[Feature Request] Multi-Agent Patterns Enhancement

[wuqunfei]

Problem Description

I found the Agent Team API can not handle complicated workflow and Hierarchical, currently agent team is only instructions.
Key Patterns like

• Orchestrator-worker pattern
• Hierarchical agent pattern
• Blackboard pattern
• Market-based pattern

Proposed Solution

Enhance the Multi agents design pattern, rethinking from existing market implementation.

• AutoGen based queue system https://microsoft.github.io/autogen/stable/user-guide/core-user-guide/design-patterns/concurrent-agents.html

• Langraph based network graph https://networkx.org/documentation/stable/tutorial.html

• CrewAI does not support advanced pattern only Sequential, Hierarchical, Consensual Process (Planned) https://docs.crewai.com/concepts/processes

I recommend system implement base on event driven like autogen, it is much easy to adapt into distribution system with existing tools like message system https://www.confluent.io/blog/event-driven-multi-agent-systems/

Alternatives Considered

using external message bus like kafka, agent listen on these topics. agno system can cross clouds providers, legacy system to integrate together. https://medium.com/data-science-collective/agentic-mesh-patterns-for-an-agent-ecosystem-ef13469b7cf7

Additional context

Include any extra information that might be helpful, such as:

Would you like to work on this?

We welcome contributions! Let us know if you’d like to help implement this feature.
[x] Yes, I’d love to work on it!
[x ] I’m open to collaborating but need guidance.
[ ] No, I’m just sharing the idea.

[Harsh-2909]

Thats a very good suggestion. I would love to work on this with you. Do tell if you need any help as I will need some guidance for this as well if i want to contribute.

[dirkbrnd]

Hi @wuqunfei
We are actively working on an improved experience of teams in Agno. This is really valuable insight! I will make sure to incorporate it in our design

[wuqunfei]

thanks @Harsh-2909 and @dirkbrnd for your feedback, I am working in one biggest insurance company in Europe , these are enough open source/commeical agent frameworks, but no one is solving integration/patterns between agent very well and very easily manner.

I do not think these agent framework will be industrialized easily. Because they do not care what have already achieved so far like micro-service, service mesh, service discovery, even-driven architecture in serious operational level. In fact, check the Autogen message system, it is quite easy to implement by event mindset.

agent_master = Agent(...)
agent_worker_1 = Agent(...)
agent_worker_2 = Agent(...)

@subscrib(topic="kafa.topic_name")
def process_master(message = ...)
       agent_master.process(message)

@subscrib(topic="kafa.topic_name")
def process_worker_1(message = ...)
       agent_worker_1(message)

@subscrib(topic="kafa.topic_name")
def process_worker_2(message = ...)
       agent_worker_2(message)

if you like, we can have remote meeting to exchange the ideas.

[ashpreetbedi]

this is an amazing recommendation @wuqunfei! reading thought and would love to meet with your team to get feedback as we're implementing!

[wuqunfei]

I am living in GMT +1, German. Free to book a session next week in working day.

And if your team is busy, there are many powerful python stream libary you can integrate into your framework with less effort.

e.g.

• https://faststream.airt.ai/latest/getting-started/subscription/
• https://faust.readthedocs.io/en/latest/