Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Prefect Community

<@U02GMEZU18B> now that Teams and Swarm is deprecated is the best approach for implementing the original intended behavior to use tools ?

using agents or any of the top level utils is broadly recommended usage

For now just getting agents to talk to one another. For example this delegates functionality doesn’t work because it depends on the marvin.Team class which has been deprecated. Currently using `3.2.2`
<https://askmarvin.ai/concepts/agents#delegates>

I have many use cases where I want a “Team” of agents to work together to solve a problem. Each agent being specialized in a specific topic. Right now I’m just sandboxing the `researcher` `writer` scenario.

if you can create an issue on the marvin repo, that'd be most helpful for tracking things!

I understand that, I’m just curious if there was discussion around the recommended way to implement Team-like runs when the decision was made to deprecate teams and swarms, or if they were just ripped out because they didn’t work.

Seems like an aggressive move to rip them out just because they didn’t work. I figured there would have been a better way to go about teams. The concept was nice to begin with.

&gt;  if they were just ripped out because they didn’t work.
pretty much. it totally should not have been released as a part of the public API of the library, since as you discovered, it was full of edge cases / bugs. to be candid, `prefect` and `fastmcp` stuff demand more of my attention these days so futzing with half-baked features in `marvin` is not in my best interests

if you want to reinstate the feature, i would be happy to review a PR, but that's not going to be priority right now. i'm a fan of pydantic-ai if you want a very thoughtfully maintained agent framework

the core utils in marvin and agent work great IME, its just those extra team/swarm features should have been experimental at the time of 3.x release

Ok I understand what you mean, so does that mean that Marvin 3 is essentially going to be deprecated ? Or trail behind pydantic-ai ?

i am passively maintaining `marvin`. that is to say, if something well-supported/familiar is broken, ill try to fix it. not going to add new surface area or spend a lot of time re-designing the fringe capabilities

all the core functionality: extract, cast, generate, classify, Agent etc are all there and not going anywhere for the time being

Thank you for sharing this insight, greatly appreciate it.