Hey @slilichenko , need a confirmation and some clarification as i have been digging into how to exclude thoughts.
So far I’ve traced this through the planning request/response flow - specifically in _NlPlanningRequestProcessor and _NlPlanningResponse. I noticed _remove_thought_from_request() already exists and clears the thought field from LlmRequest parts,

def _remove_thought_from_request(llm_request: LlmRequest):
  if not llm_request.contents:
    return

  for content in llm_request.contents:
    if not content.parts:
      continue
    for part in content.parts:
      part.thought = None

Now, to make this configurable:

Where should the exclude_thoughts: bool = False option be made ?. Adding in the BaseAgent config comes to mind, since this will apply across planners and all agents. Just wanted to know which would be the right place, or do you see other options(maybe somewhere closer to planner or processor level)?
On a related note: I see that _remove_thought_from_request() clears part.thought, in general, is thought stored strictly as a part type, or could there be other part types carrying thought-like content (e.g., a content role or metadata field)? Just want to make sure I’m not only removing the field but also filtering out any actual thought content that might come back from the LLM.

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions