow. Other policy changes retain their existing notice periods. Audio providers and separate audio-export consent are unchanged.
Version 2026-10-11: Adds default-on, site-funded AI punctuation for long subtitle passages using the existing DeepSeek text model. The original is preserved; the feature can be turned off, uses no account allowance, and does not change copy or export text. Traffic, page history, clicks and action results, campaign attribution, account activity and sign-in security records are retained long-term without automatic age-based deletion, including existing and new records. Hourly campaign-arrival and purchase-funnel counts are also kept long-term. Existing account deletion, statistics deletion and link-withdrawal controls remain available. Deleted or uncollected history cannot be recovered. Cookie, sign-in session, advertising authorization, short-lived deduplication and original-upload lifetimes are unchanged.
Version 2026-10-05: Sign-in security records now include the IP address and approximate country, region and city looked up in a local database, to identify unusual sign-ins and investigate account security. IP addresses are not sent to a geolocation provider; precise location and browser details are not collected. Records are visible to the account owner and administrators and retain the existing 90-day rolling cleanup. Older records are not backfilled, and proxy exit locations may differ from a person's actual location.
Version 2026-10-04.2: Adds a separate, default-off choice to link account identity, browsing and tasks. After this independent choice, only future events record server-verified identity; optional account statistics are not enabled by it. Account changes require a new choice; older grants are not expanded and historical browsing is not backfilled or guessed from IP addresses or times. Optional links are kept for up to 90 days; withdrawal removes the links while necessary task records remain. Earlier video-submission and audio-export consent versions remain separate.
Version 2026-10-04.1: Adds a separate, default-off option to link video submissions to visits. After explicit agreement, the administrator can group future submissions, current processing results and recorded page categories by pseudonymous browser and visit IDs, retained for up to 90 days. Signed-in accounts must also allow account activity statistics. Earlier submissions are not backfilled and older grants are not expanded; disabling the option removes the links. Existing traffic page categories can be grouped by browser and visit without displaying cookies or IP addresses. Known internal activity is excluded from this dashboard. Audio processing and its separate consent version are unchanged.
Version 2026-09-30: §3 now describes the “Video submissions” record in the admin dashboard: the operator can see each submitted video link, title, duration and processing result, and whether it was submitted while signed in (a submission made without signing in is normally shown as “Guest”; a pseudonymous visitor ID is also shown after separate permission for the optional linking described below); local uploads never show the file name. From 9 October 2026 (Beijing time), a submission made while signed in also shows the account's username, at least seven days after this notice. The kinds of data collected, recipients and retention are unchanged, and the separate audio consent version is unchanged.
Version 2026-09-29.1: Adds purchase funnel counts: when a signed-in user opens a processing confirmation or the plans page, or submits a task using purchased credits, the server counts it per hour and fixed category, without recording the account, video or amount; the one-way hashes and random values used to count an account once a day are deleted after 36 hours, and the counts are kept for 90 days. The administrator dashboard also sums orders, payments and refunds by time. Other data collected, recipients and purposes are unchanged, and the separate audio consent version is unchanged.
Version 2026-09-29: Adds campaign arrival counts: when a page is opened from an address with controlled campaign labels, the server counts opens per hour for every visitor, without a cookie, IP address or visitor ID and without linking them to traffic statistics or optional campaign attribution; requests with DNT/GPC are not counted; the counts are kept for 90 days. The optional campaign attribution section now states that the statistics script reads the labels. Other data collected, recipients and purposes are unchanged, and the separate audio consent version is unchanged.
Version 2026-09-28: Adds a separate, default-off campaign and conversion option, explains visit-to-activity links, signup markers and 90-day retention, and adds sanitized security-policy counts retained for 8 days. The audio notice now covers original audio files meeting format and size requirements as well as transcoded 16 kHz mono MP3 segments of up to 15 minutes each. Recipients and processing purposes are unchanged. Separate audio consent moves to 2026-09-28 and must be renewed before the next transcription or speaker detection.
Version 2026-09-27.1: corrects the processor names in §4. The direct recipient for speech transcription is OpenRouter (United States), which routes audio to a model provider (currently DeepInfra, Together AI or Groq); Groq, named before, is only one of the possible providers. Summaries and translation go to DeepSeek (mainland China), previously misstated as SiliconFlow. Since we began recording third-party calls these two have been the actual recipients, so this is a correction of the notice, not a change of provider. §5 adds their retention notes. Because the audio-recipient notice changed, the separate audio-export consent version moves to 2026-09-27: accounts that consented before will be asked once more before the next transcription or speaker detection. Terms of Service §11 is reworded accordingly. The categories of data, purposes and retention periods are unchanged.
Version 2026-09-27: the English Terms of Service and Privacy Policy now quote interface labels as the English interface shows them (for example “AI-generated”, “Human captions”, “Traffic statistics”) instead of quoting the Chinese labels, and their links back to Vidleaf open the English site. The meaning is unchanged: the kinds of data collected, recipients, purposes, retention and prices are the same, and the separate audio consent version is unchanged.
Version 2026-09-25: §6, “Audio files you upload”: a successfully transcribed original is no longer deleted as soon as its job ends; it is kept for at most around six hours after upload so that speaker detection can use it (previously speaker detection was available only for platform videos). Originals whose transcription failed or was cancelled are still deleted at once. The maximum retention is unchanged (a file never submitted was already kept for at most around six hours); the kinds of data collected, recipients and purposes are the same, and the separate audio consent version is unchanged. Terms of Service: the sentence on speaker detection for local uploads is updated accordingly.
Version 2026-09-24.2: Terms of Service: AAC audio in an m4a file and an AAC audio track extracted on your device from a video are no longer subject to the 60 MiB limit if the average bitrate is at most 320 kbit/s; the per-asset duration limit applies instead. Other audio files remain limited to 60 MiB. An m4a over 60 MiB is first checked on your device, and only its AAC audio track is uploaded. This policy itself is unchanged: the kinds of data collected, recipients, purposes and retention are the same, and the separate audio consent version is unchanged.
Version 2026-09-24.1: uploads are no longer limited to mp3, m4a and wav; common audio formats are accepted. When you choose a video file, its audio track is extracted on your device and only the sound is uploaded; the video picture never leaves your device. The kinds of data collected, recipients, purposes and retention are unchanged, and the separate audio consent version is unchanged.
Version 2026-09-24: adds Pack L, conditional first-order prices and cent-based refund allocations; the account area now separates purchases, orders, usage and security. Orders retain the language and policy accepted at purchase. Previous orders keep their original terms. Audio processors and the separate audio consent are unchanged.
Version 2026-09-23.2: the Terms describe prepared S and single-pass products, their validity and a 14-day refund policy based on unused components. Payment and entitlement records are necessary fulfillment records, separate from optional analytics. Audio processors and the separate audio-export consent are unchanged.
Version 2026-09-23.1: corrects the scheduled cleanup and outage wording for original uploads; clarifies account isolation of processed upload results, that hiding history does not delete results, and that expiration of the Redis hot cache does not erase database results. Current terms must be accepted before new cloud work; the separate audio-export consent version is unchanged because processors, sent data and purposes have not changed.
Version 2026-09-22.1: adds the local audio upload path. Section 1 ("Your content") now names audio files you upload, section 6 states their retention (deleted as soon as transcription ends; an upload never submitted is kept at most 6 more hours), and section 4 states that uploaded audio also leaves mainland China. The list of processors, the purposes and the retention periods are unchanged, so the separate audio-export consent keeps its version and does not need to be given again.
Version 2026-09-20.3: removes the waiting period for the initial rollout of optional account activity statistics. You can opt in after this notice is published; collection is off by default and past activity is not backfilled. This pre-launch update replaces the seven-day wait previously announced for this feature.
Version 2026-09-20.2: introduced optional account activity statistics, their separate storage, opt-in, deletion and 90-day retention.
What changed in v1.2 (2026-09-20): section 1 gained one row, “Sign-in records”; section 6 gained the matching retention line (90 days, swept on a rolling basis); and section 3 now states that the operator can also see per-account sign-in records. What is recorded is the time, the method and whether it succeeded — no IP address and no browser details; a failed sign-in is recorded only when the account actually exists.
What changed in v1.1 (2026-09-19): section 1 gained one row, “Notify-me address” — the email you type into the “Top-up pack · in the works” box on the home page.
Changes update the effective date on this page and are announced on the site. Changes that widen the use of your data, add a processor or extend retention are announced at least 7 days before they take effect. An exception applies to security measures necessary to address ongoing abuse: these may take effect when the notice is published, with notice in the affected flow. This exception does not expand optional analytics or change audio-processing recipients.

12. Contact

support@vidleaf.app