LoginSign Up
← Back to all sparks
L

Lokalise

DEVOPS
Velocity6.3

Lokalise is building the measurement layer around AI translation, not just more AI translation

ai-translationquality-analyticstranslation-memoryworkflow-automationaudit-logsapi-platform
◆Current state
Releases land weekly, tagged by type, across three tracks: AI translation options (Gemini 3.5 Flash joining GPT and Claude as selectable models), workflow plumbing around translation memory and review scope, and quality measurement — Translation Quality Analytics, per-contributor review metrics, and AI scoring. The most recent change makes AI scoring opt-in rather than something that fires on every Pro AI task.
◆Where it's heading
The movement is away from adding AI and toward accounting for it. Post-edit rate, edit distance, acceptance rate and AI scores are now first-class reports, while workflow settings increasingly let translation-memory matches pass through untouched so human review concentrates where it changes the result. Making AI scoring explicit fits that: a background cost becomes a deliberate step. The audit logs API points at a second audience entirely — the security and platform teams who approve the tool rather than use it.
◆Prediction
The new api.lokalise.com/v1/ path is stated as the home for all future public endpoints, so more surfaces should appear there while /api2 stays frozen. Translation Quality Analytics and Glossary Guard are both flagged pre-release, and general availability is the obvious next beat for each.

◆Recent moves

  1. 15d ago

    AI scoring in workflows is now opt-in

    AI scoring stops running automatically on every Pro AI task and fires only when the workflow template includes it. It reads as cost discipline as much as user control — teams relying on the implicit behavior have to move to one of the scoring templates to keep it.

  2. 20d ago

    Gemini 3.5 Flash is now available for AI translations

    A third selectable engine alongside GPT and Claude, with no price difference between them. Model choice is settling into a neutral setting here rather than a differentiator — which is precisely why the quality analytics work exists: to tell you which one is actually performing on your content.

  3. 20d ago

    Contentful Native: faster imports, smarter reference handling

    Bulk push for deeply nested references, scheduled imports that behave like manual ones, and errors that surface instead of jobs failing quietly. Connector reliability is unglamorous work, but it determines whether the AI and review workflows ever get content to operate on.

  4. 27d ago

    Stronger account protection when changing your email

    Notifications to the original address both when an email change is requested and when it completes. Minor on its own, but it lands the same day as the audit logs API — both aimed at the account-security review rather than the daily user.

  5. 27d ago

    Audit logs API: pipe your audit trail anywhere

    ⚡ SPARK

    Audit events leave the UI and become programmatic for the first time, and they arrive on a new api.lokalise.com/v1/ path declared the home for every future public endpoint. The second fact carries further than the first.

  6. 1mo ago

    More control over TM matches and review scope in Workflows

    Apply TM steps can now set the status of matched translations directly, and review tasks can exclude segments already marked Reviewed. Together they let translation-memory matches flow through a workflow hands-off, which is the mechanical half of the quality-measurement push.