Kagi

By Bill Hartzer · Updated 19 August 2026

What Kagi is, and where its results actually come from

Kagi is a paid web search engine. There is no advertising anywhere in the product, no sponsored placement at the top of a results page, and no advertiser whose interests the ranking has to accommodate. You pay a subscription; in return you get a results page with controls no conventional engine offers, most notably the ability to raise, lower, pin or permanently block any domain from your own results.

The question a reader usually arrives with is whether that subscription buys an index independent of Google and Bing. It does not, and this is the single most misunderstood thing about Kagi.

Kagi genuinely crawls the web and genuinely ranks its own index, which disqualifies the description "resold results". But it cannot serve broad commercial web coverage from its own crawl alone, and does not claim to. Kagi's own comparison documentation describes the product as including "anonymized requests to traditional search indexes, including Brave, as well as our own non-commercial index (Teclis), news index (TinyGem), and an AI for instant answers." That is a hybrid, by Kagi's own account.

The useful way to hold this: Kagi's independence is at the ranking layer, not the index layer. The re-ranking, the filtering, the personal blocklists are Kagi's, and nobody else ships them as well. The pool of pages being ranked is substantially somebody else's.

Kagibot, Teclis and TinyGem — the part Kagi crawls itself

Kagi operates a crawler called Kagibot, which identifies itself with the user-agent string Mozilla/5.0 (compatible; Kagibot/1.0; +https://kagi.com/bot). Kagi documents the crawler publicly, lists the IP ranges it crawls from, and states that Kagibot falls back to following Googlebot directives in a site's robots.txt file when no Kagibot-specific rules are present. A crawler is the program that fetches pages so they can be stored and ranked; an index is that stored copy of the web. Kagi has both.

Kagibot feeds two indexes:

Teclis — a non-commercial web index. Kagi describes it as the product of a crawl through millions of domains focused primarily on non-commercial, high-quality content. It is deliberately weighted toward what is often called the small web: personal sites, blogs, forums, the material that ad-driven ranking systematically under-surfaces. It is explicitly not a general-purpose index of the whole web, and Kagi has never said otherwise.
TinyGem — Kagi's news index, built the same way.

This is why the verdict is hybrid rather than own index. Teclis is a real index built by a real crawler, and on an obscure or non-commercial query it is often the thing producing the result that makes Kagi feel different. On a mainstream commercial query, the head of the results list is overwhelmingly partner-index material. Kagi has never published what percentage of a typical results page comes from Teclis versus a licensed index, and anyone quoting such a figure is guessing.

The blend, and the disclosure that got quieter

For general web coverage Kagi issues anonymized API calls into commercial search indexes, and states that a single query fans out to "a dozen or so different sources simultaneously" before being merged into one ranked page. Alongside the general indexes it calls specialised sources — Marginalia among them — and vertical APIs including Wolfram Alpha, Wikipedia, Apple, Open Meteo, Yelp and TripAdvisor.

Which general indexes? Here the record changes, and it is worth stating precisely, because it is the crux of the whole "is Kagi independent" question.

Until May 2024, Kagi's own documentation named its traditional-index partners outright: Google, Yandex, Mojeek and Brave. On 2 May 2024 those names were removed from the public docs and replaced with the phrase "all major search result providers worldwide." The change is visible in the commit history of Kagi's public documentation repository. Kagi did not announce a change to the underlying supplier mix — what changed, verifiably, was the disclosure, not necessarily the product.

As of 2026, Brave is still named specifically on Kagi's competition-comparison page, so Brave is confirmed current. The rest of the mix is undisclosed, and whether Google remains a source is not publicly knowable. For a company whose entire pitch is alignment with the person paying, making a central fact about the product less knowable is a poor transparency outcome, whatever the commercial logic behind it.

How Kagi makes money, and how big it is

Subscription, and only subscription. Kagi takes no advertising, sells no user data, and operates no affiliate-driven ranking. Structurally, the offering has consistently looked like this:

a free tier capped at a small fixed number of searches, functioning as a trial rather than a usable service;
consumer tiers billed monthly, differentiated by search volume — a capped tier versus unlimited searching — with higher tiers adding earlier access to AI and assistant features;
family and team plans billed per seat;
metered developer products sold separately: a Search API, and AI endpoints including FastGPT and the Universal Summarizer, billed per call and backed by Teclis and TinyGem.

Specific prices are deliberately not stated here. Kagi has restructured its tiers more than once, including moving away from an early unlimited-search plan, and each restructuring drew sustained argument from a vocal user base. What matters structurally is the direction the money flows: with no advertiser in the system, the only party with a claim on the results page is the person paying for it. On the company's own numbers, Kagi crossed $1M in annual revenue in 2023 and raised roughly $670,000 from its own users in a community fundraise that year, with total invested capital near $5M at that point.

Kagi also publishes live usage figures continuously, which almost no competitor does. Read on 19 August 2026, its public stats page reported 74,331 members and 1,098,200 queries in the preceding 24 hours. The same source showed roughly 50,000 members and 845,200 searches on 9 June 2025. Growth is real; the figures are self-reported and unaudited. The absolute scale is nonetheless tiny — about 1.1 million queries a day is the order of volume Google handles in a couple of seconds — and Kagi sits far below StatCounter's reporting threshold, so no credible market-share percentage for it exists. Around 74,000 paying members is also a small business with little commercial leverage: if a key upstream index provider raised prices or withdrew, Kagi's coverage would change materially.

Privacy, Privacy Pass, and the paid-search paradox

Kagi says it builds no advertising profiles, sells no data, and proxies queries to third-party indexes so the upstream provider sees Kagi rather than the user. There is no published independent audit of that upstream handling, so it rests on the company's own description.

The substantive technical piece is Privacy Pass, and it is real. It issues cryptographic tokens that, in Kagi's words, prove you have the right to use the service without revealing who you are, so that searches cannot be connected to your account or to each other. That is stronger than a policy promise, because it removes Kagi's own ability to link a query to an identity. The documented limits are rarely reported and they matter:

The browser extension manages a pool of 300 tokens at a time, capped at 3,000 tokens per account per month. Past that you are back to ordinary authenticated search.
It is available on the Professional, Ultimate, Family and Team plans — not on the entry-level Starter plan. The strongest privacy feature is gated behind the more expensive tiers.
Its documented scope is Kagi Search authentication only; Kagi states an intention to extend it to Assistant, Maps, Translate and Summarize, so using the AI Assistant does not currently get its protection.
Supported in Orion, Chrome, Firefox, Tor Browser and Android.

Two structural caveats deserve weight. A paid service necessarily holds a billing identity — name, email, payment relationship — that a free ad-funded engine never collects, and Privacy Pass only decouples it if you are on a qualifying plan and actually using it. And Kagi Inc. is a US company in Palo Alto, California, subject to US compulsory legal process. Its privacy position rests on not retaining query-to-identity linkage, not on jurisdiction.

AI features, and the decision not to force them

Kagi's stance on AI is unusually explicit and is a genuine differentiator in 2026: AI is a feature you invoke, not a layer applied to every result page. Compared with Google's AI Overviews or Microsoft Copilot, where the generated answer is the default surface, Kagi's ordinary query returns ordinary links.

Quick Answer — an on-demand AI summary of a results page, triggered by the user rather than fired automatically.
Kagi Assistant — a chat product with web access through Kagi's own search stack, announced 4 September 2024 and opened to all users in April 2025. It is deliberately model-agnostic: the user chooses among frontier models from several vendors. Kagi does not train a frontier model of its own.
Kagi Translate, launched 2024, and Kagi News, an AI-summarised daily digest introduced 30 September 2025.
Developer AI endpoints — FastGPT, which returns grounded answers with citations, and the Universal Summarizer, both sold metered.

Kagi's published position is that it will not add AI which degrades core search, and that AI answers must cite their sources. Which model serves a given Assistant request is a user choice rather than a hidden routing decision.

Who it suits, and who it does not

Kagi is genuinely good at a small number of things. The domain-level raise, lower and block controls are the best implementation of user-controlled ranking anyone has shipped: if you keep a personal list of content farms you never want to see again, Kagi is the only mainstream product that will honour it permanently. The absence of ads is structural rather than promotional — there is simply no commercial pressure on the top of the page. And AI being opt-in is, in 2026, a position almost nobody else holds.

It is a poor choice for several kinds of user, and that is worth saying plainly:

Anyone who wants search to be free. The free tier is a trial, not a service.
Anyone who assumes paying buys independence from Google's ecosystem. It does not. Google was a documented source until May 2024 and the current mix is undisclosed.
Anyone whose searching is mostly navigational. If you type site names to get to them, every engine works and Kagi's differentiation is invisible.
Anyone needing deep long-tail coverage in a non-English language, where both the blended sources and Teclis are at their weakest.

Common misconceptions

"Kagi has its own index of the web." Half true, and the false half is the important one. Kagi has an index — Teclis, plus TinyGem for news, both built by Kagibot. It is a curated non-commercial index, not a general web index, and Kagi supplements it with third-party commercial indexes on essentially every query.

"Kagi is independent of Google." Google was explicitly named as an anonymized API source in Kagi's own documentation until May 2024. Whether it remains in the mix in 2026 is undisclosed. Neither claim can honestly be asserted.

"Kagi is just a Google or Bing wrapper." Wrong in the other direction. A wrapper does not run a crawler, maintain indexes, or re-rank. Kagi does all three, and the re-ranking is the product.

"Paying for search means better privacy." Paying introduces a billing identity that free engines never collect. Privacy Pass is a real mitigation, but it is unavailable on the cheapest plan, capped at 3,000 tokens per account per month, and currently scoped to Search rather than Assistant, Maps, Translate or Summarize. The people most likely to buy the cheapest tier get the least of it.

"Kagi is European, so it sits outside US surveillance law." Kagi Inc. is in Palo Alto, California.

"This is the Kagi that processed shareware payments." An unrelated company that happened to share the name; see the Wikipedia article on Kagi for the distinction.

Frequently asked questions

Does Kagi have its own index?

Partly. Kagi runs its own crawler, Kagibot, which builds two Kagi-owned indexes: Teclis for the web and TinyGem for news. Teclis is deliberately non-commercial and weighted toward personal sites, blogs and forums, and Kagi does not present it as a general index of the whole web. For broad coverage Kagi makes anonymized API calls into third-party commercial indexes and merges everything into one ranked page. That blend is why Kagi is classed as a hybrid rather than an own-index engine.

Does Kagi use Google results?

Google was named in Kagi's own documentation as an anonymized API source, alongside Yandex, Mojeek and Brave, until 2 May 2024. On that date the names were removed and replaced with the phrase "all major search result providers worldwide." Kagi did not announce a change to the underlying suppliers, only to the disclosure. Brave is still named specifically as a current source. Whether Google remains in the mix in 2026 is not publicly knowable.

Why does Kagi cost money?

Because it carries no advertising at all, so subscription revenue is the only revenue. Kagi's argument is that this changes who the ranking has to satisfy: with no advertiser in the system, the only party with a claim on the results page is the person paying for it. Structurally there is a capped free trial tier, monthly consumer tiers differentiated by search volume and AI access, per-seat family and team plans, and separately metered developer APIs. Specific prices have changed more than once and are not stated here.

Is Kagi actually private?

It holds more identity than a free engine by necessity, and mitigates it technically. Because you pay, Kagi has your name, email and a payment relationship. Its Privacy Pass system issues cryptographic tokens that prove entitlement without revealing identity, so searches cannot be linked to an account or to each other. But Privacy Pass is not offered on the entry-level Starter plan, is capped at 3,000 tokens per account per month, and is documented as covering Search only, not Assistant.

Who owns Kagi?

Kagi Inc., an independent US company founded by Vladimir Prelovac in Palo Alto, California in May 2018. It has never been acquired and has no parent company. It incorporated in March 2022, launched paid public beta the same year, and converted to a Public Benefit Corporation in 2024. It is not related to the 1990s shareware payment processor that also traded as Kagi.

How many people use Kagi?

Kagi's own live stats page reported 74,331 members and 1,098,200 queries in the preceding 24 hours when read on 19 August 2026, with the latest data point dated 18 August. The same source showed roughly 50,000 members and 845,200 searches on 9 June 2025. The figures are self-reported and unaudited. Kagi is far below StatCounter's reporting threshold, so it has no published market-share percentage.

Does Kagi force AI answers on you?

No, and that is one of its clearer differences from mainstream engines. An ordinary Kagi query returns ordinary links. The Quick Answer summary is invoked by the user rather than fired automatically, and Kagi Assistant is a separate chat product that lets the user pick which vendor's model to use. Kagi's published position is that it will not add AI features which degrade core search, and that AI answers must cite their sources.

Is Kagi a good alternative to Google?

It depends entirely on what you want out of the switch. If you want control over ranking and no ads, it is the strongest product available and nothing else comes close on domain-level blocking. If you want an index that does not depend on the incumbents, Kagi does not deliver that and does not claim to. It is also a poor fit for mostly navigational searching, and weakest on deep long-tail queries in languages other than English.