I know that you can emit custom messages from tool calls or nodes that are executed in your graph but I am curious if there are any examples of getting the thought process from the reasoning and thinking models and streaming that back to the user as an update OR having another LLM generate a concise update as it is monitoring the main agent working. I am looking for more dynamic updates then Tool A was called or simple messages that you put like “Agent is thinking…” I think we are moving to provide more dynamic updates. I know this is possible but looking for examples for this more complicated/sophistcated use case.
Hi there, welcome to the community! I think if you just want to stream back the thinking messages I think you could use streaming_mode='messages' (read docs here), and then you could filter out on the type of content block you are seeing (probably need to filter on message.type='thinking' or something along those lines.
1 Like