Long-time Ultimate subscriber here. I joined when Kagi Assistant didn't exist (or was in its earliest, pre-release stages). I chose Ultimate out of a desire to support the project, not for any specific feature set. I've been actively using Kagi Assistant as it developed, so I want to contribute to this discussion with that context.
I strongly support separating Search and Assistant into distinct products. In fact, I'd go further: I would like to see a Search-only plan without Kagi Assistant entirely.
Reading this thread and observing Kagi's trajectory, I sense a strategic shift that I think warrants a closer look. Kagi was never positioned as a cheap product—it was a premium search engine. I believe subsidizing adjacent projects like Assistant from the Search subscription revenue is beginning to fracture what made Kagi exceptional in the first place.
Search quality has, in my experience, declined. The proliferation of AI-generated content ("slop") has made it harder to find signal in the noise, and while I understand this is an industry-wide problem, I'm increasingly unconvinced that my subscription dollars should flow toward an Assistant product that, candidly, feels like a raw beta.
Kagi Assistant, as it stands, lacks:
- A coherent model selection strategy (particularly for Librarian, where the model choice is opaque)
- Transparency around "magic numbers" (e.g., Librarian tool-call allowances, context compression)
- A performant UI (heavy JS, slow for long threads)
- An API or any programmatic access
- RAG-assisted workflows like Projects (a significant gap compared to competitors)
- Clear file/image parsing logic (no native DOCX/PPTX support, inconsistent image handling)
- Artifacts, Mermaid graph rendering, or Long Term Memory
- A functioning Chat Memory feature that doesn't feel bolted-on
- Reliable multi-model or agentic workload support
- MCP support for deeper integrations
- Model choice for image generation
- Video generation
- Voice/Live Mode
- A proper Coding Plan with API access
None of this reads as premium to me.
And also the search integration in Kagi Assistant is poor. If search truly costs more to serve than AI tokens (as has been noted in previous threads), then I'd rather see investment in:
A Lite search option for Assistant, with degraded result quality but a radically lower price
More sophisticated "Librarian as tool" (e.g., more filters, cache-aware querying)
We need to factor search costs into Assistant query pricing (giving users control over how their credits are allocated)
What I want, plainly, is a Kagi Search plan that includes search and—at most—Quick mode. I want a separate Kagi Assistant plan for those who need it. And I want my money to go primarily toward the product I'm paying for, not to cross-subsidise a product line that is not yet mature enough to stand on its own.
Separation would allow each product to earn its place on its own merits, which I believe is ultimately healthier for Kagi as a company and for users who value clarity in what they're paying for.
