2 / 3System to system: the LanguageDesk API

One request in.Six languages out.

No inbox, no waiting. Your systems talk to ours through the LanguageDesk API: a release is tagged, the strings are sent, the translations come back as a pull request. Follow one call from start to finish.

8 steps1,284 strings0 emails
Scroll
01Trigger

It starts with a commit.

A developer merges a feature. The build pipeline spots new and changed strings and posts a project to the API by itself — nobody has to remember to order translations.

CI pipeline · lumio/app · release 3.8POST
  • build
  • test
  • localize
POST /api/v1/projects?auth_token=••••••••{ "project": { "contact_person": "jonas@lumio.example", "content_identifier": "App UI", "source_language": "en", "target_languages": ["de-DE", "fr-FR", "es-ES", "it-IT", "pl-PL", "ja-JP"], "file_transfer_method": "http-push", "files": { … 3 files … }} }
JSON
en.json1,012 strings
JSON
onboarding.json184 strings
JSON
emails.json88 strings
Sent by the pipeline on every release tag
02Accepted

Answered in milliseconds.

The API checks the request, creates a project and replies with its number before the pipeline has moved on. It is a real project, exactly like one opened by hand — your team can see it in LanguageDesk straight away.

Response · 200 OK#48213
{ "status": 200, "project": { "id": "48213", "company_id": "1207", "status": "in_progress" }}
UI
App UI 17/11/2026 14:02 en:de-DE,… (API requests)Project name, generated from your content identifier
JSON
en.jsonreceived
JSON
onboarding.jsonreceived
JSON
emails.jsonreceived
Files can also be pulled from an sFTP HotFolder instead of pushed
03Service

The content type picks the service.

How much human goes in is agreed once, per kind of content. Changelogs: AI only. The interface: AI plus linguist review. Legal texts: full professional translation. After that, a request only has to say what it is.

Service level · by content identifier#48213
AI translationTuned on your memory and termbase · minutes
"Changelog"
AI + linguist reviewAI first, a professional checks what matters · hours
"App UI"
Professional translationSpecialist translator plus second linguist · days
"Legal"
In this request
JSON
en.jsonAI + review
JSON
onboarding.jsonAI + review
JSON
emails.jsonAI + review
Set up once with your project team, applied to every call
04Memory

Memory does the first pass.

Before any engine or linguist starts, every string is compared with your translation memory. Most of this release has been translated before — and is simply reused.

Analysis · 9,870 words#48213
JSON
en.json
0%from memory
JSON
onboarding.json
0%from memory
JSON
emails.json
0%from memory
Words0
Reused0%
New0
58% comes straight from memory. Only new and changed strings need fresh work.
05AI translation

Translated while the build runs.

AI engines tuned on your memory and termbase translate all six languages in parallel. Placeholders, plurals and length limits are protected — and every string gets a confidence score.

AI translation · en → de-DE#48213
JSON
en.json
0strings
checkout.pay_nowPay now

Jetzt bezahlen

98
cart.items_count{count} items in your cart

{count} Artikel in Ihrem Warenkorb

97
error.session_expiredYour session has expired.

Ihre Sitzung ist abgelaufen.

96
player.resumeResume

Lebenslauf

58
EN→ DEFRESITPLJA
One short string looks ambiguous — flagged for a human
06Review

A human where it matters.

Confident strings pass straight through. The uncertain few go to a linguist who sees the screen they belong to. “Resume” on a video player is not a CV — a person spots that in a second.

Review queue · 37 of 1,284 strings#48213
player.resumeResume

Lebenslauf

58

Lebenslauf→Fortsetzen

MK
“It’s the button on the video player — continue, not CV. I added a context note so the AI gets it right next time.”
Auto-approved0
Reviewed0
Edited0
Confidence scores decide what a human needs to see
07Status

Always know where it stands.

No status meetings, no “any news?” emails. Your pipeline asks the API and gets a straight answer: in progress, or completed, with a delivery date. It is the same status your team sees in LanguageDesk.

GET /api/v1/projects/48213#48213
in_progress · AI translation running
in_progress · 37 strings with a linguist
completed · all six languages ready
{ "status": 200, "project": { "id": "48213", "delivery_date": "2026-11-18T09:30", "status": "in_progress""completed" }}
Ask as often as you like — or let the pipeline check on a schedule
08Delivery

Merged before the release.

One more call returns the delivery files. The pipeline unpacks them, opens a pull request and runs the tests. Translations ship with the feature — and every approved string goes back into the memory.

Pull request #482 · lumio/appMerged
GET /api/v1/projects/48213/download_delivery_files "delivery_files_url": "https://files.languagedesk.com/…zip", "status": "completed"
M
l10n-bot · opened by the pipeline
l10n: release 3.8 in 6 languages
JSONde-DE
JSONfr-FR
JSONes-ES
JSONit-IT
JSONpl-PL
JSONja-JP
  • Tests passed
  • Placeholders intact
  • Merged
↻ Memory updated: +0 approved strings. The next release starts further ahead.
branch l10n/48213
Your turn

Your pipeline is one request away.

Tell us what your systems look like. We’ll show you the calls that connect them.

Book a meeting →
3 API calls1,284 strings37 seen by a human0 emails
Proof of concept · client, project and numbers are illustrative. Endpoints follow the public LanguageDesk API docs.