Translate long PDFs between languages using OpenAI and a friendly Gradio UI. By default the app translates English → Spanish, but you can switch to any supported pair in the interface.
Show loading state screenshot
- Extracts text from PDFs of arbitrary length using
pypdf - Splits content into configurable chunks so 100+ page documents can be processed safely
- Uses the OpenAI Chat Completions API for high-quality translations while preserving structure
- Provides a browser-based Gradio interface for uploading PDFs, reading the translated output, and downloading it as a fresh PDF whose filename mirrors the original with a
{src}_to_{dst}suffix
- Python 3.10+
- OpenAI API key with access to GPT-5-mini (or another translation-capable model)
fonts/DejaVuSans.ttfis bundled for Unicode-safe PDF exports (sourced from the DejaVu fonts project and redistributed underfonts/DejaVuSans-LICENSE.txt)- Generated PDFs are stored locally under
translated_pdfs/(ignored by git) using the original filename plus the language suffix
- (Optional) Create and activate a virtual environment.
python -m venv .venv source .venv/bin/activate - Install dependencies.
pip install -r requirements.txt
- Provide your OpenAI API key.
cp .env.example .env # edit .env and set OPENAI_API_KEY
python app.pyGradio starts a local server and prints URLs in the terminal. Open the http://127.0.0.1:7860 link, upload a PDF, choose source/target languages, and click Translate PDF. The translated text appears in the textbox, and a Download translated PDF button lets you grab a PDF version of the translation.
- Uploaded PDFs are read locally and the extracted text is sent to OpenAI's Chat Completions API for translation. Avoid using the app with documents that contain confidential or regulated information unless your OpenAI account allows it.
- Translated PDFs are written to
translated_pdfs/on your machine and are never uploaded anywhere.
Environment variables (set them in .env or your shell):
| Name | Description | Default |
|---|---|---|
OPENAI_API_KEY |
Required OpenAI credential | — |
OPENAI_TRANSLATION_MODEL |
Model used for translation | gpt-5-mini |
MAX_CHARS_PER_CHUNK |
Character limit for each translation chunk | 3500 |
Lowering MAX_CHARS_PER_CHUNK trades throughput for higher reliability on very large files.
Run the automated test suite (also executed in CircleCI) with:
pytestIntegration tests hit the real OpenAI API; they run only when both OPENAI_API_KEY and RUN_OPENAI_TESTS=1 are set (CircleCI sets both via project settings). Locally you can omit RUN_OPENAI_TESTS to skip them.
- Add support for more advanced formatting (tables, images) in the generated PDF if needed.
- Extend the language list or add automatic language detection.
Issues and pull requests are welcome. Please run pytest (and set RUN_OPENAI_TESTS=1 if you want to include integration tests) before opening a PR.
Released under the MIT License.