Create and configure a background agent
This guide creates an agent definition for your workflow. To follow one chosen example from creation through a real tool call, use Create your first background agent.
Before you start
You need a role with Create access to Background Agents and an enabled runtime under Agent Types. Open Build > Background Agents.
Create an agent
- Select Create Agent.
- Enter a Name and an optional Description that states the job.
- Choose the Platform. Manage beside the field opens organization settings if the runtime you need is unavailable.
- Choose the Channel. Select None for an agent that does not need Slack; selecting Slack also requires its own app connection before it can respond there.
- Select Create agent.
You can choose an avatar in the same dialog. The platform choices are limited to those enabled for your organization; use the platform reference for setup requirements.

Use a ready-made agent
- Select Discover on the inventory, or Browse ready-made agents from the creation dialog.
- Search or choose a category. Only with tools I have narrows the catalog to your available tools.
- Open an agent to inspect its required and optional integrations, tools, skills, and instructions.
- Select Use this agent and complete the creation flow.
- Review its capabilities and connect the integrations it needs before deploying.
A template's connected-integration labels help identify prerequisites. Confirm the authentication used by the new agent itself in its Tools tab.

Create with AI
Enter the name and a concrete description in the creation form, then open More create options beside Create agent and select Create with AI. This option also picks tools and skills from the name and description. Review the resulting instructions and capabilities before connecting accounts or deploying.
Configure the definition
On Overview, enter the System Prompt and select Save. Describe the job, the inputs the agent needs, and the result it should return.
Open Tools to assign and authenticate the tools. Add skills through Knowledge if the job needs reusable instructions. Under Settings > Platform Configuration, check the runtime's model and other required configuration, and save any changes. For Claude Managed Agents and Willow Agents, a Reasoning effort dropdown below the model picker sets how much effort the model spends. It stays on Model default until you pick a level, and it is hidden for models without an effort control. Cursor Cloud Agents show the model's own parameter dropdowns in the same place.
For external clients, use the agent's Connect dialog to obtain its endpoint and a manual token from Settings > Tokens for gateway authentication. A secret is shown once at creation or when you create or recreate a token. Store it before closing; deploying does not invalidate manual tokens. See Agent credentials.
Keep the agent scoped
Assign the tools the job requires and review any capabilities selected by AI or a template. A prompt requesting read-only behavior does not remove write tools from the agent's capabilities; remove those tools if the job should only read.
Deploy it
The definition is ready when the instructions are saved, the intended capabilities are assigned, and each required integration has an authentication identity. Next, deploy and test the agent. An Active badge alone does not confirm deployment or working tool access.