Description
Have you considered implementing a comprehensive workflow that integrates agents and regular functions in a graph-like structure, similar to CrewAI’s approach?
This approach becomes particularly useful when LLMs encounter challenges in processing the output of external tools due to their substantial size. For instance, consider a tool that queries a remote database and generates a 4M token output. This input exceeds the capabilities of an LLM. To address this limitation, we need an LLM capable of identifying the necessary arguments for the tool and then executing it as a regular function without any token restrictions. This collaboration between regular functions and agents becomes essential for effectively handling extremely large contexts.