Deepl Translate Node
/install deepl-translate-node
DeepL Translate (confidence-gated fallback)
A thin wrapper around the DeepL API. The point of this skill is the trigger logic, not the API call: translate with DeepL when your own translation might be wrong and being wrong matters.
When to reach for DeepL instead of translating yourself
Translate it yourself when the text is everyday prose and you are confident. Call DeepL when any of these is true:
- Proper nouns / brand / product names that have official localized forms
- Domain terminology — legal, medical, financial, technical specs, patents
- Ambiguous source where one word maps to several target words and context doesn't disambiguate
- Idioms / fixed expressions that don't translate literally
- Low-resource or distant language pairs where your training signal is thin
- High-stakes output — anything the user will publish, sign, or send to a third party, where a subtle error is costly
- The user explicitly asks for DeepL.
If you're confident and the cost of a minor error is low, just translate directly — don't burn an API call.
Prerequisite: API key
The key is read from the DEEPL_API_KEY environment variable (never hardcode it).
Default endpoint is the Free tier host api-free.deepl.com; for Pro, set
DEEPL_API_HOST=api.deepl.com.
Set it once:
# macOS / Linux — add to ~/.bashrc or ~/.zshrc to persist
export DEEPL_API_KEY="your-deepl-auth-key-here"
# Windows (persistent, user scope) — then open a NEW shell
setx DEEPL_API_KEY "your-deepl-auth-key-here"
Verify in a fresh shell: echo $DEEPL_API_KEY (bash) / $env:DEEPL_API_KEY (PowerShell).
If DEEPL_API_KEY is unset, stop and tell the user to set it — do not invent a key.
How to call it
Use the bundled Node helper (cross-platform, Node 18+). --source is optional;
omit it to let DeepL auto-detect.
Language codes (DeepL next-gen model — broad coverage)
DeepL supports far more than the classic ~30 languages. Don't assume a language is unsupported — check before falling back to a self-translation.
- Core (source + target):
AR BG CS DA DE EL EN ES ET FI FR HU ID IT JA KO LT LV NB NL PL PT RO RU SK SL SV TR UK ZH - Extended (~100 languages, incl. many low-resource ones):
AF BN FA HE HI HR HY KA ML MR MS MY NE PA SW TA TE TH UR VI YUE ZU… and many more. - Target-only regional variants — prefer these when precision matters:
- Chinese:
ZH-HANS(Simplified),ZH-HANT(Traditional) — better than bareZH - English:
EN-US,EN-GB - Portuguese:
PT-BR,PT-PT - Spanish:
ES-419(Latin American) FR-CA(Canadian French, beta),DE-CH(Swiss German, beta) — beta = not billed
- Chinese:
For the authoritative, always-current list, query the API. Use the v3 languages
endpoint (recommended for new integrations; replaces /v2/languages). The resource
query parameter is required — omitting it returns 400:
GET https://api-free.deepl.com/v3/languages?resource=translate_text with the
Authorization: DeepL-Auth-Key \x3Ckey> header. Valid resource values:
translate_text, translate_document, voice, write, glossary, style_rules.
It returns one entry per language (100+ for translate_text) with source/target
support flags. Or see
https://developers.deepl.com/docs/getting-started/supported-languages
Code-format caveat: /v3/languages returns BCP 47 codes (en-US, pt-BR,
zh-Hant — lowercase base, uppercase region, title-case script). But the
translation endpoint is still /v2/translate, whose target_lang accepts the
v2-style UPPERCASE codes (EN-US, PT-BR, ZH-HANT). Don't mix the two casings when
feeding a value from the languages endpoint into a translate call.
Versioning note: Only glossaries and the languages list have moved to v3. Text
translation has no v3 endpoint — /v2/translate is current and not deprecated.
node "{SKILL_DIR}/translate.mjs" --target ZH \
--text "The plaintiff filed a motion to compel discovery."
With an explicit source language:
node "{SKILL_DIR}/translate.mjs" --source JA --target EN-US --text "持分会社"
The script prints only the translated text on success, or an error line
(prefixed ERROR:) on failure. Multiple lines / paragraphs are preserved.
Workflow inside a task
- You're translating something and hit a passage you're unsure about.
- Confirm
DEEPL_API_KEYis set (the script checks and errors clearly if not). - Run
translate.mjsfor the uncertain passage (or the whole text). - Use DeepL's output. If it conflicts with your own read, surface both to the user and explain the discrepancy rather than silently picking one.
Glossary / terminology consistency (optional)
For repeated terminology, pass --glossary as a term=translation;term=translation
list and the script will post-process exact matches after DeepL returns. For real
glossary support DeepL has a Glossary API; ask the user if they need it and we can
extend the script.
Notes
- Free tier: 500,000 chars/month. The script does not track quota; if you get a
456response that means quota exhausted. 403means a bad/missing key.- For DeepL Pro, set
DEEPL_API_HOST=api.deepl.com(default isapi-free.deepl.com).
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install deepl-translate-node - After installation, invoke the skill by name or use
/deepl-translate-node - Provide required inputs per the skill's parameter spec and get structured output
What is Deepl Translate Node?
Use DeepL's neural MT API as a fallback when you are NOT confident in your own translation — proper nouns, ambiguous phrasing, domain/legal/medical terminolo... It is an AI Agent Skill for Claude Code / OpenClaw, with 33 downloads so far.
How do I install Deepl Translate Node?
Run "/install deepl-translate-node" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Deepl Translate Node free?
Yes, Deepl Translate Node is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Deepl Translate Node support?
Deepl Translate Node is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Deepl Translate Node?
It is built and maintained by rockbenben (@rockbenben); the current version is v1.1.1.