translatarr-client

A typed Python client for the Translatarr API — synchronous and asynchronous, with pydantic response models generated from the server's OpenAPI spec.

Install

pip install translatarr-client

Published to PyPI for each v* tag (the version matches the app's tag). The same wheel + sdist are also attached to the corresponding GitHub Release, and you can install straight from this repo:

# from a published GitHub Release (version matches the app's v* tag)
pip install https://github.com/joshrmcdaniel/translatarr/releases/download/v1.2.3/translatarr_client-1.2.3-py3-none-any.whl

# or straight from this repo
pip install -e clients/python

Requires Python 3.10+.

Authentication

Mint a personal API key in the web UI under Settings → API keys (tokens are prefixed tra_ and shown only once). Pass it as token:

from translatarr import TranslatarrClient

with TranslatarrClient("https://translatarr.example", token="tra_…") as tra:
    result = tra.translate("Good morning", source_lang="en", target_lang="ja")
    best = result.translations[0]
    print(best.text, best.romanization)

A browser session cookie works too, via session_cookie="…" instead of token.

Async

AsyncTranslatarrClient mirrors the sync client method-for-method:

import asyncio
from translatarr import AsyncTranslatarrClient

async def main() -> None:
    async with AsyncTranslatarrClient("https://translatarr.example", token="tra_…") as tra:
        result = await tra.translate("Bonjour", source_lang="auto", target_lang="en")
        print(result.detected_source_language, result.translations[0].text)

asyncio.run(main())

What you can do

Persisting a chat without paying for a second translation

translate does not store anything; add_turn translates and persists. To avoid a duplicate LLM call, pass the result you already have:

chat = tra.create_chat(source_lang="en", target_lang="ja")
preview = tra.translate("Let's ship it", source_lang="en", target_lang="ja", chat_id=chat.id)
chat = tra.add_turn(chat.id, "Let's ship it", source_lang="en", target_lang="ja", result=preview)

Errors

Failures raise a subclass of translatarr.APIError, each carrying status_code, the server's code (when present), and the raw response:

from translatarr import NotFoundError

try:
    tra.get_chat("does-not-exist")
except NotFoundError:
    ...

Development

The response models in src/translatarr/_models.py are generated from the server's OpenAPI document — do not edit them by hand. To regenerate after the API schema changes:

# one-time tooling install
python -m venv clients/python/.venv
clients/python/.venv/bin/pip install -e 'clients/python[codegen]'

# regenerate openapi.json + _models.py
clients/python/scripts/regenerate.sh

The script dumps the spec via bun scripts/dump-openapi.ts (the same document served at /api/docs/openapi.json) and runs datamodel-codegen over it. Only _models.py is committed, so the client builds and installs without a running server; openapi.json is a regenerated, gitignored artifact.