They have a dedicated chat interface at https://kagi.com/assistant https://kagi.com/assistant You can access this from any search bar by adding !expert to your

They have a dedicated chat interface at https://kagi.com/assistant https://kagi.com/assistant You can access this from any search bar by adding !expert to your query. That’s the general case… there’s also !code for development questions and you can create your own more focused flavors. You can provide custom instructions and set up multiple assistant flavors. For example, you can have one that highlights controversial information about a topic, another one that jokes about everything, etc. All the assistants have optional web access. They use Kagi search to look up relevant information for you. You can apply a search lens to restrict web results to particular domains. You can set the assistant to use different LLM APIs. They currently support Anthropic (Claude 3 Haiku, Claude 3.5 Sonnet, Claude 3 Opus), OpenAI (GPT 4o, GPT 4 mini), Google (Gemini Pro), Mistral (Pixtral, Large) and Meta (Llama 3.1 405B). Kagi maintains their own LLM benchmark. Their goal is to (1) ensure that models haven’t ever seen the questions before and (2) emphasize reasoning and instruction following, which they’ve found to be most important for the assistant. https://help.kagi.com/kagi/ai/llm-benchmark.html https://help.kagi.com/kagi/ai/llm-benchmark.html