Kagi

license

Proprietary(API-only)

source

closed(no self-hostable index

managed

cloud SaaS with API keys, $12 per 1,000 search requests, $4 per 1,000 extracted pages

Kagi's search index is closed and reachable only through a paid API key, with no self-hostable core. The OpenAPI specification document itself carries an Apache-2.0 licence, but that covers the spec file rather than the service; the generated Python and TypeScript clients carry no licence, and only the separate MCP server is MIT.

• https://kagi.com/api/pricing recorded 2026-09-17

  The v1 API price list: Search "$12 /1k requests", Extract "$4 /1k pages", and an Enterprise "Get a quote" tier. Establishes a metered closed service with no self-hosting option.

Kagi publishes a subscriber count for its consumer search product, not for API callers, so that figure measures the wrong population. The `kagiapi` package on PyPI sees too little use to serve as a proxy, and Kagi's own generated v1 clients are not published to any package registry, so no channel measures how the API itself is used.

• https://kagi.com/api/pricing recorded 2026-09-17

  The vendor's own pricing and product documentation, carrying rates and limits and no request, customer or developer count - which is why no level is recorded.

Kagi's `/search` endpoint can run its `extract` option inline, replacing each result's snippet with the full page markdown, so one call returns both references and their content. That puts it level with Firecrawl, which sells the same find-plus-content combination; FastGPT, which composes a cited answer, is a separate product at its own price and is not part of what is scored here.

• https://kagi.com/api/docs/_spec/openapi.yaml recorded 2026-09-17

  The v1 OpenAPI document. Two paths: `/search` "Perform a web search" returning searchResult objects of url, title and snippet, and `/extract` "Extract page content as markdown from URLs". The `/search` request's `extract` property: "Configuration for extracting page content from search results. When provided, the API will fetch and extract the content from the specified number of result pages. The resulting page markdown will update the value of the snippet field on the respective result item."