Some Questions and Suggestions

[juhengzhe]

I think this project has great potential, and I commend you on your work. Below are some questions I have, and while my suggestions may add complexity, they are indeed issues that an agent framework needs to address.

1、I have a question regarding how the planner is embodied and what distinguishes it from the ManagedAgent.
2、Impressive examples: It is recommended to include the capability to load and analyze data, and to implement a data analysis task where the results can be presented graphically, such as generating bar charts and other visual representations.
3、Are there plans to introduce a simple memory mechanism?
4、Are there plans to add orchestration features for multiple agents?
5、Are there plans to introduce more types of agents, such as a ReactCodeAgent?

[aymeric-roucher]

Thank you for your feedback @juhengzhe!

1. The planning step is an orthogonal direction to the ManagedAgent: the planning step is about having a specific step of an agent dedicated to planning forward towards solving its task, while ManagedAgent is about letting an agent call another agent by assigning it a task.
2. Data scientist agent example: that's a good idea, will do it!
3. We already have a simple memory! You can inspect it under function write_inner_memory_from_logs
4. Yes, we have this with ManagedAgent, take a look at the orchestration example
5. ReactCodeAgent from transformers.agents has become CodeAgent in this version: the CodeAgent has all the ReactCodeAgent had