I keep having issues with Claygent visiting other websites than the one provided for retrieving information/context on the site. Does anyone have any tips on how to prompt to avoid this from ever happening? Current instructions I'm using are in short "Visit the provided company page /website to retrieve X. Do NOT use any other sources". But when looking at the Steps taken, it clearly is visiting other websites based on "google search". Alternatively, does anyone experience any difference on how this instructions are met depending on which model we use? Normally use Claygent's own models or Gemini 2.0 Flash
Thank you! Will give it a go
Mathias S. Yeah that happens sometimes when claygent tries to self-correct missing context by searching externally you can easily fix it add this line at the start of your prompt: Strictly use only the provided url disable all external searches or google lookups. also include the company domain as a column reference like {Company Domain} instead of plain text that helps Claygent lock onto the exact site If it still happens switch the model to GPT-4 via Claygent it follows strict browsing rules much better than Gemini Flash.
