I've always felt the Privacy section to be a bit confusing, but since the switch from a five-star style system to a "Learn more" button we've lost the distinction between Quick and Research. Clicking through the button takes me to LLMs & Privacy where neither Kagi Quick nor Kagi Research are listed. While they have some documentation in Kagi Research Assistants there is still no information about the providers involved.
My memory is that Quick showed 5/5 and Research showed 4/5 or less, though I cannot recall for sure.
It's also never been clear to me how or why a specific model was rated a specific way -- Claude and GPT are both listed as "30 days" but GPT was usually ranked much lower in Privacy, but that's less important. This distinction has now been lost
Expected behavior / proposed fix: add a section in the LLMs & Privacy article documenting which providers may be used to power Kagi Research Assistants
I understand the specific models and prompts may be considered a 'trade secret' and change over time, but it would be good to know the possible providers (Anthropic, Azure, Bedrock, Cerebras, etc.) or at least their retention policy (not stored, 24 hours, 30 days).
To reiterate, I do NOT expect or need the specific model names (Sonnet vs Fable, etc.) especially since that may be subject to change as "upgrades" become available from the same providers. Would be interesting, but I don't think it's ever been publicly documented to that level of detail anyways.
