Agents

An Agent is a reusable project identity and configuration that can be assigned to Workspaces and automated work. Open Advanced → Agents to review each agent's memory, security mode, status, description, and health.

Usually managed automatically

Workspace setup automatically creates and configures the default Agent. Most teams manage its day-to-day behavior through Personalization in the Workspace UI. Use this page when you need to inspect how that Agent connects memory, session records, permissions, and execution Flows, or when you are creating an additional project Agent.

⤢Expand

Screenshot previewThe light-mode Agents page with the active default Workspace agent

×

Select an agent name to edit it. Select New Agent to create another agent.

Agent settings

Section	What it controls
Agent Name, icon, and Description	Identify the agent to users and administrators. Use the description to state its role and boundary.
AGENT.md	Supplies the agent's durable operating instructions. Keep credentials out of this text.
Persisted Memory	Links an Index used to search long-term memory. The Agent's Documents folder stores durable memory files and per-session logs that can be indexed for later retrieval. The page can create or repair the folder and Index.
Security	Defines the Agent's permission boundary and chooses whether execution uses the initiating user's access or an independent agent identity.
Workspace Defaults	Selects the Flow used for Agent execution, the scheduled Flow, primary Skill Collection, and Flow-property defaults used when the Agent works in a Workspace.

The list summarizes Memory, Security, and Status. Health warnings can identify an agent with no Flow, no background approver, or no mapped identity user.

Agent identity and security

Mode	Identity and access	Choose it when
Use initiating user's security (copilot style)	Each run uses the initiating user's effective Connection, Skill, and project permissions. Different users can therefore give the same Agent different access. Actions that require user approval remain tied to that interactive user's security context.	The Agent assists a signed-in person and should never have broader access than that person.
Use independent agent identity	The Agent runs as a dedicated identity with its own groups, resource permissions, status, background approver, and notification configuration. Its access remains consistent regardless of which user or schedule starts the work.	The Agent performs scheduled, shared, or unattended work that needs a stable, explicitly reviewed permission boundary.

Independent identity does not mean unrestricted access. Administrators must keep the identity active, map its identity user correctly, choose a background approver, and grant only the required resources. Connection tool enablement and approval policies still apply in both modes.

Only account or project administrators and owners can manage the sensitive security controls. Suspend an identity before investigating unexpected access; archive an Agent only after checking Workspaces, API keys, queues, templates, and scheduled work that reference it.