Hi,
I've been using Kagi Assistant on the Ultimate plan for about 2 months. I primarily use the assistant for working with attached files. Until recently, I was very satisfied with the responses - particularly with their quality and precision.
On average, I uploaded up to 3 PDF files (max. 5 MB, 10–200 pages each).
However, over the past few days I've noticed that the assistant's ability to analyze attached files and generate responses based on them has significantly deteriorated:
1/ The assistant claims that the information I'm looking for is not present in the attached file - and this applies to genuinely simple files where the information is clearly structured and visible.
2/ The assistant, when it does find some information, gives the impression of a very shallow analysis - the retrieved data feels random, and the response is further contaminated with fabricated details, making it incorrect
The problem occurs both when I use individual AI models like Claude Sonnet, Claude Opus, Gemini 2.5, etc., and when I use custom agents I've created. This is particularly significant because instructions emphasizing thorough file searching and providing only reliable information don't help.
Interestingly, when I choose the 'Research Assistant' created by Kagi for file analysis, it analyzes the files well. I'm currently using that assistant as a temporary workaround for file analysis.
I'm not sure whether the description I've provided is sufficient, but I wanted to ask whether anyone else has noticed issues with files?
I would be very grateful for any response, as the file analysis feature is extremely important to me.
Thanks!
