Privacy Policy

Last updated: 2026-06-23

📋 Site-wide privacy (cookies, advertising, third-party services on the blog and tool pages): mathsystem.dev/blog/privacy/. This page covers only the LaTeXFlow tool itself (Add-on permissions, Web app data processing).

This Privacy Policy describes how LaTeXFlow — both the Google Docs™ Add-on (“the Add-on”) and the Web app at mathsystem.dev (“the Web app”) — collects, uses, and protects your information. Where a section differs between the two products, the distinction is called out explicitly.

1. Who We Are

LaTeXFlow is developed and maintained by an individual developer (contact: sung2417@gmail.com). The Add-on is published on the Google Workspace Marketplace; the Web app is hosted at mathsystem.dev/latexflow/web/. The same developer is the data controller for both products.

2. Google Workspace Data Access (Add-on)

The Add-on accesses your Google Docs™ document solely to provide its core functionality:

• Document reading: The Add-on scans your document body to detect LaTeX expressions enclosed in $...$ and $$...$$ delimiters.
• Document writing: Detected expressions are replaced in-place with rendered equation images. The Add-on also reads inline image metadata to support reverting images back to editable LaTeX text.
• Sidebar and dialogs: The Add-on displays a sidebar panel for LaTeX input and live preview, and a modal dialog for managing data collection consent, within Google Docs™.
• External rendering libraries: The Add-on loads Temml and MathJax from jsDelivr CDN to render LaTeX expressions as images within the sidebar. No document content is included in these requests.
• User identification: The Add-on reads your Google account email address solely to associate anonymized error logs and optional training data with your session. Your email is never shared with third parties or stored in plain text outside of Google’s infrastructure.

Your document content is processed within Google’s infrastructure (Apps Script) and is not transmitted to external servers. The Add-on never stores or shares the text content of your document.

2b. Web App Data Access

The Web app runs entirely in your browser. The .docx file you upload is parsed and converted client-side using JSZip and MathJax loaded from jsDelivr CDN; the document body text is never transmitted to our servers.

When you convert equations, the LaTeX source for each equation and its rendered PNG image are sent to our collection endpoint (a Google Cloud Function named collect-web-pair) for inclusion in a training dataset. The non-equation contents of your document — paragraphs, tables, images, file name, author metadata — are never transmitted. See §3a for the exact fields collected.

External rendering: MathJax is loaded from jsDelivr CDN. The .docx content itself is processed locally in your browser.

2c. Web App — Google Drive Import (Optional)

The Web app provides an optional “Import from Google Drive” button alongside the standard drag-and-drop input. This path is for users (notably on iPad) whose Google Drive app only exports .gdoc as PDF. The button is dormant until you click it — if you only use drag-and-drop, no Google Drive access ever occurs and this section does not apply.

• OAuth scope — https://www.googleapis.com/auth/drive.file only. This is a narrowly-scoped permission: the app can read only the file you explicitly pick through Google’s Drive Picker dialog. It cannot list, search, or read any other file in your Drive.
• Sign-in flow — Google handles the sign-in and consent screen. You may see an “App isn’t verified” warning while the app is in unverified status; you can proceed via Advanced → Continue or cancel at any point.
• File handling — Once you select a Google Doc (or .docx / .odt / .rtf), Google exports it as .docx and sends it directly to your browser. The file then enters the same client-side pipeline described in §2b. File content never transits our servers.
• Access token — The OAuth access token returned by Google is held in browser memory only, scoped to the current page. Closing the tab or refreshing discards it. We do not store, log, or transmit the token.
• Credentials embedded in the page — The Web app embeds an OAuth Client ID and a Google API Key. Both are spec-public identifiers: the OAuth Client ID is public by design under the OAuth 2.0 specification, and the API Key is restricted by HTTP referrer to mathsystem.dev (and localhost for development). No client_secret is used (the Web app uses PKCE / implicit flows).
• Revoking access — You can revoke the Web app’s Drive access at any time at myaccount.google.com/permissions.

For the underlying Google services and their policies, see Google API Services User Data Policy and Google’s Privacy Policy.

3. Information We Collect

3a. Training Data

The collection mechanism differs between the two products:

For the legal basis of the Web app’s automatic collection, see §5b for the rationale and your opt-out options.

Fields collected:

This data is used solely for training future OCR/recognition models that improve LaTeXFlow’s ability to automatically detect and convert equations.

Personally identifiable information (name, IP address, document content) is never collected by the Web app. The Add-on collects your Google account email only to allow you to manage your consent and request deletion.

3b. Error Logs

If a conversion error occurs in the Add-on, it automatically records:

• The LaTeX string that caused the error
• The error message
• A timestamp

This data is stored in a private Google Sheet accessible only to the developer, and is used for debugging purposes only. The Web app does not transmit error logs.

4. How We Store Data

Training data (§3a) from both products is stored in Google Cloud Storage (GCS), in a single bucket owned by the developer’s GCP project. Access is restricted to the developer’s service account.

Object layout:

Error logs (§3b) are stored in a private Google Sheets file.

5. Your Consent and Control

5a. Add-on — Consent Dialog

• On first launch, the Add-on displays a consent dialog asking whether you agree to share anonymized training data.
• You may decline at any time — the Add-on’s core functionality (equation insertion) works fully without consent.
• You may withdraw or change your consent at any time via: Extensions → LaTeX Converter → Data Collection Settings

We do not collect Add-on training data unless you have explicitly agreed.

5b. Web App — How to Opt Out

The Web app does not display a consent dialog because the data is fully pseudonymized — no email, name, IP address, or document content is collected. The only client-side identifier is a 16-hex random UUID stored in your browser’s localStorage under the key latexflow_anon_hash. This UUID is generated locally with crypto.getRandomValues and is not derived from any identifying information; it is used only to group submissions from the same browser for statistical analysis (e.g., to detect duplicate equations from the same session). It is not linked to your identity and is cleared whenever you clear site data.

This automatic processing relies on the lawful bases summarized in §3a. Under GDPR Art.21 you have the right to object at any time; under PIPA Art.37 you have an equivalent right to stop processing. Both can be exercised without contacting us, as follows.

To opt out at any time, open your browser DevTools (press F12), go to the Console tab, and run:

1
localStorage.setItem('latexflow_collect_optout', '1')

The opt-out is enforced immediately by the Web app and persists until you clear site data. While the opt-out is set, no further submissions are sent. You may also simply avoid using the Web app if you prefer not to contribute.

To request deletion of previously submitted Web app records, email sung2417@gmail.com together with the value of latexflow_anon_hash from your browser (read it from the Console using localStorage.getItem('latexflow_anon_hash')). Without that identifier we cannot locate your submissions, because no email, name, or IP address is stored alongside them.

6. Data Retention

Training data — from both the Add-on and the Web app — is retained while it remains useful for model training. Records that have been unused for more than 12 months are reviewed quarterly and deleted if they are no longer needed.

Error logs are retained for up to 12 months and then deleted.

You may request deletion of your anonymized training data by contacting us at sung2417@gmail.com. For the Web app, please include the latexflow_anon_hash value as described in §5b.

7. Third-Party Services

LaTeXFlow uses the following third-party services:

These services are governed by their own privacy policies.

8. Children’s Privacy

LaTeXFlow is not directed at children under 13. We do not knowingly collect data from children under 13.

9. Changes to This Policy

We may update this policy from time to time. The “Last updated” date at the top of this page will reflect any changes. Continued use of the Add-on or the Web app after changes constitutes acceptance.

10. Contact

For privacy-related questions or data deletion requests:

Email: sung2417@gmail.com GitHub Issues: github.com/Seong6496/google-docs-latex-addon/issues