mdxlate 1.1.5

Translate Markdown docs into multiple languages using LLMs with smart caching and custom prompts

mdxlate

Translate Markdown docs into multiple languages using LLMs.
Batteries included: prompt template, CLI, OpenAI/OpenRouter provider switch, and a simple change-detection cache.

📚 Documentation

Read the full documentation →

• Getting Started – Installation and quick start
• CLI Reference – Complete command-line guide
• Programmatic Usage – Python API and examples
• Caching System – How the cache works
• Custom Prompt – Customize translations
• Error Handling – Failure recovery
• Development Guide – Contributing
• FAQ – Troubleshooting

Install

pip install -e .

Quick start

1. Initialize the editable prompt (creates ~/.mdxlate/translation_instruction.txt):

mdx init

2. Run translations:

export OPENAI_API_KEY=sk-...   # or use OPEN_ROUTER_API_KEY when provider=openrouter
mdx run docs_src out --languages de fr --model gpt-4o-mini

Result: translated files under out/<lang>/..., preserving the original folder structure. A cache file .mdxlate.hashes.json is written in docs_src.

CLI

mdx run [OPTIONS] DOCS_SRC OUT_DIR

Options

• --base-language TEXT – Base language (default: en)
• --languages TEXT... – Target languages, space-separated (default: de)
• --model TEXT – Model name (default: gpt-4o-mini)
• --provider [openai|openrouter] – Backend provider (default: openai)
• --api-key TEXT – API key (overrides env)
• --api-env-key TEXT – Env var to read (default: OPENAI_API_KEY)
• --base-url TEXT – Custom base URL (e.g., OpenRouter)
• --prompt-path PATH – Use a custom prompt file instead of the default
• --force – Force re-translation, bypassing cache
• --cache-dir PATH – Directory for cache file (defaults to source directory)

Examples

OpenAI (env var):

export OPENAI_API_KEY=sk-...
mdx run docs_src out --languages de fr --model gpt-4o-mini

OpenRouter:

export OPEN_ROUTER_API_KEY=or-...
mdx run docs_src out --languages de --provider openrouter --model google/gemini-2.5-pro

Custom prompt:

mdx run docs_src out --languages de --prompt-path ./my_prompt.txt

Custom cache directory (for read-only CI/CD):

mdx run docs_src out --languages de --cache-dir /tmp

Error Handling

If any file fails to translate (e.g., due to API errors, rate limits, or network issues), mdxlate will:

1. Continue processing other files instead of crashing
2. Save the cache for successful translations
3. Generate a failure report at .mdxlate.failures.json with details about what failed

Example failure report:

{
  "failures": [
    {
      "file": "docs/advanced.md",
      "error": "Rate limit exceeded",
      "error_type": "RateLimitError"
    }
  ]
}

After fixing the issue (e.g., waiting for rate limits to reset), re-run the translation. Only failed files will be retried thanks to the cache.

Behavior

• Prompt: default lives at ~/.mdxlate/translation_instruction.txt (created by mdx init). You can edit it freely or pass --prompt-path.
• Cache: re-translation is skipped if file bytes + prompt content + model + language are unchanged. By default, cache is written to source directory as .mdxlate.hashes.json. Use --cache-dir for read-only environments.
• Structure: each language gets its own mirror tree under OUT_DIR/<lang>/.

Programmatic use

from pathlib import Path
from mdxlate.start_translation import start_translation

start_translation(
    docs_src=Path("docs_src"),
    out_dir=Path("out"),
    base_language="en",
    languages=["de", "fr"],
    model="gpt-4o-mini",
    provider="openai",  # or "openrouter"
    api_key=None,       # pass explicitly or rely on env
    base_url=None,
    prompt_path=None,
    cache_dir=None,     # optional: specify custom cache directory
)

Integrations

• Jekyll – Complete guide for translating Jekyll sites with frontmatter preservation

Files of interest

• mdxlate/cli.py – Typer CLI (mdx init, mdx run)
• mdxlate/client.py – make_client() factory (OpenAI/OpenRouter)
• mdxlate/translator.py – translation, hashing, and I/O
• mdxlate/translation_instruction.txt – default prompt template

Development

Setup

pip install -e .
pip install ruff mypy pytest

Code Quality

This project uses Ruff for linting and formatting, and Mypy for type checking:

# Lint code
ruff check src tests

# Auto-fix linting issues
ruff check --fix src tests

# Format code
ruff format src tests

# Type check
mypy src --ignore-missing-imports

# Run tests
pytest tests/

CI/CD

The .github/workflows/quality.yml workflow runs automatically on every push and PR:

• ✅ Ruff linting
• ✅ Ruff formatting check
• ✅ Mypy type checking

License

MIT

Layout

repo/
  pyproject.toml
  README.md
  src/
    mdxlate/
      __init__.py
      cli.py
      client.py
      translator.py
      start_translation.py
      translation_instruction.txt
  tests/            # optional
  main.py           # optional local test runner

---

# pyproject.toml
```toml
[build-system]
requires = ["setuptools>=69", "wheel"]
build-backend = "setuptools.build_meta"

[project]
name = "mdxlate"
version = "0.1.0"
description = "Translate Markdown docs into multiple languages using LLMs."
readme = "README.md"
requires-python = ">=3.10"
license = { text = "MIT" }
authors = [{ name = "Tobias Bück" }]
dependencies = [
  "typer>=0.12",
  "openai>=1.40",
  "tenacity>=8.2",
]

[project.scripts]
mdx = "mdxlate.cli:app"

[tool.setuptools]
package-dir = {"" = "src"}

[tool.setuptools.packages.find]
where = ["src"]
include = ["mdxlate*"]

[tool.setuptools.package-data]
mdxlate = ["translation_instruction.txt"]

[tool.pytest.ini_options]
addopts = "-q"
testpaths = ["tests"]