Kagi Search and Kagi Assistant Explained: Plans, Models, Sources and Privacy

Kagi is a paid search engine with no ads, and its Assistant gives subscribers access to models from many AI providers in one place, which changes the usual trade between price, privacy and choice. Kagi Search is a subscription web search engine from Kagi, a Delaware public benefit corporation founded in 2018, and Kagi Assistant is its AI chat product that combines large language models with optional Kagi Search results under the same subscription. Kagi's company page dates its launch to June 2022. This page sets out the plans and what each includes, where Kagi's results come from, how Quick Answer and the Assistant work, which model providers are offered, what Kagi says it keeps, and how the service has developed. It is written for individuals, families, teams, privacy teams and publishers who want facts from Kagi's own pages rather than commentary.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It draws on Kagi's public help documentation as of November 2026; plans and features change, so confirm details against Kagi's current pages.

Key takeaways

• Paid, no ads. Kagi is funded by subscriptions: as of 2026-11-15 the pricing page lists a free trial of 100 searches, Starter at $5, Professional at $10 and Ultimate at $25 a month, plus sales tax.
• Many sources per query. Kagi says each search typically queries about a dozen sources at once: its own Teclis web index and TinyGem news index, anonymised queries to third-party search providers, and vertical data sources and APIs.
• Multi-provider Assistant. Kagi Assistant lists models from 12 providers; it is available to all members, while premium models need the Ultimate plan.
• Quick Answer from a question mark. Ending a query with "?" triggers Quick Answer, an AI summary of the returned pages with references.
• Short default retention. Kagi says Assistant threads are deleted after 24 hours of inactivity by default, that data is never used to train AI models by Kagi or the model providers, and that account information is not shared with providers.

What is Kagi?

Kagi runs a family of products around paid search. Kagi Search returns results with no ads, Quick Answer adds AI summaries to the results page, and Kagi Assistant is a chat interface for many large language models with optional web search. Kagi states that its products are designed without ads, trackers or distractions.

How do you access Kagi?

Kagi needs an account: searches run at kagi.com once signed in, starting with a trial of 100 searches and 100 AI interactions. Privacy Pass lets members search without the request being tied to their account, and Tor access is offered. Members can also tune results, as below.

How do you shape Kagi results?

Lenses restrict which sites and parameters appear in results. Kagi enables eight built-in lenses by default: Forums, Programming, News 360, Fediverse Forums, Usenet/Archive, Academic, PDFs and Kagi Documentation. Cyber Security, Small Web and Recipes can be turned on in Lenses settings.

• Custom lenses: set a region, up to 10 included and 10 excluded websites, up to 5 included and 5 excluded keywords, a file type, and before or after date limits.
• Scope: lenses do not yet apply to image or video searches, per Kagi's help page.
• Sharing: a lens can be shared by link; recipients get an editable copy that does not receive later updates, and turning off sharing invalidates the link.
• Personalised results: Kagi applies a member's Lenses and Personalized Results settings to Assistant searches too.

How does Kagi find and cite sources?

Kagi's search sources page names three kinds of source and says a typical search queries about a dozen of them at once.

Kagi also says it down-ranks pages heavy with ads and trackers. That ranking choice, and the mix of providers, means a publisher's visibility in Kagi depends partly on the third-party engines it queries as well as on Kagi's own indexes.

What is Quick Answer?

Quick Answer summarises the pages returned for a search and provides references to the pages used, so the reader can check sources.

The automatic question mark trigger can be turned off with the Auto Quick Answer toggle on the Search AI Settings page (kagi.com/settings/ai). The Quick Answer help page does not name the model behind it.

Which models power Kagi?

Kagi Assistant combines large language models with optional Kagi Search results. Kagi describes these features:

• Web access toggle: powered by Kagi Search, and it can be switched on or off mid-thread.
• Model switching: change model within a thread.
• Custom assistants: each with its own model, instructions and search settings.
• Inputs: file uploads, voice input and code syntax highlighting.
• Sharing and export: threads export as Markdown or JSON and can be shared by link.
• Research mode: listed on the pricing page for the Ultimate plan.

As of 2026-11-15. The model list changes often; the provider names above are taken from Kagi's Assistant help page, and the pricing page names Claude, DeepSeek, Gemini, GPT, Grok and Mistral model versions among Ultimate models. Check Kagi's page for exact model versions on a given day.

How much does Kagi cost?

Prices are from Kagi's pricing page and plan help page, in US dollars before sales tax, as of 2026-11-15.

• Annual billing: the pricing page offers 10% off for annual billing; a yearly Starter plan includes 3,600 searches and 3,600 AI interactions for the year.
• What counts as a Starter search: bangs and search suggestions do not count, reloading the same search within about 2 minutes does not count, each lens and content type counts separately, loading more results counts, and unused searches do not roll over.
• Changing plans: upgrades and downgrades can happen at any time with the unused balance prorated. PayPal and OpenNode payments do not create a recurring subscription.
• AI usage limits: the Assistant help page describes usage as value-based and tied to the plan price, with a 20% markup on token costs; when the limit is exceeded, AI interactions are disabled until the next billing cycle. The pricing page also refers to a Fair Use Policy and to extra payments to resume access.

How do ads work in Kagi?

They do not appear. Kagi's pricing page says Kagi "has no ads, no tracking, no noise", and the service is funded by subscriptions. Kagi's search sources page also says it down-ranks pages heavy with ads and trackers.

What can publishers control?

Kagi's search sources page names its own indexes, Teclis for the web and TinyGem for news, alongside anonymised calls to third-party search providers, specialised engines such as Marginalia and vertical sources. It does not describe a crawler user agent or any publisher opt-in or opt-out. Domain promotion and blocking are member settings that shape one member's results, not publisher controls.

What privacy controls exist?

From Kagi's privacy policy and Assistant help page, as of 2026-11-15.

The terms are governed by Delaware law. GDPR and CCPA rights are set out in a separate Privacy Rights Addendum. Because Kagi is a paid account, the account itself links a person to a subscription; Privacy Pass is the mechanism Kagi offers for searching without tying requests to that account.

When was Kagi released, and how has it changed?

As of 2026-11-15.

What are Kagi's limitations?

• Usage caps: Starter has 300 searches a month, and AI use on every plan is capped by value; when it runs out, AI interactions stop until the next cycle.
• Premium models need Ultimate: other plans use standard models.
• Summaries can be wrong: Quick Answer and the Assistant summarise sources; open the referenced pages for anything legal, medical, financial or security related.
• Long threads lose detail: history beyond about 32k tokens is summarised or truncated.
• Retention depends on settings: a member who switches thread retention to permanent keeps threads beyond the 24 hour default.
• Model behaviour varies: answers depend on which provider and model is chosen in a thread.

Common questions about the service

Is Kagi free?

No, apart from the trial of 100 searches and 100 AI interactions. Kagi is subscription-funded and carries no ads; paid plans start at $5 a month plus sales tax as of 2026-11-15.

Does Kagi use its own index?

Partly. Kagi runs its own Teclis web index and TinyGem news index, and combines them with anonymised queries to third-party search providers and vertical data sources for each search.

Where identity and security fit

Kagi Assistant lets one account reach models from many providers, so an organisation's AI use policy should name which tools and models staff may use and what data may never be pasted in; the Safe Use of AI Chatbots Guide covers this. Shared thread links and permanent retention settings deserve a rule too. Assistant can fetch URLs and search the web, so treat retrieved content as untrusted, as explained in LLM Grooming and AI Search Poisoning.

Practitioner checklist

• Pick a plan by expected volume: Starter caps searches at 300 a month, Professional and Ultimate are unlimited for search.
• Decide whether premium models are needed before choosing Ultimate over Professional.
• Keep the 24 hour Assistant thread deletion default unless records are needed, and set a rule for shared thread links.
• Turn off Auto Quick Answer if AI summaries are not wanted on question-style queries.
• Use Privacy Pass where searches should not be tied to the account.
• Set a policy on which Assistant models and data types staff may use for work.
• Recheck prices, limits and the model list on Kagi's pages before decisions; they change.

Standards and references

• Kagi: Pricing
• Kagi Help: Plan types
• Kagi Help: Kagi Assistant
• Kagi Help: Quick Answer
• Kagi Help: Search sources
• Kagi Help: Lenses
• Kagi: Privacy policy
• Kagi Help: Company

Related NHI Mgmt Group resources: AI Answer Engines Hub · Safe Use of AI Chatbots Guide · LLM Grooming and AI Search Poisoning