langgraph/concepts/multi_agent/

[giscus[bot]]

langgraph/concepts/multi_agent/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/concepts/multi_agent/

[wvanhuffel]

I have to say, this is one of the best if not the best multi-agent articles I have found. Thank you.

[asifmd1806]

100%

[santosomar]

I completely agree! Outstanding work by the LangChain/LangGraph team!

[oelmady]

This is exciting

[sabbadino]

Nice article , thank you.

[DK-KD-DK]

Langraph’s modularity and specialization for multi-agent systems are a strong step toward managing the complexity of LLM-driven workflows. The ability to design custom handoff mechanisms and hierarchical architectures is particularly useful.

However, two questions come to mind:

1. How does Langraph handle the computational overhead when scaling agents with different state schemas?
2. What strategies exist for optimizing handoffs between agents with shared vs. private scratchpads?
   On a related note, Klover.ai takes a different approach by integrating ensemble-based Artificial General Decision Making™ (AGD™) with a multi-agent core. Could concepts from AGD™, like leveraging inter-agent ensembles for decision optimization, enhance the flexibility of Langraph's supervisor models? Curious to hear thoughts!

[waynehuu]

You forgot to import Command :)

[magallardo]

In the context of the multi-agent architectures discussed above, how does the Lqangchain Agent Protocol integrate or contribute? Any examples of its use in this context?