Parallel Agent Execution

[martimfasantos]

Hello,

Is it possible to execute two SocietyOfMindAgents in parallel, so each one executes their inner-team members and each one does different tool calls in parallel? I am struggling to find documentation about that. Thanks :)

[ekzhu]

How about this?

task1 = asyncio.create_task(agent1.run(task="..."))
task2 = asyncio.create_task(agent2.run(task="..."))

results = await asyncio.gather(task1, task2)

Curious what's your context? If you want to run two parallel teams in AgentChat, you need to create a custom agent and have the above code executed within that agent. If you are using Core API, this is natively supported via agent runtime.

https://microsoft.github.io/autogen/stable/user-guide/core-user-guide/design-patterns/concurrent-agents.html

[martimfasantos]

Hi @ekzhu

Thanks for your reply! I’m trying to run two SocietyOfMindAgent in parallel so each can run their own team at the same time and return their individual results to a GroupChat, but I’m stuck. Every group chat option (Selector, Swarm, RoundRobin) seems to only pick one speaker at a time. I’ve tried a lot of prompt engineering, but no luck.

My approach involves having a planner agent recognize when a task should be subdivided across multiple agents (based on their expertise) and then execute them in parallel, retrieving their results afterward.

Should I dive deeper into the creation of a custom agent to allow for this?

Thank you!