Skip to main content
Commander is Army of Agents’ company-scoped coordination assistant. Use it when you want to ask about company context, plan work, inspect tasks, use skills, or trigger governed actions from a conversational cockpit. Commander does not bypass company access, role permissions, approval gates, tool trust rules, memory visibility, or budget controls.

​
When to use Commander

Use Commander to:
  • ask questions about company work and context
  • summarize tasks, discussions, goals, and memory
  • plan the next set of tasks from a vague request
  • invoke skills attached to Commander
  • inspect your accountable work through the Cockpit
  • trigger governed actions when your role allows them

​
Start and verify Commander

1

Open Commander

Select Commander from the sidebar while the intended company is active.
2

Check runtime setup

If the selected runtime is not ready, follow the setup prompt to authenticate and verify the local CLI or configured provider path.
3

Send a small context question

Ask Commander to summarize the current company, open tasks, or your accountable work. This verifies streaming and context access before you ask it to take action.
Connection verification reports whether the configured runtime and authentication are usable. A successful sign-in does not grant Commander permissions beyond the authenticated operator context.

​
Use the Cockpit

The Cockpit summarizes work accountable to the current user:
  • Mine: work directly assigned or accountable to you
  • Managed: work in your human and agent responsibility hierarchy
  • Awaiting review: work that expects your review
If one source fails, Commander marks the result partial instead of presenting an all-clear state.

​
Compose a useful turn

Good Commander requests include the outcome, scope, and constraints.
The composer supports:
  • @ mentions for company agents
  • / skill tokens or skills selected from the add menu
  • up to five supported attachments
  • retry, edit, and discard after a failed request
Plain text, Markdown, and JSON attachments can contribute text to the runtime turn. Images and PDFs are stored but are not currently shown to the model.

​
Tools, permissions, and trust

Commander tools remain company-scoped and pass normal role and entity checks. Governed actions can stop for confirmation. Tool permissions determine which capabilities are available; trust rules can remember an approved decision for eligible repeated actions. Review or remove trust rules when the operating boundary changes.
Treat trusted Commander actions like product permissions. If the context, role, or workflow changes, remove old trust rules and approve the action again.

​
Questions, review, and completion

Agent questions, review requests, task state, and run completion are separate concepts:
  • a work question can appear in Commander, Inbox, Task Work, Workspace, or its source Discussion
  • answering a question can request continuation of parked work
  • a technically completed run does not automatically mean the task is complete
  • review and acceptance follow the task’s assigned reviewer and completion policy
Use Awaiting review for accountable review work. Check the task and its output before inferring completion from a streamed Commander response.

​
Troubleshooting

Confirm the skill is installed, attached to Commander where required, and allowed by the current tool manifest.
Check company selection, role permissions, page context, department context, and memory visibility.
This is expected for governed actions. Review the proposed action and approve only when the scope is correct.

Company Brain

Understand the memory context Commander can use.

Commander API

Review request and stream contracts.