Recommend way of returning the last tool result as the agent's result/ preventing a last reasoning step in the Agent

[MannyDiera]

As an example, I have a personal assistant agent that uses a sub agents as tools.

In one scenario, my main / orchestrator agent calls an education agent which answers the users question. The main agent then receives this result as a tool output, calls an LLM and then responds to the user.

My goal is for the response from the education agent to be sent as-is back to the user. This way I can avoid an extra call to an LLM, reducing latency and cost.

Have looked into setting the callback handler to None, then capturing the tool result event with a Hook but in this case the hook event doesn't have enough information I need in order to respond to the user.

Does anyone have any suggestions?

[Unshure]

Hi, we have a feature request for a similar feature to allow one agent to handoff control to another agent. You can track that here: #911

[MannyDiera]

Good deal, tracking. Thanks!