Background

Currently, when an agent calls a tool, the tool must fully complete before any output is returned.
However, some tools internally call LLMs or perform long-running tasks, and users have to wait without seeing any intermediate results.

Request

Please support streaming tool outputs, similar to model response streaming.
Allow tools to send incremental outputs while they are still running, so users can see progress in real time.

Benefits

• Improve user experience by reducing perceived waiting time.
• Enable better interaction for tools involving LLM calls or large content generation.
• Make tool behavior consistent with LLM output streaming.