A2A Protocol - langgraph and deployment ideas

[multib0t]

Hi has anyone played around with a2a protocol and langgraph. I was wondering if there are any commmon approaches people would recommend. There is a basic example on google's github but I was thinking about more advanced usecase where there would be a swarm/fleetof agents working together to complete a more complex task.
I was also wondering about the deployment of such swarm. What would be in people's opinion the best environment for it that is reliable and cost effective. I was thinking of aws lambda but it would be limited to quick question - answer and not good for long running tasks.

Any ideas would be appreciated.

[marco-sciatta]

Hi, there is an example, in the official a2a repository,
here: langgraph example

However, i'm struggleling to understand how to replicate the same streaming behavior without using structured response.

Check it out the examples, are very huge help

[nithinjp1997]

You can pass the task updater inside the runnable config of the graph.invoke() or .stream() and do the updater.update_status() or updater.complete() in the tools or node functions inside langgraph itself.

[sawradip]

In my case, I also thought lambda is a solution. But if I go beyond very simple agents, Lambda is not enough. One is your case, Long running tasks. Second is, if I want to stream the response. And taking a VPS for deploying it might be overkill, considering the costs and scale.

Finally the solution I came up with, is building a api abstraction[along with streaming], which can be deployed anywhere, even serverless. I am organizing the code @ RunAgent. Let me know if it can be of use.