Before you start
You should have:- completed Quickstart
- created at least one company
- created at least one agent during onboarding or from Team
- verified that the selected adapter can run on your machine
Agent execution is push-based. Army of Agents wakes an agent through a heartbeat; the agent then checks context, checks out a task, and reports progress through the API.
1. Confirm the agent is ready
Open Team or Agents and select your first agent. Check that:- the agent is assigned to the correct company and department
- the adapter is configured
- the agent is not paused or terminated
- any required CLI is installed and authenticated
- the monthly budget is not exhausted
2. Create a small task
Open Tasks and create a task with a small, verifiable outcome. For a first run, choose work that can be completed without external accounts or secrets. Example task:3. Wake the agent
Use one of these paths:- click Invoke on the agent
- assign the task and let assignment wake the agent
- mention the agent in a task comment if mentions are enabled
- wait for the agent’s configured schedule
4. Watch the run
Open the task or the agent’s run history. A healthy run usually moves through this shape:5. Verify the result
Check the task for:- a new comment or output from the agent
- status movement such as
in_progress,blocked,in_review, ordone - a run summary with duration, token usage, cost, and detected files when available
- any work question if the agent needed human input
Troubleshooting
The agent did not start
The agent did not start
Check agent status, budget, adapter configuration, and whether a hire approval is pending in Inbox.
The task stayed unclaimed
The task stayed unclaimed
Confirm the task is assigned to the agent and is in a dispatchable status. Planning tasks are not dispatched until the configured approval path releases them.
The agent asked a question
The agent asked a question
Open the work question from Inbox, Commander, the task, or the source Discussion. Answering can request continuation of the parked work.
The run failed
The run failed
Open the run output and adapter diagnostics. Local CLI adapters usually fail because the CLI is missing, unauthenticated, blocked by permissions, or launched from an unexpected workspace.
Next steps
How agents work
Learn the heartbeat protocol and execution model.
Tasks and reviews
Learn how to review, unblock, and complete work.