Hello, I would like the ability to loop through a custom tool / sub-workflow with an AI agent. In the workflow below, the sub-workflow iterates through the loop but the AI agent only receives the first iteration. Ideally, the sub-workflow would feed the agent each iteration, allowing for the agent to provide an item-by-item response with less error and more specific focus than cramming all the items in at once or using an arbitrary (character/token) splitter. How can I go about this? Thank you
hey @Jonathan_Sherman
Not exactly what you’re asking for but looping tools is probably not advisable. Agents have something called a max iteration rate which the agent will force-stop when this is reached - eg. 32 is the current limit would be inadequate for say 100 student assignments. The max iteration rate is adjustable however but consider the following example.
Agent scores on all criteria at the same time.
Much simpler than using a scoring custom workflow tool.
Hey @Jim_Le, thanks for the examples. I agree in the case I presented that this makes sense, but I am still curious about how to possibly loop a tool, even if it’s not a good idea, ha. I can think of some interesting scenarios I’d just like to experiment with. Any ideas on how to implement/force it?