Jarvis guide
Try “supplier bill”, “overdue invoice”, “approval”, or a feature name.

Learning and review

Let Jarvis suggest useful patterns, then decide what may become a shared rule.

For
Jarvis administrators and knowledge reviewers
Availability
Managed learning depends on workspace and plan
Reading time
2 min
Verified
2 August 2026

In one sentence

Learning looks for repeated ways your team works and puts suggestions in front of a reviewer.

What it does

When available on a managed workspace, Jarvis can review a limited amount of recent history during a chosen overnight window. It proposes patterns such as a repeated classification or approval habit.

It does not silently turn every pattern into a rule. Suggestions go to Review, and approved changes still need an explicit Apply before they reach Jarvis.

Analysis

An administrator can open Skills > Analysis to:

  • Turn behavioural learning on or off.
  • Choose the daily analysis window.
  • Limit how many findings can be raised in a run.
  • Run an analysis now.
  • Check the latest run and the next scheduled time.

Self-hosted and some plan types may not include this managed overnight analysis.

The Analysis area with behavioural-learning settings, the latest run and findings summary, custom-app learning guidance, and a link to Personalise.
Analysis keeps the learning window, run coverage, findings, and custom-app boundary visible to administrators.
Learning and review in 30 secondsInspect behavioural-learning coverage, move to the Review queue, and return to Analysis without starting a run.Download MP4

Review

Reviewers can see proposed Skill candidates and requests to widen knowledge or Skills beyond one person. A reviewer can edit and approve, snooze, reject, or ask the named user. Approved Skill suggestions remain pending until someone chooses Apply learned skills. The decided log keeps earlier review outcomes visible.

Personal findings can become a question for the named user instead of company-wide knowledge.

Wiki and Skill promotion requests have their own review queues. Approving a wider audience is a deliberate decision; it is not the same as agreeing that a personal note was useful once.

Learn from a custom app

Administrators can install the Custom App Learning Agent. A run reads only the apps selected for that run and writes business-focused pages to the company Wiki. The selected source is sent to the AI provider connected to the site, so choose only apps whose code you trust.

A real example

Jarvis notices that freight bills for the South warehouse are repeatedly assigned to the same cost centre after manual correction. It proposes a rule. The finance reviewer checks the examples, narrows the wording, approves it, and applies it as a learned Skill.

A repeated habit is not automatically a good rule. It may be a workaround or a mistake. Review the examples before approving it.

Try a realistic task

Follow Learn how your custom app is used.