v0.89.0
September 23, 2026
**Jev joins the engine:** an AI task can now simply *decide* — classify, estimate > urgency, check whether something essential is missing — with a dedicated, very fast model > that answers with a probability instead of writing text. The rules you already know take > care of the rest.
-
Feature
A model that decides, not that writes
The AI task now has a second mode, **Jev**, alongside the familiar free-prompt one. Three question TYPES — true/false, a choice among options, a score —, up to 20 questions per task, and the answer arrives in a fraction of a second, with a probability that says how confident the model is. You decide how low that confidence can go before an answer counts as "undetermined" instead of being taken at face value, and send those cases to a human review instead of routing them as if the model were certain. You need a Jev provider (TypeSafe or OpenRouter, see below); the model performs best in English, even when the item's data stays in another language.
-
Feature
Try it before you publish
In the designer, the **Try** button really runs the question on an item you pick and shows you, answer by answer, what would get written — without touching anything. It is the very same engine the real task uses, with its own row in the AI call log, so you can tune thresholds and criteria before going live.
-
Feature
A demo scenario to see it in action right away
The new **"Request triage"** scenario (Admin → Organizations → Demo Data tab → Realistic scenarios) routes six sample support requests among technical support, back office, needs-more-information and manual review, with a single Jev task and three questions. It installs (and can be removed) from the same screen even without a Jev provider configured, ready for a live demo as soon as you add one.
-
Integration
A Jev provider ready in two presets
In Settings → AI settings, the **Jev / System One** category offers two ready-made presets — **TypeSafe (Jev)** (`https://api.typesafe.ai`, model `jev-latest`) and **OpenRouter (Jev)** (`https://openrouter.ai/api`, model `typesafe/jev-1.13`) — with the provider key and a **Test connection** button to verify it right away. FlowSharp automatically uses the organization's first active Jev provider for tasks in this mode, and a dedicated panel sets the confidence threshold and review outcome for everyone. Note: the model performs best in English, even when the item's data stays in another language.