Core Concept - Agents
An agent is a persistent AI worker with instructions, conversation history, files, and configured tools. It starts work in response to messages, schedules, and supported events; it does not run an uninterrupted reasoning loop.
The Agents page shows the current workspace's agents. Check the My workspace or Team label in the sidebar and breadcrumb. If the workspace is empty, Browse Teams helps you find agents in other Teams you can access. See Workspaces And Navigation.
They can:
- Browse websites
- Click, type, scroll, and select elements
- Extract structured information
- Fill out forms
- Download files
- Make reasoning-based decisions
- Use multiple tabs
- Maintain state inside long workflows
Browser work depends on the website, authentication, available tools, and deployment configuration. Connected apps provide additional workflows; for example, you may connect Google Sheets and grant access to a spreadsheet the agent should update.
Recurring work requires a configured schedule or event source. An agent's active state, credentials, approvals, and usage limits determine whether it can process a trigger. Use Stop, Pause, And Restore to manage its lifecycle.
You can communicate through web chat and configured email channels. Channel availability and contact permissions apply. The REST API and Remote MCP support documented programmatic workflows.
When creating an agent, enter an Agent name such as Finance Review, or leave it blank for a random name. Organizations can require names, in which case the field must be filled before creation, including when launching a template. Chat labels show the complete name. You can rename an existing agent in its settings.