microsoft / autogen

A programming framework for agentic AI 🤖
https://microsoft.github.io/autogen/
Creative Commons Attribution 4.0 International
34.69k stars 5.01k forks source link

[Feature Request]: make it easy to manage agents #2678

Open taozhiyuai opened 6 months ago

taozhiyuai commented 6 months ago

Is your feature request related to a problem? Please describe.

I have created many agents in autogen studio. but it is difficult to mange them. to find it or to delete a group of agents.

Describe the solution you'd like

no

Additional context

no

wangruxun commented 6 months ago

I have a similar idea and can consider using drag-and-drop combination

jtoy commented 6 months ago

@taozhiyuai and @wangruxun can you share some ideas on how this may look like at all? Do you want to manage them from code or in a database? I think we need more insight to be able to move forward with this.

taozhiyuai commented 6 months ago

@taozhiyuai and @wangruxun can you share some ideas on how this may look like at all? Do you want to manage them from code or in a database? I think we need more insight to be able to move forward with this.

1,If I modify an agent, it will be automatically synchronized wherever the agent is referenced in the group chat.

2.At present, the agent does not follow the requirement of the prompt word very much. If you don't let him do it, he will still do it.

3There would be many agents for multiple workflows. All of them are placed in an agent column. It looks very messy. If you want to find an agent, it is very difficult.Can I mark it with a color? For example, for a certain workflow, I mark it in red. For all agents, the buttons are framed in red. . Or by the agent's name prefix. Sort the agents.

4.It is in the chat interface of playground. The layout of the box showing the chat history is very unreasonable. A large part of it doesn't display anything, but it takes up a lot of screen area.

taozhiyuai commented 6 months ago

Another problem is that as soon as the workflow starts working, there is no way to interrupt it, but to wait for it to finish its work, which is sometimes a waste of time.

taozhiyuai commented 6 months ago

At present, there will be a summary after a group chat. Multiple rounds of dialogue will be folded below the summary. I don't think that's a good idea. The information can not be displayed completely, which is very inconvenient.

The conversation of each agent is displayed in the normal group chat mode. The summary of the dialogue by llms is displayed in the form of a group chat after the agent dialogue. That is to say, the summary of llm is one of the speakers in the group chat.