Back to the changelog

• Feature

  Human fallback inside the task

  In AI, Script, Wait and Sub-process tasks you can define when to ask for a human decision, who makes it, instructions and allowed outcomes. The person finds the work in the To do queue and the dashboard widget, calendar and Kanban, chooses an outcome and adds a note when required. The reason is also available through the keyboard and screen readers. The same task continues with the existing rules. Transient errors first use the configured retries.

• Feature

  Control over suggestions

  An Admin can hide the model suggestion to avoid influencing the decision; individual tasks can make the same choice. When visible, probability and confidence remain explicit and sensitive fields stay redacted. The same rule applies to the API, MCP and extension.

• Design

  From manual to Jev, keeping human fallback

  “Keep the manual decision” preserves the original instructions and outcomes in the Jev task’s fallback: it no longer creates a review phase. The flowchart and PDF mark tasks with fallback enabled. The Triage scenario demonstrates the new behaviour.

• Feature

  Measuring decisions

  Analytics and KPIs distinguish automated and human decisions and calculate the fallback rate. Amending an outcome does not count as a new decision. Earlier tasks without attribution stay separate, without guessing who decided.