How a person stays in control

An AI team does the repetitive work. The decisions stay with you. This page lists the places where Agentivity stops and lets a person decide.

• The Assistant works on a draft. What it builds is not in your workspace until you select Save. Before you save, you see what will be added and where it will go. If you remove the draft, nothing in your workspace changes.
• You publish your changes yourself. In an editor, a change is saved when you select Publish.

• A step can wait for a person. In a workflow, you add a step that sends a question and stops. The workflow continues only when someone has answered. See Ask a person to approve or answer.
• An agent can ask before it decides. An agent that has a channel to reach a person can stop and ask for a choice or an approval.
• You can stop an execution. The bar above the canvas has Pause, Resume and Stop. The section Executions can stop several executions at once.
• An execution starts when a person starts it. Nothing starts by itself. Starting executions on a schedule is planned, and it is not available yet.

• Every execution is listed. The section Executions shows the executions of your workspace and their state. An execution that waits for a person has the label ACTION REQUIRED.
• Every step can be read. In the Execution view of an editor, the Inspector shows, for each step, what it received and what it produced.

An AI model can answer differently to the same request. When a rule must always give the same result, do not leave it to a model: put it in a workflow, as a fixed step. A condition, for example, always takes the same path for the same input. See Add a condition to a workflow.

Agentivity has no login today. It does not know who is in front of the screen, so it cannot check who approved a step. Anyone who can reach your server can build, run and answer. Keep the server on a network you trust, or behind your own access control.

What the Community edition includes.