Agent Controls

Stop individual agents or all agents during an emergency.

Agent Controls shows every agent's runtime state. Use it to stop current work in one project. You can stop one agent or every agent.

Use it when:

• An agent behaves unexpectedly.

• A scheduled load test needs intervention.

• You must stop active work before changing configuration.

Open Governance → Controls & Policies → Agent Controls.

Agent Controls, with emergency stop controls and per-agent runtime status.

Key terms

Term	Meaning	What you can do here	Scope
Agent	A configured AI worker with an identity, skills, knowledge, triggers, and integrations	Stop its active runs	One agent
Project	The boundary for the current work	Stop active runs	One project
Run	One execution after a trigger or request	Ends when complete or stopped	One execution

Agent Controls does not build, edit, disable, delete, or deactivate agents. Switch projects to review or stop work elsewhere. This view has no cross-project scope.

Stopping work only interrupts current runs. The agent runs again after its next trigger.

Stop work safely

Need	Action	Confirmation	Result
Stop one agent	Select Stop for that agent	None	Interrupts that agent's active runs
Stop every agent	Select Stop all agents	Required	Interrupts active runs throughout the project

The single-agent action takes effect immediately. The project-wide action asks you to confirm the affected project.

Narrow the agent list with Search agents… before stopping work. This reduces the chance of selecting the wrong agent.

Stop a running agent

For an agent, open More Actions (⋯). Choose View Agent or Stop. These actions remain available when activity shows Completed or Unknown.

View Agent opens the agent in a new browser tab at /platform/agent-builder/<agentId>?…. Configure its graph, skills, triggers, knowledge, and integrations in Agent Builder. Your current activity view stays open.

Stop interrupts the agent's active runs. Work in progress is lost. If none exist, the request has no effect. The service still sends the request and displays Stopping the agent's executions.

Stopping the agent's executions can remain visible and can appear twice. It does not indicate ongoing work. The request runs once.

Stop every agent in the project

Select Stop all agents to interrupt active runs throughout the current project. The confirmation states in this project and warns that the action cannot be undone. Select Cancel to leave work running. New runs can start immediately after stopping.

The confirmation applies only to the current project. To stop work elsewhere, switch projects and repeat the action.

Find agents

Use Search agents… to find agents by name. Matching ignores case and accepts part of a name. The filter is saved as ?search=<term>, so you can bookmark or share it. Search covers the full project, including agents beyond the default first page.

Unlike Search, changing the page does not update the URL—page state is client-only, so a specific page cannot be bookmarked or shared. Use Search to narrow the list when you need a shareable view.

Choose the right destination

Use these destinations for work that does not stop active runs:

Goal	Destination	Available action	Limitation
Build or edit an agent	Agent Builder	Change configuration	Does not stop work automatically
Stop selected agents	Agent Controls	Stop each agent separately	Selected-agent bulk stopping is unavailable

Related

Title

Governed Graph

Description

Explore how agents, apps, tools, and data connect

Link

Governed Graph

Title

Agent Builder

Description

Configure agents and their connected resources

Link

Agent Builder

Title

Glossary

Description

Look up agent, project, run, and related terms

Link

Glossary