Agents are reusable AI teammates with their own instructions, skills, workspace access, and optional schedules. Use an agent when the same role or process should continue across conversations.
Start with a blueprint
Open Workspace → Agents → Blueprints, choose a role that fits the work, and review its description before installing. A blueprint is a starting point; you can change the agent's name, instructions, model, skills, and access later.
Choose where the agent belongs
Install the agent for your personal work or for an organization you belong to. The agent remains owned and managed by the person who installed it unless the product explicitly shows a different ownership model.
Review workspace access
An agent can use only workspaces available to its owner and allowed in the agent's Workspaces settings. Narrow the list when the agent needs only one project or team space.
Use the control panel
Overview — recent activity and schedules.
Settings — name, description, model, instructions, run limits, and memory options.
Memory — durable facts you intentionally save for the agent.
Skills — reusable procedures and guidance.
Pinned — resources the agent should consider regularly.
MCP — external tools the agent may use.
Workspaces — the workspaces the agent may access.
Start and review work
Select the agent from Copilot or Chats, give it a clear outcome, and attach the resources it should use. Review the conversation or activity history to confirm what completed; do not assume a resource changed merely because a task was requested.