An agent is an assistant that belongs to your organization. It chats, it acts in the apps you’ve connected, and it runs Agent Automations on its own. Think of each agent as a role you’ve hired for, not a tool you open.

​
What makes an agent an agent

Each agent has its own:
  • Instructions: prose you write telling it who it is and how to work. These apply to everything it does
  • Connections: the apps it’s authorised to act in. Connections belong to the agent, so anything the agent does can use them
  • Agent Automations: its standing jobs
  • History: the record of every run
Agents are separate on purpose. Your finance agent’s connections and instructions don’t leak into your support agent, and vice versa.

​
One agent or several?

Every organization starts with one. Add more when a distinct part of the business wants its own instructions and its own apps: one per function is the common shape. Switch between them in the agent rail on the left edge, and create another with the plus button.

​
Who can use an agent

Agent access is granted per member. Sharing an agent with a colleague gives them the full agent: its chats surface, its Agent Automations, its connections. Two colleagues can see entirely different agents, and that’s by design. Admins can see the full roster of every agent in the organization under Settings → Agents.

​
Archiving

Archiving an agent shelves it: it leaves the rail and the roster’s active list, but its Agent Automations keep running and keep spending credits. If you want work to stop, pause the Agent Automations themselves. See Pause, resume, and run now.