Concurrent Workflow Sample Code Not Working

[liquidgreeting]

I copied the Concurrent Workflow sample code from

https://github.com/microsoft/agent-framework/blob/main/dotnet/samples/GettingStarted/Workflows/Concurrent/Concurrent/Program.cs

The only modification I made was to comment out the section that sets up the Azure OpenAI client. I replaced it with code that creates the client with OllamaApiClient so that I can test locally (IChatClient chatClient = new OllamaApiClient(new Uri("http://localhost:11434"), "mistral:7b-instruct");). I do not yet have an Azure account. The code executes but does not produce any output. When the code finishes running, the only thing displayed on the console is:

C:\Users\DevUser\Documents\Visual Studio 2026\Projects\NewStore.Agent\NewStore.Concurrent\bin\Debug\net10.0\NewStore.Concurrent.exe (process 2276) exited with code 0 (0x0).
Press any key to close this window . . .

As you can see, there is no output from the aggregationExecutor. I've tried using different LLM Models. No change. Has anyone else experienced this while testing the Agent Framework Concurrent Workflow code locally or with an Azure account?

Any assistance will be greatly appreciated. Thank you in advance.

[moonbox3]

Adding @lokitoth

[jmurkothintel]

@liquidgreeting Ran into the same problem and I have raised an issue in the repo

[liquidgreeting]

Thank you. I would really like to move ahead with testing that functionality for a use-case I have.

[lokitoth]

Got my wires crossed a bit here earlier. There are a couple of related issues here:

The underlying issue is that the fan-in does not properly understanding the ChatProtocol way of doing things. The way this works for the orchestrations is we aggregate the messages upfront, then feed the pre-aggregated List<ChatMessage> into the output. With the way the workflow in the sample is currently set up, especially post-changes to how AgentRunResponseUpdate is aggregated, we it should work around the FanInEdge issue.

I'm going to make a change which will allow multiple "Reduction Strategies" to be used with FanIn (including null, which would be the equivalent of having added the edges manually), but it is a bit of a redesign, and will take a bit more time.

In the meanwhile, I've put out a PR to fix the sample: #1871 fix pulled into another PR to reduce conflicts #1612

[lokitoth]

Immediate fix is merged in. "fan-in" API updates to follow