Grpc Distributed Example with AssitantAgents/Teams

[asheeshgarg]

Do we have any examples of using the AgentChat and Teams in a Distributed Runtime via GRPC?

[ekzhu]

Not at this moment. It will be possible once #4075 is done in the next milestone. This still requires a bit of design in terms of the API, but the mechanism is that each agent in a group chat can be running from a separate worker runtime.

[asheeshgarg]

@ekzhu how about the group chat manager is it going to be running in its own worker runtime process and all communicating via runtime host?

[ekzhu]

That's the idea.

[ekzhu]

You can already do this with Core API. We don't have this in AgentChat at the moment.

[asheeshgarg]

@ekzhu how to achieve long running agents. Let say we have few agents that are developed by different teams and hosted it as a services. Now different groups want to use these agents in different workflows based on the needs or discovery.
Any idea how to achieve this? Seems like is a nested runtime?
Where top level runtime is the group want to use and the underlying is a runtime for running a long running runtime service?

[ekzhu]

I think a single distributed runtime is sufficient. We do need a separate discovery service for the runtime to make sure agents are discoverable. Track the progress here: #4241