Back to Kagi

Original Article ↗ Hacker News Discussion ↗

Paid search engine Kagi is drawing praise for its customizable results, lack of ads, and integrated AI tools, with many users saying it feels closer to “old Google” and gives them more control over what they see. Others remain unconvinced, arguing that its results are often no better than Google or DuckDuckGo, its pricing tiers are too restrictive for heavy searchers, and that large language models now cover many of their needs. Underlying the debate is a broader concern that the quality of the web itself has declined due to SEO spam and AI-generated content, limiting how much any search engine can improve the experience.

Overall sentiment on Kagi

  • Many self-identified long‑term subscribers say Kagi delivers consistently high‑quality, “Google‑before-ads” style results.
  • Others find results similar to Google/DDG/Brave and don’t see enough improvement to justify paying.
  • Some trial users report Kagi lacked certain results versus Google, especially in earlier years.

Search engines vs LLMs

  • Several commenters now use LLMs (including ChatGPT, Claude, Perplexity, Kagi Assistant) for most “answer” queries and traditional search only for specific documents, URLs, or mission‑critical accuracy.
  • One common view:
    • “Seeking documents → search engine”
    • “Seeking answers → LLM”
  • Critics say LLMs hallucinate, are untrustworthy for serious search, and often just sit on top of traditional engines anyway.
  • Some users describe massive productivity gains from LLMs in debugging and API comprehension, others think this erodes “real” problem‑solving or just layers band‑aids on bad complexity.

Features people like (or want) in Kagi

  • Strong praise for:
    • Per‑user curation: pin/raise/lower/block domains; account‑wide blocklists.
    • SlopStop and community/crowd blocking of LLM‑generated “slop” sites (though some think review is slow and too many reports are rejected).
    • Custom lenses (scoped search), URL rewriting, vim keybindings, optional AI answers via ?, Kagi Assistant, and Kagi Translate.
  • Some want more/better widgets (sports, finance, weather), better follow‑through on promises, and deeper SlopStop automation/ML.

Pricing, limits, and usage patterns

  • $10/month is seen by some as trivial (compared to coffee, entertainment, or broadband), and by others as too high, especially given subscription fatigue and lower search volume due to LLMs.
  • The 300‑search/month plan works for light users; heavy users report 800–2,500+ searches/month and hit limits quickly.
  • Requests appear for: cheaper bare‑bones plans, bundled API/MCP access, and more generous quotas.

Regional performance and language issues

  • Mixed reports: good experiences in France, Germany, Netherlands, Canada; poor for Finland and Arabic‑language news.
  • Some still rely on Google for local products, maps, and reviews, or on specific Arabic news sites directly.

Privacy, indexing, and ecosystem concerns

  • Some value Kagi’s paid, non‑ad, more private model and customization as a way to “vote” for a better ecosystem.
  • Skepticism exists about its reliance on other engines’ indexes, though others argue re‑indexing the web is unnecessary.
  • Alternative mentions: Brave Search (own index, good for some), DuckDuckGo, Startpage, Qwant/Ecosia/Staan.ai, Marginalia Search.