Description
Currently, neither the consolechat nor webchat interface in any reference implementation surfaces tool call activity to the user. During agent execution, users see only a generic "thinking" or loading indicator, with no visibility into which tools are being invoked, what inputs were provided, or what results were returned. The tool calls execute silently and only the final response is shown.
This is a significant UX gap, especially as AFM agents can be powered by multiple MCP tools with non-trivial side effects. Users have no way to understand what the agent is doing on their behalf, or to build trust in its actions.
Current Behavior
- consolechat (Textual TUI): Shows a LoadingIndicator widget while the agent runs, then renders the final reply.
- webchat (HTML/JS): Shows a typing indicator during the blocking HTTP request, then renders the final agent response.
- Both implementations execute tool calls entirely within the backend with no streaming or intermediate feedback surfaced to the UI.
Description
Currently, neither the consolechat nor webchat interface in any reference implementation surfaces tool call activity to the user. During agent execution, users see only a generic "thinking" or loading indicator, with no visibility into which tools are being invoked, what inputs were provided, or what results were returned. The tool calls execute silently and only the final response is shown.
This is a significant UX gap, especially as AFM agents can be powered by multiple MCP tools with non-trivial side effects. Users have no way to understand what the agent is doing on their behalf, or to build trust in its actions.
Current Behavior