The Research Assistant is great, but too expensive. I would like to select a system which is very similar to the Research Assistant in terms of agentic capabilities and tool use, but with a much cheaper model powering it. Meaning access to python, PDF reading, image analysis, repeated searching and web page access and analysis cycles, revision, etc. If it can be done cheaply, image generation too (Flux?). As of this writing, that would mean instead of Claude Sonnet, using something like Deepseek, or perhaps Qwen, z.AI GLM, or maybe even Gemini Flash, depending on the cost.
This would be perfect for everyday use when you don't need an answer quickly (which the Kagi Quick Assistant is great for) but the research project isn't worth spending the token cost for Kagi Research Assistant. It would be especially valuable for it to be thorough in performing extensive search, repeatedly, to reduce the chance of missing something. Basically, a cheap, slow, extensive model with capability to access all Kagi agents.
