Problem
stream_agent_response already emits workflow updates for reasoning summaries, allowing clients to display the model’s thought process as it unfolds.
However, when the agent invokes the web_search tool, no workflow events are produced. This results in:
- Reasoning steps progressively appearing ✅
- No information about web search tool being called
Users could perceive this as a freeze or unexplained pause.
Requested Enhancement
Treat web_search_call items as first-class workflow tasks.
As search calls begin and progress, emit:
WorkflowTaskAdded when a search startsWorkflowTaskUpdated as the status evolves (loading → complete)- Include metadata such as:
- Active query or URL
- Current status indicator
- Discovered or referenced URLs
Why It Matters
- Consistent streaming UX — Reasoning steps and search steps should appear in the same timeline.
- User clarity — When the assistant pauses, users should see “Searching for X…” instead of guessing.
- Immediate access to search metadata — Clients can render queries and links without waiting for final content.
Problem
stream_agent_responsealready emits workflow updates for reasoning summaries, allowing clients to display the model’s thought process as it unfolds.However, when the agent invokes the
web_searchtool, no workflow events are produced. This results in:Users could perceive this as a freeze or unexplained pause.
Requested Enhancement
Treat
web_search_callitems as first-class workflow tasks.As search calls begin and progress, emit:
WorkflowTaskAddedwhen a search startsWorkflowTaskUpdatedas the status evolves (loading → complete)Why It Matters