abersetz 1.0.14

No project description provided

---

this_file: README.md

abersetz

Minimalist file translator that reuses proven machine translation engines while keeping configuration portable and repeatable. The tool walks through a simple locate → chunk → translate → merge pipeline and exposes both a Python API and a fire-powered CLI.

Why abersetz?

• Focuses on translating files, not single strings.
• Reuses stable engines from translators and deep-translator, plus pluggable LLM-based engines for consistent terminology.
• Persists engine preferences and API secrets with platformdirs, supporting either raw values or the environment variable that stores them.
• Shares voc between chunks so long documents stay consistent.
• Keeps a lean codebase: no custom infrastructure, just clear building blocks.

Key Features

• Recursive file discovery with include/xclude filters.
• Automatic HTML vs. plain-text detection to preserve markup when possible.
• Semantic chunking via semantic-text-splitter, with configurable lengths per engine.
• voc-aware translation pipeline that merges <voc> JSON emitted by LLM engines.
• Offline-friendly dry-run mode for testing and demos.
• Optional voc sidecar files when --save-voc is set.

Installation

pip install abersetz

Quick Start

First-time Setup

# Automatically discover and configure available translation services
abersetz setup

This will scan your environment for API keys, test endpoints, and create an optimized configuration.

Basic Translation

# Using the main CLI
abersetz tr pl ./docs --engine translators/google --output ./build/pl

# Or using the shorthand command
abtr pl ./docs --engine translators/google --output ./build/pl

CLI Options (preview)

• to_lang: first positional argument selecting the target language.
• --from-lang: source language (defaults to auto).
• --engine: one of
  • translators/<provider> (e.g. translators/google)
  • deep-translator/<provider> (e.g. deep-translator/deepl)
  • hysf
  • ullm/<profile> where profiles are defined in config.
• --recurse/--no-recurse: recurse into subdirectories (defaults to on).
• --write_over: replace input files instead of writing to output dir.
• --save-voc: drop merged voc JSON next to each translated file.
• --chunk-size / --html-chunk-size: override default chunk lengths.
• --verbose: enable debug logging via loguru.

Configuration

abersetz stores runtime configuration under the user config path determined by platformdirs. The config file keeps:

• Global defaults (engine, languages, chunk sizes).
• Engine-specific settings (API endpoints, retry policies, HTML behaviour).
• Credential entries, each allowing either { "env": "ENV_NAME" } or { "value": "actual-secret" }.

Example snippet (stored in config.toml):

[defaults]
engine = "translators/google"
from_lang = "auto"
to_lang = "en"
chunk_size = 1200
html_chunk_size = 1800

[credentials.siliconflow]
name = "siliconflow"
env = "SILICONFLOW_API_KEY"

[engines.hysf]
chunk_size = 2400

[engines.hysf.credential]
name = "siliconflow"

[engines.hysf.options]
model = "tencent/Hunyuan-MT-7B"
base_url = "https://api.siliconflow.com/v1"
temperature = 0.3

[engines.ullm]
chunk_size = 2400

[engines.ullm.credential]
name = "siliconflow"

[engines.ullm.options.profiles.default]
base_url = "https://api.siliconflow.com/v1"
model = "tencent/Hunyuan-MT-7B"
temperature = 0.3
max_input_tokens = 32000

[engines.ullm.options.profiles.default.prolog]

Use abersetz config show and abersetz config path to inspect the file.

CLI Tools

• abersetz: Main CLI with tr (translate) and config commands
• abtr: Direct translation shorthand (equivalent to abersetz tr)

Python API

from abersetz import translate_path, TranslatorOptions

translate_path(
    path="docs",
    options=TranslatorOptions(to_lang="de", engine="translators/google"),
)

Examples

The examples/ directory holds ready-to-run demos:

• poem_en.txt: source text.
• poem_pl.txt: translated sample output.
• vocab.json: voc generated during translation.
• walkthrough.md: step-by-step CLI invocation log.

Development Workflow

uv sync
python -m pytest --cov=. --cov-report=term-missing
ruff check src tests
ruff format src tests

Testing Philosophy

• Every helper has direct unit coverage.
• Integration tests exercise the pipeline with a stub engine.
• Network calls are mocked; real APIs are never hit in CI.

License

MIT