Skip to content

Internal Agent Responses Leaked in SSE Output with partial: true #4152

@abhijith2599

Description

@abhijith2599

When using Google ADK (google-adk==1.17.0) with SSE streaming over FastAPI, the agent is emitting internal or intermediate messages in the response stream. These messages are marked with partial: true and appear to include internal thinking or non-final agent output, which should not be exposed to the client. The expected behavior is that partial SSE responses should only contain user-safe, externally visible content, and internal reasoning should never be streamed regardless of the partial flag.

Metadata

Metadata

Assignees

No one assigned

    Labels

    web[Component] This issue will be transferred to adk-web

    Type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions