DeepAgent capabilities

When DeepAgent is enabled, the Supervisor Agent gains access to additional capabilities that help it coordinate sub-agents and tools more effectively.

These capabilities are available when using DeepAgent:

  • Token-level streaming in DeepAgent
  • Sub-agent activity display

Token-level streaming in DeepAgent

DeepAgent supports token-level streaming. When token streaming is enabled, the GenAI Assistant UI shows responses as they are generated.

To enable token-level streaming, include the "streamMode" variable in the chat request:

  • "streamMode": "token"

    Enables token-level streaming. Partial AIStreamMessage events are emitted with partial: true as the supervisor model generates its response.

  • "streamMode": "complete"

    Returns the full response message before it is displayed. This is the default behavior.

Note: Token-level streaming applies only to the supervisor model responses. Sub-agent responses and internal tool LLM calls are not streamed token by token.

Sub-agent activity display

In the GenAI Assistant, the sub-agent menu header is dynamically updated to show the most recent action performed by the sub-agent. This helps you see the sub-agent thought process in real time.