Jev 1.14 opens multimodal danger zone

Jev is evolving beyond narrow text decisions, but TypeSafe’s danger-zone framing makes clear these features are promising previews rather than production guarantees.

• –Multimodal inputs let developers apply Jev’s typed decisions to images without changing the core API.
• –The new box output type points toward visual localization and lightweight computer-vision workflows.
• –Longer context expands possible use cases, while TypeSafe openly warns that reliability degrades as context grows.
• –`jev-preview` gives builders access to improvements before they reach the stable `jev-latest` channel.
• –Opt-in query sharing could accelerate debugging without using customer data for training.