Open jcoombes opened 1 month ago
Have you tried this?
I saw this and got the impression that it would only ask for input directly before giving the final answer, but this might just be myself reading the docs incorrectly.
I would want it to ask for help the moment it gets stuck and uses the tool, and then potentially do a bunch more task calls or crew member calls afterwards?
It currently works after each task is completed but only once per task. It may be something that we can improve on in a future release.
You can also use allow delegation https://docs.crewai.com/core-concepts/Agents/#agent-attributes
https://python.langchain.com/v0.1/docs/integrations/tools/human_tools/
Human are AGI so they can certainly be used as a tool to help out AI agent when it is confused.
I think my current pipeline requires too much thought to complete this task successfully, so whilst debugging I would like the model to query a human for answers when it gets stuck.
This might be as simple as asking for an input() from stdin, and then composing this into the prompt?