Introduction

Translate PDF Online — architecture, features, security, and open-source compliance.

What is Translate PDF Online?

Translate PDF Online is a web service for translating PDF documents while preserving layout, figures, and tables where possible. It is designed for academic and professional PDFs across English, Chinese (Simplified), Spanish, French, Italian, Greek, Japanese, Korean, German, Russian, and more language pairs over time.

Website: www.translatepdfonline.com
Contact: hello@translatepdfonline.com

Open source and credits

Our translation pipeline builds on community open-source work, in particular zotero-pdf2zh and the wider pdf2zh / BabelDOC ecosystem for layout-aware PDF translation.

We encourage you to review the upstream project for implementation details and license terms.

AGPL-3.0 compliance

Components derived from pdf2zh (AGPL-3.0) may require that users who interact with the service over a network can obtain corresponding source code for the AGPL-covered parts, or that the project otherwise complies with AGPL obligations.

Source access: Upstream reference implementation and related code: github.com/guaguastandup/zotero-pdf2zh. For questions about our deployment-specific source offerings, contact hello@translatepdfonline.com.

This section is for transparency; it is not legal advice. Consult counsel for your AGPL obligations.

System architecture

We use a separated front end and back end.

Front end (edge)

Stack: React / Next.js application.
Hosting: Cloudflare Pages / Workers (edge delivery).
Role: UI, authentication flows, billing UI, PDF preview and comparison in the browser, and calls to backend APIs.

Object storage

Cloudflare R2 stores uploaded PDFs and generated outputs, with S3-compatible APIs.
R2 helps control egress costs for large PDF downloads compared to many object stores.

Backend API (application servers)

Stack: Python (FastAPI or similar) on elastic cloud servers (ECS / VPS-class hosting).
Role: User accounts, credits and subscriptions metadata, task creation, orchestration, and integration with the translation model provider (DeepSeek API) and worker queues.

Async workers

Celery (or equivalent) workers run CPU- and IO-heavy steps: layout analysis (e.g. DocLayout-YOLO), text extraction, translation calls, and BabelDOC-style re-rendering so translated text is placed back into the PDF vector structure where supported.

Data flow (summary): Browser → Next.js on Cloudflare → Backend API → R2, task queue, DeepSeek API → workers → R2.

Core features

A. PDF viewing and area selection

Library: react-pdf-selection (or equivalent) overlays an interaction layer on the PDF view.
Flow: The user draws a rectangle; the client reads normalized coordinates for that page region and sends them to the backend.
Partial translation: The backend uses PyMuPDF (or similar) to extract text from the region, calls DeepSeek for translation, and returns text for inline or overlay display.

B. Side-by-side comparison and scroll sync

Engine: pdf.js (or equivalent) renders two PDF panes (source and translated).
Sync: The left pane’s scroll position is read, a scroll ratio is computed, and the right pane is updated so both views stay aligned even when page lengths differ slightly.

C. Task pipeline

Parse / layout: Workers run layout detection (paragraphs, equations, tables) as configured.
Translate: Batches are sent to DeepSeek (e.g. DeepSeek-V3). Token costs vary with document size and model pricing—treat public pricing as indicative and subject to change.
Render: BabelDOC (and related tooling) writes translated strings back into the PDF structure while preserving style where possible.

Using the translate workspace (History panel)

On the Translate PDF screen, the History column is central to your workflow:

It lists uploads and translation tasks tied to your current browser session (and to your account when you are signed in), including status, source/target languages, and filenames.
Open a past task from History to resume preview, download the translated PDF, or monitor progress—without uploading the same file again.
History works together with the PDF preview and task controls on the right: selecting an entry loads that task’s context so you can continue where you left off.

If you do not see an older item, it may have been removed by retention policy (see below) or you may be on a different device or session.

Data security and retention

Transport: HTTPS end-to-end (including Cloudflare to origin).
Retention: Translated outputs and related objects in R2 are scheduled for automatic deletion after approximately 7 days unless a different retention rule is stated in product settings or law requires otherwise.
Security: We apply reasonable access controls and encryption in transit; no online service can guarantee absolute security.

For legal terms, privacy, and payment processing, see our Privacy Policy and Terms of Service.