Agents

Admin → Agents is where you register the agents your organization uses and decide who can run each one and what it's allowed to do. You also check an agent's status and edit, restart, or delete it on the same screen. Only people with the Owner or Admin role can use it. Personal agents that people create for themselves don't appear in this list.

The agent grid

Agents appear as cards. Each card shows the agent's name, its role, a scope badge (Global or the name of its team), model information, and a What this agent can do: summary of its permissions. When the agent's status can be checked, a status dot appears too. The top of the screen shows the counts for All agents, Global, and Team only, and you can filter the list to All, Global only, or Team only.

Creating an agent

Click Create agent to open the What should this agent do? window. Describe the work in plain language and click Create draft. The AI fills in a draft name, role, and work instructions, and marks the permissions it thinks are needed as Recommended, with a reason. To fill everything in yourself from the start, click Enter manually.

In Review and refine, check the Agent name, Role, and Work instructions, set the Creation scope to Global or Team (for a team, pick the team too), then click Create agent. Ian looks up agents by Role when it assigns work. All permissions start off, so check the Recommended marks and turn on only the ones you need yourself. Knowledge access starts at Assigned team only. A new agent is created with Inherit to child teams turned on; you can change that afterwards in the card's Scope settings.

Scope and permissions

Select a card to open its settings window. In Scope settings, set Scope to Global (all users) or Team (one team only), and for a team, pick it in Select team. Turn on Inherit to child teams and the agent is also visible in that team's child teams. Permissions come in three groups.

In Knowledge access, set Knowledge base access to one of Assigned team only, All teams, Global only, Custom, or No access.

Tool permissions has settings for Browse the web, Run code, Access files, Call external APIs, and Query databases. You can turn each one on or off.

Operational limits has the fields Maximum tokens per request (leave blank for unlimited), Maximum tokens per day (leave blank for unlimited), and Maximum concurrent sessions (leave blank for unlimited).

After changing the scope and permissions, click Save. The changes are recorded on the Audit screen.

Status and operations

When the status can be checked, the top of the settings window shows one of Active, Idle, Busy, Stopped, or Error. AI model information shows the Current process model, Configured model, and Backend. If the current model can't be checked, it shows Current process unavailable; if no model has been set for the agent, it shows Automatic.

From the ⋮ menu at the top right of the window, choose Edit, Restart, or Delete. In Edit, you change the Name, Role, Work instructions, and Tools (comma-separated) fields and turn Modify files, Delegate to other agents, and Keep session on or off. Restart and Delete each ask you to confirm once more, and a delete can't be undone. For an agent the runtime isn't running, this menu is missing or shows only Delete.

Common situations

• A new team can't see an agent — in the card's Scope settings, check which team the agent belongs to (Select team) and Inherit to child teams. An agent that belongs to a parent team shows up in child teams only when Inherit to child teams is on.
• The list won't load — click Try again next to the Couldn't load the agent list. message. If you see Runtime unreachable, only the registered agents are listed, without their status, and the Edit and Restart menu items don't appear either.