Until this week, "translate" was one step. If you wanted a human to edit the output, you ran a separate job. If you wanted an AI to re-check that the human respected your glossary, you wrote a custom scorer. If you wanted to catch semantic drift, you wrote your own back-translation harness. The localization pipeline collapses all of that into optional stages around the core translate step – each one independently toggleable per engine and overridable per job. Turn on what you need, leave the rest off.
The pipeline stages:
- Pre-localization AI edit. An AI agent cleans the source payload before translation, so a single source error doesn't propagate across every target locale.
- Post-localization human edit. Sends the translation to a qualified human translator. The job pauses on a webhook until the edit returns.
- Post-localization AI review. Reconciles the human output against the engine's glossary, brand voice, and instructions.
- Back-translation check. Translates the final output back into the source locale and compares. The agent flags semantic drift by severity and auto-adjusts on major or critical drift.
The same week, engine provisioning ships – an async API that takes a fresh engine plus a few URLs or content samples and auto-configures the brand voice, glossary, and instructions for you. Multi-step manual setup becomes one call.
Also shipped#
- Pipeline configuration in the dashboard. Toggle stages alongside the rest of the engine config.
- Daily token limits per org. Cap that prevents runaway usage from a misconfigured integration.