[Assistant API] Where is the right place to tell the API how to process the output from tools?

I have an assistant API that can use a tool that returns a big json. When the assistant gets this big json I want it to process it in a certain way. I am currently telling the assistant this on his main instructions, however, this feels wrong for a few reasons:

  • When the tool is not being used, this information is just crowding the context window;
  • Sometimes the Assistant forgets to follow these instructions, I think if the instructions came right before or after the tool output it would work better.
  • This solution does not scale with more tools.
1 Like

Create a step by step process flow (numbered bullets) and ask the Assistant to go step by step.