Skip to contentv0.1.0
PIMgate
TRANSLATION & LOCALIZATIONLIVE

Translate with your agency, AI and your team — on one clear path.

PIMgate turns translation into orders: product data from your PIM per export and target language, content from Ibexa DXP collected into one package. Your agency translates, AI fills gaps or drafts, your reviewers check every value in review rounds, and each order ends with a release and a tamper-evident protocol.

ORDERS PER LANGUAGEREVIEW ROUNDSRELEASE PROTOCOL

Illustrative · fictional data

In production use at a DACH manufacturer with 20,000+ products.

01 · The problem

Translation fails between the systems, not in the text.

Most companies have good translators. What goes wrong is everything around them: which values to send, who checks what, what came back, and whether anyone may publish it.

01

Spreadsheets by email.

Exports are pulled by hand, sent as attachments and come back in a different shape. Nobody can say which version is current.

02

Everything, every time.

A full catalogue goes out although only a few attributes changed. The same dictionary value appears on hundreds of products and is checked again on each one.

03

Review without context.

Reviewers see isolated strings in a table, not the product in the layout where it will appear — so terms drift and errors reach the shop.

04

No one knows whose turn it is.

Corrections go back and forth in mail threads. When something is finally published, there is no record of who checked and released what.

02 · Two flows

Product data and CMS content — two sources, one way to review and release.

Product texts come from your PIM, page content comes from your CMS. PIMgate runs both as translation orders with the same phases, the same roles and the same release — so the product page and the website speak with one voice.

Two flows into one review and release pathTop lane: a PIM export becomes product orders per export and target language — automatic, delta, selection or follow-up orders. Bottom lane: content from Ibexa DXP areas and items is collected into an Ibexa order and sent as a package. Both merge into translation by agency, AI or in house, review rounds and release, and split again: release status to catalogue, portal and API, and take-over into Ibexa as a new version.Product data · from the PIMCMS content · Ibexa DXPPIM exportFull and delta exportsProduct orderPer export × languageAuto · delta · selection · follow-upIbexa DXPAreas · items · fieldsIbexa orderCollect · send as packageXLIFF 1.2 / 2.0 · XML · JSONTranslationAgency · AI · in houseReview roundsFits · objectionReleaseProtocol · SHA-256Catalogue · portalRelease status · APIBack into IbexaNew version · datable

Illustrative · fictional data

Product data from the PIM

LIVE

Every imported PIM export creates one order per target language with all translatable entries — product texts, dictionary values and table cells. Further orders cover a narrower scope.

  • Automatic order per export and target language
  • Delta exports change only what they contain; after a release, changes land in a follow-up order
  • Manual orders over categories, a product list or a selection of PIM IDs — variants come along
  • The agency returns its work through the PIM import or by uploading the file it received

Content from Ibexa DXP

LIVE

Pages, articles and other content items are translated field by field. You create the order first — languages, due date, team — then collect content from any area into it and send it as one package.

  • Whole content items or single marked fields, from several areas
  • One package to the agency: XLIFF 1.2, XLIFF 2.0, XML or JSON
  • Or translated in house: AI drafts, a person confirms every draft
  • Confirmed translations are taken over into Ibexa as a new version — now or on a date

03 · How it works

Every order runs through the same phases — and always shows whose turn it is.

New order, in translation, review round with corrections, release, done. Review and corrections repeat as a round until everything fits. The phase bar names the person who has to act, and that person gets the mail.

Lifecycle of a translation orderSix phases in a row: New order, In translation, Review, Corrections, Release, Released. Review and corrections sit in a round that repeats. The export package travels to the agency, delivery 1 comes back into review round 1, a correction list goes to the agency, delivery 2 starts review round 2, and the order is released with a protocol and its status is published to catalogue, portal and API.MESSARA · Order 13 · DE › FR-CHYour teamAgencyRound 1Round 2repeats until everything fitsNew orderResponsibleIn translationAgencyReviewReviewerCorrectionsAgencyReleaseResponsibleReleasedEveryone informedExportDeliveryCorrection listTranslation agencyOwn TMS · CAT tool · memoryTranslatingCorrectingPDF · CSV · JSONPortal · API · IbexaXLIFFR1FixR2What just happenedOrder created · 1,240 entries · due 12.12.Export sent to the agency · ZIP with manifestDelivery 1 received · review round 1 · 2 objectionsCorrection list sent · the agency answers in the threadDelivery 2 · review round 2 · everything fitsRelease · protocol as PDF, CSV and JSONReleased · status to catalogue, portal and API

Illustrative · order 13 at tenant MESSARA · fictional data

  1. 01 · System or responsible

    Order created

    Automatically per export and target language, or by hand: data source, scope, target languages, due date. The preview shows products, entries, already delivered and missing per language.

  2. 02 · Responsible

    Team assigned

    Reviewers, agency people and a responsible person are chosen per order from the people of your tenant. New team members get a mail.

  3. 03 · Agency or AI

    Translation

    The export goes to the agency as a ZIP with a manifest and a signed, expiring link. For CMS content, AI can produce the first draft instead.

  4. 04 · Agency

    Delivery

    The agency delivers through your PIM import or uploads the filled-in file in the order. PIMgate compares each entry and writes a delivery report.

  5. 05 · Reviewer

    Review round

    Each agency delivery that changed values opens a review round. Reviewers mark every value as fits, does not fit (with a comment) or suggest a correction.

  6. 06 · Responsible and agency

    Corrections

    Objections are decided and sent back to the agency as a correction list with due dates. The next delivery opens the next round.

  7. 07 · Decider

    Release

    When nothing is open, products are released — optionally by two different people. Then the order is released and the protocol is generated.

  8. 08 · System

    Status and take-over

    The release status reaches catalogue, portal and API; a webhook announces it. Confirmed CMS content is taken over into Ibexa as a new version.

04 · Product orders

Orders that follow your PIM — per export, per language, per change.

PIMgate reads your PIM exports and derives the translation work from them. You decide what goes into an order; PIMgate makes sure nothing is sent twice and nothing is lost between deliveries.

LIVE

Automatic per export and language

An imported export creates one order per target language with all translatable entries. If an order for that language is still open, the delivery goes into it instead of starting a new one.

Delta and follow-up orders

A delta export changes only what it contains. If every order of a language is already released, changes land in a follow-up order with only those changes. A changed master text puts its translations back to «awaiting decision».

Manual orders over a selection

Choose an export, then the scope: the whole export, categories including subcategories, a frozen product list or a selection of PIM IDs. Variants of a selected product come along automatically.

Many orders in parallel

Several orders may run for the same export and language as long as they cover different parts of the catalogue. Shared entries are named in the preview and a delivery updates every open order that contains them.

Each value once

A dictionary value used on 300 products is one entry and one line for the agency. The panel shows its reach — «affects 46 products» — so a correction is made once, for all of them.

Review in the shop layout

Reviewers check the product as the website shows it, master and target side by side. Variants inherit from their product, so inherited values are reviewed once, on the product. Review also works inside a portal.

ENTRY STATES · NEW · MISSING · FITS · CORRECTION REQUESTED · ASKED FOR · AWAITING DECISION · SOURCE DATA · AI INTERIM

05 · AI and agency

Your agency, AI and your reviewers — combined the way each language needs.

PIMgate does not replace your translation agency. It sends it smaller, better prepared work and brings the answers back into review. AI is used where it helps: drafts for CMS content and interim texts for gaps in product data.

Product data

Agency translates, your team reviews

The agency works in its own TMS or CAT tool with its own translation memory. PIMgate prepares the export, receives the delivery, runs the review rounds and sends corrections back.

Product data

AI bridges the gap

Where a translation is still missing, AI can fill it as a marked interim text so the website stays readable — or send it only as a suggestion to the agency. The next delivery replaces it; the entry stays open work.

CMS content

AI first, people decide — or no agency at all

For Ibexa content, AI translates first as a draft. A person reads, edits and confirms each draft before it goes to Ibexa. Text can also be typed in by hand, or the cells go to the agency as a package.

How the AI is steered

  • Translation memory first

    Exact matches from the translation memory are applied before any AI call and cost no tokens. Fuzzy matches are passed to the model as preferred earlier translations.

  • Your glossary, as a must

    Glossary terms that appear in the text are passed to the model as mandatory terminology, per field group and language.

  • AI contexts per field group

    Instruction, length, glossary use and language rules per field group, with a preview of the finished prompt. A field category can be locked against AI entirely.

  • Language rules

    Regional rules are applied on every result — Swiss German always gets «ss» instead of the sharp s, even from the memory.

  • Budget and usage

    Tokens of the month against the budget, and who may use AI, are visible in the settings.

What the agency gets

  • A work basket of its own

    Agency people land on their work basket: the corrections they owe per order and the files sent to them. Catalogue, sync and other areas stay closed to them.

  • Packages it can work with

    A ZIP with manifest.json (order, round, format, count, SHA-256 per file) in XLIFF 1.2 (default), XLIFF 2.0, TMX or a TMS-XML draft; for Ibexa also plain XML or JSON.

  • Due dates and reminders

    Due date per order, correction deadlines in working days, a reminder if a link is not downloaded and a weekly follow-up while nothing is delivered.

  • Two ways back

    Deliver through your PIM import, or upload the filled-in file in the order — as a delivery that enters review, or as a proposal that changes no state.

  • A thread per value

    The agency answers objections in the comments. If the answer convinces, the reviewer sets the value to «fits» with one click — logged with the name.

06 · Team and parallel orders

Several people, many orders — and everyone sees what is theirs.

Each order has its own team. Mails go to the person whose turn it is, the work basket collects every task in one list, and many orders run in parallel per language, export or content area.

Parallel orders with their teamsFour orders as lanes: product order 13 to French (Switzerland) in corrections, product order 14 to Italian (Switzerland) in review, Ibexa order 7 to English in release and delta follow-up order 15 in translation. Each lane shows its team as initials — responsible, reviewers and several agency translators. A marker moves from lane to lane to the person whose turn it is, and the mail that tells them is shown below.OrderPhaseTeamOrder 13 · SmartHomeDE › FR-CH · product orderCorrectionsLMOSKRT1AT2AOrder 14 · SensorsDE › IT-CH · categoriesReviewLMORBRT3AIbexa order 7 · Landing pagesDE › EN · CMS contentReleaseJHOSKRT4AOrder 15 · DeltaDE › FR-CH · follow-up orderTranslationLMONBRT1AO responsible · R reviewer · A agency translatorMail → T1 (agency): 3 corrections in order 13, due 05.12.Mail → RB (reviewer): delivery 2 of order 14 is ready for reviewMail → JH (responsible): Ibexa order 7 is ready for releaseMail → T1 (agency): order 15 · delta export with 86 entries

Illustrative · fictional data

A team per order

Reviewer, agency and responsible are set per order — several people per role, for example two translators for one language. An empty role falls back to the general rule.

Mails say whose turn it is

Correction lists, comments and answers go only to the team of that order. Each person chooses immediate, daily, weekly or no notifications.

Work basket for every person

One row per order with language, phase, due date, open values and exactly one main action. Overdue orders come first; open Ibexa orders follow below.

Simple and Expert

«Simple» shows a phase bar, a «Next step» box with «Your turn» and the one action due now. «Expert» opens every tab — entries, deliveries, objections, history.

Comments and @mentions

Comment on a single value or on the whole order. Typing @ suggests the people of the order; mentions are sent as a collected mail with a link straight to the comment.

Many orders at once

Orders per language, per export and per catalogue or content area run side by side. The order list and each order header show the scope, so parallel orders stay apart.

07 · Review rounds and release

Every round is traceable. Every release is documented.

A review round is one agency delivery. Reviewers judge every value, objections go back as a correction list, and the release produces a protocol that cannot be changed afterwards.

  • The round as a checklist

    Each round shows what this delivery brought, in groups: master changed, objections corrected, objections returned unchanged, wish met, newly delivered, unchanged, still missing — with «fits» directly in the row.

  • Three judgements per value

    Fits. Does not fit — always with a comment and a reason. Or suggest a correction with the wanted wording and a word diff against the delivery.

  • Decisions, then the correction list

    Each objection gets exactly one answer: request a correction from the agency, fix it in the PIM source data, or keep the value. Reviewers can send the marked objections themselves, with an Excel file listing every value.

  • Release in two steps

    First the values (fits), then the product. Optionally under the four-eyes principle: two different people of the deciding role, in any order.

  • The release protocol

    PDF, CSV and JSON, each with a SHA-256 checksum, stored unchangeably. Per entry: master, released value, path of states, rounds, comments, corrections and whether AI was involved.

  • Reopen without rewriting history

    An order can be reopened with a reason. The new release adds an addendum; the original protocol stays byte-identical. The release is announced by webhook — with checksums, without translated values.

RELEASE PROTOCOL · ORDER 13 · FR-CH
TENANT      MESSARAORDER       13 · SmartHome · DE › FR-CHROUNDS      2 · deliveries 2 · objections 3ENTRIES     1,240 · fits 1,236 · source data 4AI          interim texts 18 · replaced by delivery 18RELEASED    product release by LM · four eyes: NBFILES       protocol.pdf · protocol.csv · protocol.jsonSHA-256     9c41…e07 · 2b7d…a19 · f03e…5c2WEBHOOK     translation.order.released

Illustrative excerpt · fictional data

08 · CMS content in Ibexa DXP

Ibexa content: collect it into an order, translate, review, take it back.

For content in Ibexa DXP the unit of work is the cell — one field of a content item in one target language. PIMgate writes confirmed translations back into Ibexa as a new version, and every take-over stays in the history.

LIVE
  1. 01

    Create the order.

    «New order» first: target languages, due dates and the team with reviewers and agency.

  2. 02

    Collect content.

    Tick content in the overview — from several areas — or add a whole content item or marked fields to the order.

  3. 03

    Send one package.

    «Send to agency» sends everything in one package: XLIFF 1.2 or 2.0, or plain XML or JSON for agencies without a CAT tool.

  4. 04

    Return and review.

    The return is uploaded in the order — by you or by the agency itself — and lands as an agency draft. Reviewers confirm each cell.

  5. 05

    Or translate in house.

    Without an agency, AI translates the selected cells as drafts; a person reads, edits and confirms them.

  6. 06

    Take over into Ibexa.

    Confirmed cells are written to Ibexa as a new version — immediately or on a date. Only confirmed cells are taken over.

Conflicts are visible

If Ibexa holds a different text than PIMgate took over, or the master changed after the translation, the cell says so. The old text stays available in the history.

Do not translate

A field can be excluded for one language on purpose. It then no longer counts as missing — also for future changes in Ibexa.

Languages per siteaccess

Master and target languages are set per Ibexa siteaccess, with exceptions for folders. PIMgate shows each language with its name and BCP-47 code.

Ibexa DXP is the CMS connector built today. Other CMS and DXP systems, such as Drupal, can be connected on request — talk to us about your content model.

09 · Formats and interfaces

Standard exchange formats, a manifest for every file, an API for the result.

Agencies get the formats their tools read. Every package travels with a manifest and checksums, and the outcome of a release is available to your other systems.

FormatProduct ordersIbexa ordersUse
XLIFF 1.2DefaultYesThe common CAT exchange format (e.g. Trados, memoQ, Phrase)
XLIFF 2.0YesYesWhen the agency supports the newer version
TMX 1.4Yes—Source and target for translation memories; always with the manifest
TMS-XMLDraft—PIM dictionary form, agreed with the agency before use
XML · JSON—YesPlain formats for agencies without a CAT tool, every language in one document
CSV / ExcelCorrection list—Checklist for people working in Excel; not a return path
ZIP + manifest.jsonAlways—Order, round, format, count and SHA-256 per file
ProtocolPDF · CSV · JSON—Release protocol with checksums
Webhook · REST APIRelease status—translation.order.released; status per product via /api/v1

In product orders, language codes in the files follow the form your PIM uses, so deliveries go back without conversion. Formats returned by the agency are read in the same format they were sent.

10 · Security

Agencies see their orders. Nothing more.

External translators work in your tenant with a narrow role. Every file, download and decision leaves a trace.

Narrow agency access

Agency logins see their work basket, their orders and the single product being reviewed — not the catalogue, sync, media or other modules, and not through the API either.

Signed, expiring links

Packages are sent as signed links that expire. Every download is logged; the first one notifies the reviewers and the admins.

Your PIM stays untouched

PIMgate never writes into your PIM. Corrections are proposals on the order; the source changes only through your own PIM and its next export.

EU / CH hostingGDPRrevDSGTenant isolationAppend-only audit

Security & sovereignty →

11 · FAQ

Questions about translation in PIMgate

12 · Translation workshop

Map your translation process in 90 minutes.

A remote working session with the PIMgate team. We look at your languages, your PIM and CMS, your agencies and your review process — and sketch how the orders would run.

  1. 01 Your languages, markets and regional variants
  2. 02 Sources: PIM exports and CMS content
  3. 03 Agencies, formats and who reviews what
  4. 04 Where AI helps and where people decide
  5. 05 Next steps and setup scope

// TRANSLATION WORKSHOP

Request a translation workshop

We reply within two business days with proposed dates.

LIVE

AI workflows

Where AI helps across the gate — and where people decide.

Put a gate between your catalog and chaos.