Integration Releases 2026

Click Follow/New articles and comments on the section level (one up from the article level) to receive notification of when new notes are released.

Earlier release notes can be found in the archive.

For latest changes to Phrase APIs, SDKs, and developer tools, see Phrase Developer Hub.

v26.9 - May 12, 2026

Job Sync was failing and no project was created when setting up new projects. This has been fixed.

API: Added msWordParams.altText and msPowerPointParams.altText flags to .DOCX and .PPTX files import settings to control extraction of image and object alt text for translation. The option is disabled by default.

Four new fields are added to the segment_statistic table in Phrase Data:

Field name	Description
latest_editing_source	Identifies which segments were processed by Phrase post-editing AI features or a partner integration service.
locking_source	Identifies who/what has locked the segment. Possible values: `mt` - machine translation, set only by pre-translation `tm` - translation memory, set only by pre-translation `nt` - non-translatable, set only by pre-translation `ir` - import rule, set during import for some file formats, by import settings `ex` - external, set in all other cases - especially when locked via import xliff/mxlif
locking_source_detail	Identifies the service for cases when the locked by value is `ex` - external
quality_estimation_score	This is tied to `latest_editing_source` field. It represents the QPS score generated via Phrase AI features, or the score provided by a partner service integration. Whenever `latest_editing_source` is cleared or updated, this field (if present) should follow the same behaviour. If a later enhancement (e.g. Auto Adapt) happens afterwards, the value in this field is removed since it no longer applies.

Available as of March 18, 2026

Phrase Language AI is now available directly in Salesforce Service Cloud through the Verbis integration.

Support agents can translate chats and cases in real time without switching tools or copying content between systems. Agents can work entirely in their own language, while customers receive responses in theirs. Automatic language detection helps ensure conversations continue smoothly.

Some REST API requests to the connector service timed out or failed with intermittent 503 and 504 errors. This has been fixed.

Phrase Data:

Four new columns have been added to the lqa_v2 table: job_id, job_uid, workflow_step_level, and job_part_inner_id. These fields enable reliable joins between LQA data and job-related tables (for example, job_part_v2) in scenarios where job_part_id and job_part_uid may be inconsistent due to sharing.

When joining lqa_v2 to job_part_v2, project vendors and project buyers should use job_uid (or job_id), workflow_step_level, and job_part_inner_id. Providers assigned to jobs should use job_uid and job_part_inner_id only, as access to the relevant workflow step may not be available

Tridion: Import was failing with UNKNOWN_ERROR and APC created more projects. This has been fixed.
Zendesk: The authentication window was staying open and TMS appeared twice. This has been fixed.
Job Sync: The due date of Strings jobs was not being imported via APC. This has been fixed.

Have you come across any bugs?

Thank you for letting our Support team know.