[BUG] Agentflow Not Functioning Properly in Flowise 2.0.0

[seonghobae]

Description: After upgrading to Flowise 2.0.0, we have encountered an issue where the existing Agentflow is not functioning correctly. Specifically, messages generated by the Worker are not appearing in the final output.

Steps to Reproduce:

1. Upgrade to Flowise 2.0.0.
2. Execute an existing Agentflow.
3. Observe that messages from the Worker do not appear in the final result.

Expected Behavior: Messages generated by the Worker should be included in the final output.

Actual Behavior: Messages from the Worker are missing in the final output.

Environment:

• Flowise Version: 2.0.0
• Operating System: Ubuntu
• Browser: Edge

Additional Context: This issue was not present in the previous version of Flowise. It appears to be a regression introduced in the 2.0.0 update.

Screenshots: If applicable, add screenshots to help explain your problem.

[HenryHengZJ]

This is the new behaviour. Since multi agents are working back and forth between workers, we think the final output should be a summarisation of the whole conversation.

For example. If you have a coder agent and a reviewer agent, and you ask the team to generate piece of code. The expected final output should be the code generated by coder agent. But it will be the reviewer's output.

This new behaviour is somewhat aligns with AutoGen studio

[seonghobae]

The problem with this approach is that when you send a POST request, it doesn't get the right text. I think this behaviour is quite misleading because the current Flowise native chat tool doesn't show all the Agent results either, and the Agent results are collapsed.

This is not proper result what I intended as final result! I've spent countless hours trying to create complete Markdown documents using Agent, and my current belief that the end result should be a summary has undermined my efforts. It's not up to the service framework to decide how the service should be delivered. By that logic, even sequential agents should have summarised results, but senquential agents are getting the ‘normal result’, aren't they? The instruction Remember to give reasonings, instructions, and summarisation in multiple agents is not necessary, it's just causing misbehaviour. Summarising only makes sense when you specify some sort of specification, and summarising without any specification is meaningless - what's the point when even the example agent you created is misbehaving?

The reason why you shouldn't summarise is more obvious than you might think. In a model that values economics, summarisation leads to a loss of context.

[HenryHengZJ]

Understood, will see if we can create an optional field for supervisor, and provide summary only when user select so

[HenryHengZJ]

@seonghobae changes has been pushed to 2.0.1, can you try and see?

[seonghobae]

@seonghobae changes has been pushed to 2.0.1, can you try and see?

I haven't reviewed the regression test in detail so I can't say for sure, but there doesn't seem to be anything out of the ordinary in our normal testing routine. If there is, I'll reopen the issue.

---

Query

Find the weather in Seoul today specifically and tell me in English.

---

Response:

Current Weather in Seoul

• Current Temperature: 26°C
• Feels Like: 33°C
• Today's High: 31°C
• Today's Low: 26°C
• Weather Condition: Mostly clear and humid

There is a forecast for thunderstorms from Thursday afternoon to Friday night.

For more detailed weather information, you can check the following links:

• [Seoul Weather Forecast](https://www.accuweather.com/en/kr/seoul/226081/weather-forecast/226081)
• [Hourly Weather Forecast](https://www.accuweather.com/en/kr/seoul/226081/hourly-weather-forecast/226081)
• [Daily Weather Forecast](https://www.accuweather.com/en/kr/seoul/226081/daily-weather-forecast/226081)
• [Seoul Weather Radar](https://www.accuweather.com/en/kr/seoul/226081/weather-radar/226081)
• [Air Quality Index](https://www.accuweather.com/en/kr/seoul/226081/air-quality-index/226081)