I am working on completing an Agent that assesses document contents and then reports its findings via email. The email body (HTML) varies greatly it seems, even with some fairly explicit instructions. Every now and then, it hits a GRAND SLAM as far as formatting and contexts it generates (sometimes i get colored edits, etc. and though rare, sometimes no edits at all - no joy). Could you make some suggestions for making the results somewhat more consistent? Not looking for templated results as the document’s content vary greatly in scope, making that somewhat unfeasible. I think my model settings may help … i am currently liking the results from Gemini 2.5 Flash having moved there from GPT-5 whiich was slow, expensive and often overzealous. I’ve atttached the Gemini 2.5 Flash setttings iin the event you see something that might help? Thank you good sirs.
Hi @Avocet,
Unless you’re using an HTML template, the result will likely vary because the AI model generates new HTML each time the Agent runs.
That said, the more detailed your instructions on style and formatting, the more consistent the results will be. Lowering the temperature setting can also help improve consistency.
Hi Alex … thanks for the prompt response! I will look into getting a tool to build me an HTML template. Talk soon!
Best,