Summary/Report Agent

[moritz-baumgart]

Currently, the final response implementation is very basic:
The LLM is provided only with the user request, result node code, and (potentially truncated) output.
See:

AutoRecLab/treesearch/minimal_agent.py

Line 579 in 5bc53b3

async def _summarize(self, user_request: str, node: Node) -> str:

We should consider adding a small agent at the end of the pipeline to improve the quality of the final response. This agent might have capabilities like:

• File I/O to read/peek into result files.
• Access to a Python interpreter to potentially load results and calculate e.g. further statistics.
• A more structured multi-step/iterative approach to generating the response.