Typesafe Has a New AI Update for Testing, With Optional Extras

Typesafe has released Jev 1.14.0 for preview testing. The company announced it on Oct. 14, 2026, and said it brings small improvements in many areas, with the biggest gain in support for many languages Typesafe.
The preview label matters. This version is for testing only, not the standard version for live products.
This release adds experimental "danger zone" switches, like light switches you flip on only for testing. They are special settings sent with a software request to turn on unfinished features Typesafe does not approve for live use. If you send no switch, nothing changes for you.
One switch turns on image support with TypeSafe-Danger-Zone-Images: true. Typesafe said it works by find-and-replace in stored files, called assets, and can be used in state, instructions, or criteria. The company said the earlier model jev-1.13.0 could already handle images and text, so this continues that work. A new box type, which returns a single rectangle showing where something is in a picture, works only in jev-1.14.0.
Longer context, or how much text the system can look at at once, has its own switch. You turn it on with TypeSafe-Danger-Zone-Long-Context: true, without changing the normal setting for others. Typesafe also added a switch to share user questions to help find jaggedness, or uneven answers. The company said it will still not train on those questions. The goal is to find problems, not reuse data.
The broader context here helps explain the plan. In my view, Typesafe is keeping the regular setup steady while testing new ideas behind switches. That means less extra work for most users and extra risk only for those who opt in.
Looking at what this means if you test the preview, keep tests separate. Test images, longer text, and the single-rectangle output alone, because combined switches make problems hard to track. Test other languages alone too, since that is the main claimed gain. Sharing questions needs a clear decision, even with the no-training promise. Live use is out of scope. The value now is in small tests you can undo, like trying images only in criteria, trying longer text on a little traffic, and seeing if one rectangle is easier to use.


