Opal by Welocalize: AI Localization, Trained on Your Content

Opal is the AI localization platform from Welocalize. Welocalize trains it on your own translation memories, glossaries, and style guides, so the output follows your terminology and tone instead of a generic model's defaults.

This integration connects Opal to your Crowdin projects. When a file reaches 100% translation in a target language, Crowdin sends it to Opal. Opal edits the translations where needed and returns them to the project, with "Opal" recorded as the author of every change. Translations that Opal's quality evaluation marks as final are approved automatically. You choose in project settings whether Opal-edited translations get approved too or wait for your reviewers.

Key Features

• Custom-trained AI: Welocalize trains Opal on your linguistic assets and issues an API key and a content group for your organization. The connector uses them to route your content to the model configuration built for you.
• Consistent brand voice: Opal applies your terminology, tone, and style rules across languages and content types.
• Feedback loop: On file approval, the connector sends the final XLIFF to Opal. Opal learns from what your reviewers actually approved, not from intermediate drafts.
• Human work stays untouched: The connector never sends human-authored translations, perfect (101%) TM matches, or approved TM matches to Opal.
• Incremental processing: The connector keeps a watermark per file and language. If a file has not changed since the last run, the event is skipped and no Opal capacity is used.
• Traceable changes: Every translation from Opal is authored by the Opal agent and visible in the project Activity Log.
• You control approvals: Segments that pass Opal's quality evaluation are approved automatically. For segments Opal has edited, the project setting Approve Opal-edited translations decides: approve them right away, or leave them for human review. Keep it off if your workflow requires a person to approve every string.

How It Works

1. A file reaches 100% translation for a target language in Crowdin (via pre-translation, TM, or human translators).
2. Crowdin sends the file to Opal through the connector.
3. Opal evaluates and edits the translations using the model trained on your assets, then returns them to the project. Segments its quality evaluation marks as final are approved. Edited segments are approved only if Approve Opal-edited translations is enabled.
4. Your team reviews the translations in the Crowdin Editor as usual.
5. When the file is approved, the connector sends the final XLIFF back to Opal as feedback.

Use Cases

• UI strings and product copy that must stay consistent from release to release
• Legal and compliance content where approved terminology is mandatory
• Help center articles and support macros updated in high volume
• Automated content pipelines that run without a manual trigger

How to Set Up

1. Request access from Welocalize to receive your Opal API key, base URL, and content group ID.
2. Install the app from the Crowdin Store.
3. In Organization Settings → Opal Settings, enter the API key, base URL, and default content group.
4. In each project where you want Opal to run, open Project Tools → Opal and enable the integration. Decide whether to turn on Approve Opal-edited translations – this setting is not recommended for human-in-the-loop workflows. You can override the content group at the project level if needed.