Legal and privacy

1. Who we are and what this covers

Blab Translation (the "Service") is operated by FutrixDev ("we", "us", "our"), the publisher of the AI Academic Paper Translator Chrome extension and of the website at blab-translation.com. We are the controller of the personal data described below.

This policy covers three things, which are deliberately kept separate:

• The website — the marketing, tutorial, docs and legal pages you are reading now.
• The extension — the browser software, including its free translation features.
• The account service — sign-in, usage allowances, and comic and PDF page translation at blab-translation.com.

It does not cover the third-party AI endpoints you choose to configure in the extension, the sites you visit while translating, or the Chrome Web Store. Those have their own policies.

2. The two halves of the product

This is the single most important thing to understand about how your data flows, so it comes before the list of what we collect.

2.1 Free translation — nothing reaches us

Selection translation, hover translation, full-page translation, input translation and YouTube subtitle translation are performed by the extension in your browser. It sends the text to be translated directly to the API endpoint you entered in Settings — your own OpenAI-compatible key, your Claude key, an OpenRouter gateway, or a local model such as Ollama or LM Studio. The request does not pass through our servers and we have no copy of it. We do not see the pages you visit, the text you select, the subtitles you translate, your API keys, or your endpoint URLs. No account is required and none of this is logged by us.

Your extension settings — including your API key — are stored by Chrome in chrome.storage.sync, which means Chrome may sync them across the browsers where you are signed into your Google profile. That is a Google feature operating on your own profile; the values are never transmitted to us.

2.2 Comic page translation — this one uses our servers

Redrawing a comic page cannot run in your browser: it needs an image model, credentials and 30–120 seconds of server time. So when you translate a comic page, the extension uploads the image bytes of that page to blab-translation.com, we forward them to a third-party image model to be redrawn in your target language, and we store the redrawn result so your browser can fetch it. This requires an account, draws on your monthly free page allowance, and is described in detail in sections 3 to 7.

2.3 PDF translation — this one uses our servers too

Translating a PDF and rebuilding it as a readable document takes minutes of server time and a text model, so it works the same way. The difference is where the bytes go: a PDF is too large to travel through our application, so your browser uploads the file directly to our private storage bucket using a signed link that expires after 10 minutes. We never see the file in transit.

A container running on Cloudflare then downloads that file, sends its text to a third-party AI gateway to be translated, rebuilds the document, and writes the result — bilingual, translation-only, or both — back to the same private bucket. That container gets its own signed links, valid for the length of the job plus a margin — 45 minutes as configured today — and used by nothing else. You download the result through a signed link that expires within 30 minutes. Unlike a comic page, your source PDF is stored at rest for as long as the job may need it; both the source and the results are deleted automatically 7 days after upload by a storage lifecycle rule (section 7). This also requires an account and draws on your monthly free PDF page allowance.

2.4 Live television captions — this one uses the microphone on your television

Live captions are produced by the app you install on your own Android TV, and there are two paths through it. The no-microphone path: you type the film's title and playback position on your phone or in the web console, the television renders an existing subtitle track, and it never asks for recording permission at all. The live-listening path: the television opens its own microphone and uploads what it hears to us to be recognised and translated.

On the live-listening path the television shows a three-second notice on screen before the microphone opens. It keeps listening after recognition succeeds. Even once the film has been identified and the session has switched to following a subtitle track or the player, the microphone on the television does not close by itself — playback can seek, change film, or run past the end of what the subtitles cover at any moment. To close it you stop the captions yourself: the stop control on the television, or the one in the web console. Quitting the app or ending the session releases it too.

While a session runs, audio is uploaded as a compressed stream to the recognition container we operate. We do not keep the audio. It is recognised into text in memory and discarded; nothing is written to storage. What we do keep is the record of the session (its length, its mode, the film it turned out to be) and the recognised English text windows — and those are indexed against the film, so that a later viewing of the same film can identify it faster, not against your account. Recognition happens in our own container; translating that text into the language you chose is done by a third-party AI gateway, see section 6. A translated subtitle track is only ever rendered into a running session; we do not offer it as a download.

3. What we collect

3.1 Account data

Sign-in is via Google or GitHub OAuth only. There is deliberately no email-and-password option, so we never hold a password for you. From your chosen provider we receive and store: your provider account identifier, your email address, your display name, and your profile image URL if the provider supplies one. We also store session records so you stay signed in.

3.2 Extension link tokens

Connecting the extension to your account mints a bearer token for the extension. We store only the SHA-256 hash of that token, plus a label and timestamps — a copy of our database therefore contains nothing replayable. Tokens expire 90 days after their last use and you can revoke them at any time.

3.3 Translation job records

For each comic page you translate we store a job record containing:

• a job identifier and your account identifier;
• the idempotency key sent by the extension;
• job status and timestamps;
• the URL of the page you were reading, if the extension sends it;
• source and target language;
• the pixel dimensions of the source and the result;
• credits charged, whether free quota was applied, the number of upstream attempts, and which gateway served it;
• an error code and a truncated diagnostic message when a job fails.

For each PDF you translate we store a job record containing:

• a job identifier, your account identifier and the idempotency key sent by the client;
• job status, a progress percentage, the current stage label and timestamps;
• the storage keys of the source file and of each result file;
• the page count, source and target language, and your output choices (bilingual or translation-only, and the bilingual layout);
• the number of model tokens the job used, the credits charged and whether free quota was applied;
• an error code and a truncated diagnostic message when a job fails.

We do not store the name of the file you uploaded, and — unlike a comic job — no page URL is involved.

3.4 Document and image data

The page you upload is held in memory only for as long as the redraw takes. We do not archive it: there is no copy of your original page in our object storage. (An earlier design did keep one; it was removed because nothing read it.)

The redrawn result is written to a private object-storage bucket that is not publicly listable or readable. Your browser fetches it through a signed link that expires after 30 minutes, and the object itself is deleted automatically 7 days after it is created (section 7).

The PDF you upload is different: it is stored at rest, in a private bucket that is not publicly listable or readable, because the container that translates it downloads it there rather than receiving it through our application. The result documents are written to the same bucket. Neither is reachable without a signed link: the ones you use to download a result expire within 30 minutes, and the ones the translation container uses to read the source and write the results last for the length of the job plus a margin (45 minutes with the current settings). Both files are deleted automatically 7 days after upload (section 7). The text inside your PDF is sent to a third-party AI gateway to be translated; the file itself is not.

3.5 Payment data

Paying is optional — the free allowance and your own API key both work without it. When you do buy a subscription or a credit pack, the checkout is handled by Creem, which sells the Service to you as merchant of record: Creem, not us, charges your card, issues the invoice and appears on your statement. We never see or store your card number, CVC or expiry. What we hold is an order record: order id, the plan or credit pack bought, amount, currency, status, the Creem checkout and subscription identifiers, and timestamps. Creem receives your email address so it can send the receipt, and handles your payment details under its own privacy policy.

Payments made before this were processed by Stripe on the same terms — card data went to Stripe directly, never through us. Those older order records are kept for the accounting and tax period described in section 7, but nothing new is sent to Stripe.

3.6 Technical and operational data

Cloudflare, which serves this site, processes connection metadata such as IP address, user agent and request timing for delivery, caching and abuse prevention. Our own application logs record job identifiers, error codes, gateway names and durations — operational diagnostics, not content.

3.7 Cookies

We set one category of cookie: the session cookie that keeps you signed in after OAuth. It is strictly necessary for the account area to function. We run no analytics, advertising, profiling or third-party tracking cookies on this site, which is why you are not being shown a consent banner. The marketing pages do load a web font from Google Fonts, which means your browser makes a request to Google's font servers; no cookie is set by us for that.

4. How we use this data

• To provide the Service — authenticating you, running comic redraws and PDF translation jobs, delivering results.
• To count usage correctly — reserving, settling and refunding pages against your monthly free allowance, and against any credits your subscription or a credit pack granted.
• To prevent abuse — enforcing quotas and the acceptable-use rules in the Terms of Service, and detecting automated or shared-account use.
• To debug and improve — investigating failed jobs and upstream gateway problems.
• To comply with law — accounting and tax records, and responding to valid legal requests and copyright complaints.

We do not sell your personal data. We do not use your uploaded images or your job history to train our own models, and we do not use them for advertising or profiling.

5. Legal bases (EEA/UK users)

• Performance of a contract — account data, job records, image processing and payment records, all of which are necessary to deliver a service you asked for.
• Legitimate interests — security, abuse prevention, diagnostics and keeping the Service running, balanced against your rights.
• Legal obligation — retaining transaction records and responding to lawful requests.
• Consent — where we ask for it explicitly, such as the consent click that links the extension to your account. You may withdraw it by revoking the token.

6. Who else processes your data

We share the minimum necessary with the following providers. We do not sell data to anyone, and no provider below is permitted to use your data for its own marketing.

We may also disclose data where we are legally required to, or where it is necessary to establish, exercise or defend legal claims — including responding to a copyright complaint under the procedure in our Terms of Service.

International transfers. These providers operate globally, so your data — including a comic page or a PDF you upload — is processed outside your own country, and outside the EEA/UK. Where a transfer of personal data leaves the EEA/UK we rely on the European Commission's Standard Contractual Clauses or an equivalent safeguard offered by the provider. If you are not comfortable with a particular page or document being processed this way, do not upload it.

7. How long we keep things

8. Your rights

Depending on where you live you may have the right to access, correct, delete, export or restrict the processing of your personal data, to object to processing based on legitimate interests, and to withdraw consent. Californian residents have the rights to know, delete, correct and opt out of "sale" or "sharing" — we do neither — and not to be discriminated against for exercising them.

To exercise any of these, email dylan@magic-design.art from the address on your account. We will respond within 30 days. Note that deleting your account does not delete the transaction records we are legally required to keep, and that unused credits are forfeited on deletion (see the Terms of Service).

If you are in the EEA or UK and think we have handled your data badly, you may complain to your local supervisory authority. We would appreciate the chance to fix it first.

9. Security

• Sign-in is OAuth-only, so there is no password of yours for us to lose.
• Extension tokens are stored as SHA-256 hashes, never in plaintext.
• Result images live in a private bucket and are only ever reachable through a short-lived signed link tied to your job.
• All traffic is served over HTTPS with HSTS.
• Your own AI provider keys are never sent to us, so they cannot be exposed by a breach of ours.

No system is perfectly secure, and we cannot guarantee absolute security. Keep your account and your extension tokens to yourself; you are responsible for activity carried out through them.

10. What you should not upload

11. Children

The Service is not directed at children under 13 (or under 16 where local law sets that threshold), and we do not knowingly collect their personal data. If you believe a child has given us personal data, contact us and we will delete it.

12. Changes to this policy

We may update this policy as the Service changes. The "last updated" date at the top always reflects the current version. If a change materially reduces your rights or materially widens what we collect, we will give notice on this site or by email before it takes effect. Continuing to use the Service after a change takes effect means you accept the updated policy.

13. Contact

Privacy questions, data requests and complaints: dylan@magic-design.art.