eastern-cyan•3mo ago
Multiple Agents in Agent Network with Supervisor Agent
In this video explains nicely how to use a central supervisor agent to decide when to use a knowledge base or route to other expert agents for other tasks:
https://www.youtube.com/watch?v=-rDAGu4nVGc
I would love to get the supervisor agent instructions and take a look at how the workflow is set up to better understand how it works and what are the best practices.
The Documentation only briefly mentions using Multiple Agents but does not explain it. The Template Library does not seem to contain anything with a multi-agent network like this.
Where to find the example prompt for the supervisor agent, or the template workflow for this? Or a page or documentation that describes how to best do this? Anyone succesful at implementing this?
7 Replies
genetic-orange•3mo ago
Experienced the same issue, especially with tool calling. I am unable to force the agent to take a specific path.
its all in the prompt
genetic-orange•3mo ago
I dont think so, because the prompt was very detailed on my side plus i wrote to always call the tool when the user type something
Hey @Steve B. | IntelleSync.ai what are your thoughts making tool calling more precise on AI agents?
As @blackadder5343 mentioned, I think it really is down to the prompting. I don't use tooling calling I prefer to use paths at this point as I've had better consistent results.
eastern-cyanOP•3mo ago
From what I’ve got to work so far you need to:
- add paths with a specific name, and add a description on when to use that path
Then, inside your prompt, also mention the scenario on when to use that path with a clear description.
Just adding the paths did not work. When mentioning the path in the description it did start to work.
I also add a variable to the path to store the information I find out that does help
Bit still some time it refuse to the the paths...
exotic-emerald•3mo ago
I have been going around and around with this same concept. Is anyone else noticing that when I run the full network start button —> supervisor agent —> other sub agents things seem to be working well but when I test individual sub agents I get wild and inconsistent results when looking at paths and loops (like for setting variables)
If anyone has any feedback it would be appreciate