Make Custom Assistants and Presets Bang-Addressable

CrunchyFritos

Make custom assistants in the new Kagi Assistant v2 bang-addressable similar to in Kagi Assistant v1

artemvmin

+1, especially with the latest update making Assistant v1 inaccessible, I feel like I'm taking a considerable productivity hit not being to select any of the presets or models directly in my search query. For example, I find GPT 4o is still best at writing code, and I am no longer able to ask GPT directly without using my mouse and dropdown for every single search.

CrunchyFritos

artemvmin I agree. Bang-addressability with presets in v2 would be great. Specifically, I'd like to be able to use bangs to determine any combination of model + internet access along with custom assistants. This would put it on more than parity with v1 since that would allow users to create virtual "Research," "Code," and various "Chat" modes.

sudoer-777

One of the main things I am interested in when prompting the assistant is I want to see what its massive amount of training data leads it to output, and the search feature that it defaults to completely ruins that and summarizes like 2 sources that I could just read myself. The output with just training data may not be very reliable or up-to-date, but it is far more consistent than the web searching feature. If I need to be sure that something is correct, then I'll do my own research and probably explore at least 10 different sources instead of the 2-3 top results that the assistant uses. I wish there was a way I could make it default to the non-search mode like the !chat bang on the older assistant.

Luis

Hey! We're currently working in building a better UX for Custom Assistant bangs (relevant thread: #2183)
In the meantime, you can use the profile and internet query parameters in the Assistant URL to achieve your goal -- for example: https://kagi.com/assistant?q=%s&profile=gpt-4o&internet=on

&profile with value,

for custom assistants: the id in the URL when you open your custom assistant in the settings page
for standard models: uses the model's name, for e.g. gpt-4o
- full list: code, claude-3-haiku, claude-3-opus, claude-3-sonnet, gemini-pro, gpt-4, gpt-4-turbo, gpt-4o, gpt-4o-mini, llama-3-405b, mistral-large, mistral-nemo

&internet + &personalizations use values on and off