Hi Team,
We’ve recently started using generation_text_action within Moveworks compound actions, incorporating LLMs with reasoning capabilities—specifically GPT-5.1.
During this implementation, we noticed frequent time outs when handling high-reasoning requests and even occasionally with low-reasoning tasks, particularly when large contexts are provided to the LLM.
As we embrace these new demands for reasoning models and more complex analyses, we would like to discuss the possibility of increasing the time out thresholds for compound actions. This change would not only enhance the capabilities of reasoning LLMs but would also support listeners that are highly dependent on compound actions. For instance, raising the time out limit would enable us to notify more users (e.g., beyond the current cap of 50) without compromising performance.
We believe this adjustment will significantly benefit our workflows, particularly as we adopt more advanced models and use cases.
Looking forward to your thoughts!
