Ah okay. You actually have a few options. You could just fetch the information via an execute code card, save it to a variable in Botpress, and then include that variable in the AI Task Input. You could also fetch the information, and do an API call to a LLM like OpenAI or Claude etc. directly in the same execute code card, and then save the response from that LLM to a Botpress variable, and then use it. The reason why you might do your own API call instead of the AI Task is that it grants you a bit more control. You can pick which LLM you want to use, not limited to OpenAI, which allows you to test multiple models, to pick the one that works the best compared to the costs. You set all of the information as you want it. You can prompt it 100% to your liking, which might be helpful here🤷‍♀️ I would recommend you to do some testing, maybe try and consult with someone or something, like ChatGPT. I haven't used JSON a lot in my prompts, so I can't really tell you what will work the best.