Skip to main content
简体中文 DESIGN → Agents manages identity, behavior and execution. Reuse an Agent in Chat, Issues, Teams and Endpoints. Saving its definition does not start work.

​
Choose an execution type

The work entry points are shared, but models, tools, recovery and configuration projection differ. Catalog visibility does not guarantee every conversation or dispatch capability. See the dedicated Managed, Hosted and External guides.

​
Create an Agent

  1. Start creation and enter its name, purpose and Instructions.
  2. Link a Workspace when it should reuse operating guidance, skills, tools or subagent definitions.
  3. Explicitly choose Runtime under Execution. An online Hosted provider may be selected initially; choose AgentScope Managed for a Managed Agent.
  4. Leave Model empty for the runtime default, or enter a model identifier supported by that provider.
  5. For Managed Agents, choose an Environment in Advanced settings. The automatic local default requires administrator permission for Local execution.
  6. Select Create & open agent, then review the resulting configuration.
Agent key is a stable identity within the scope. The display name communicates purpose to colleagues.

​
Configure the detail pages

​
Add capabilities incrementally

Verify conversation first, then file access, an external tool and a specialist skill. Store credentials in Vault and shared knowledge in Memory. Configuration does not automatically install executable tools or grant external permissions. A shared Workspace can affect multiple Agents. Check its consumers before editing and verify with a new Chat or Issue. Do not assume that running work switches definitions immediately. Next: Teams · Workspaces.