Claude Projects integration with Kagi Assistant

kyleb

I'd love to be able to use Kagi Assistant, but I'd want the assistant to reflect and update projects in Claude.

Claude.ai doesn't search the web, so using a Claude model in Kagi Assistant would be helpful, I think. However, I'm already really invested in the Projects functionality of Claude, so I'd want some sort of syncing between Projects in Claude in and work with Kagi Assistant.

laiz

kyleb
Could you further illustrate and describe the suggestion. As someone who is not familiar with Claude Projects, how would what you're suggesting work, and what would the use cases be?

mullinggroove

laiz EDIT: wrong thread

kyleb

laiz Hi — Here’s more information on Claude Projects. I’m not sure how this would work with Kagi — I’m not familiar enough with how the search engine works or how it might integrate with Claude (I’m not a developer). The use case would be to use Kagi as a replacement for Claude.ai. The projects feature allows one to upload reference docs and images so that replies are informed by the context of the inquiry. For example, if you’re working on a blog migration project, you can upload docs about your business, goals, examples of successful writing, style guide information, and more. The assistant/project takes these pieces into consideration when evaluating and making recommendations as to what content would be migrated, what can be merged, how the information architecture might be planned, and so on.

patchmonkey

Hi! I know this is old, but I have just asked the same question to Kagi support.

I have been using Claude's Projects to create my own helpdesks and search within my own archives of material. For example, I have all my notes related to Japanese Sake (brewing, tasting, pairing, etc.) within a Claude project. Since Kagi Assistant supports private focused assistants, it would be positive to be able to upload my own documents here and use Kagi in place of subscriptions to multiple providers.

This is effectively just allowing Kagi to have a private dataset of my own material, which can then be limited when processing a response.