Actions and results

Everything the agent does, it does through actions: create a part, wire a circuit, run a simulation, take a screenshot, open an item. Each action shows in its reply as a card with the action's name.

Card states

Icon	State	Meaning
Gear	Running	The action is running now.
Hourglass	Pending	Waiting for your approval.
Tick	Done	It ran and succeeded.
Red mark	Failed	It ran and returned an error. The agent reads the error and usually tries again.
Crossed circle	Cancelled	It never ran: the turn was stopped, the action was rejected, or you sent a new message first.

A failed action is normal: it is how the agent learns that a script did not build or an input was wrong. Only worry when the same action fails again and again; then stop the agent and describe the problem differently.

Inputs and results

Click a card to open it. It shows the Input the agent sent and the result that came back, as text. This is the quickest way to see exactly what changed, or why something failed.

Results you can see

Some results add a preview under the card:

• Workspace: a thumbnail of each part the agent built, and View on Canvas after it changes the assembly.
• Schematic: a preview of each new part, a card for each measurement it added, and View on Canvas after it changes the circuit.
• Workspace and scene: screenshots and recordings the agent takes of the view.

Files an action produces, such as screenshots, show as images; click one to see it full size.

Background jobs

Long work runs as a job on the server: physics checks, G-code, splitting a STEP file, and builds that need more than the editor allows. Its card shows Queued, Queued #2 (its place in line) or Running, then the result. You can keep working, or leave and come back; the result is in the chat when it is done.

Actions in the window

Some actions run in your browser window instead of on the server: screenshots, simulations, reading the circuit's measurements, reading this guide. They need the page to stay open. In a hidden or minimised window they can be slow.

Approvals

In a workspace or a scene, with Require approval for AI actions on (the shield in the message box), the agent stops before each change and its card shows Pending. The bar above the message box offers Approve or Reject, or Approve all and Reject all when several are waiting. A rejected action does not run, and the agent continues from there.

Approvals suit careful edits of a design you care about. For a first draft, leave them off and use History to roll back.