- Search / session-token HTML search - fully implemented and live-verified for base search, lens-aware search, and runtime search filters (
r,dr,from_date,to_date,order,verbatim,personalized) - Search / official API-token path - implemented for base search only; if Kagi rejects the API-token search path, base search falls back to session-token search when available
- Universal Summarizer API - implemented on the documented paid public API path
- FastGPT API - implemented on the documented paid public API path
- Web and News Enrichment API - implemented on the documented paid public API path
- Small Web RSS feed - implemented and live-verified
- Subscriber web Summarizer - implemented on the session-token web-product path via
kagi summarize --subscriber ... - Kagi News public product endpoints - implemented via
kagi news ... - Subscriber web Quick Answer flow - implemented on Kagi's authenticated Quick Answer stream via
kagi quick ... - Subscriber web Assistant prompt flow - implemented on Kagi Assistant's authenticated tagged stream via
kagi assistant ... - Subscriber web Assistant thread list/open/delete/export flows - implemented on the authenticated Assistant thread endpoints via
kagi assistant thread ... - Kagi Translate text mode - implemented via
kagi translate ...with runtime bootstrap fromKAGI_SESSION_TOKEN
According to Kagi's public API docs, the documented API families are:
- Kagi Search API
- Web and News Enrichment API
- Universal Summarizer API
- FastGPT API
- Kagi Small Web RSS feed
This CLI also implements non-public or product-only seams:
- subscriber web Summarizer via Kagi session-token auth
- subscriber web Quick Answer via Kagi session-token auth
- subscriber web Assistant prompt flow via Kagi session-token auth
- subscriber web Assistant thread management via Kagi session-token auth
- Kagi News product endpoints
- Lens support and runtime search filters are not documented on the official Search API. In this CLI they work through Kagi's live HTML/session flow using the
l,r,dr,from_date,to_date,order,verbatim, andpersonalizedquery parameters. - The official Search API uses
Authorization: Bot <token>onhttps://kagi.com/api/v0/search. - Search API access is still account-gated in practice, and API-token search can also fail for billing reasons.
- Base-search fallback to session-token search happens on the user-facing
searchcommand only.auth checkvalidates the selected primary credential without fallback. - The paid public Summarizer, FastGPT, and Enrichment APIs require
KAGI_API_TOKENand sufficient API credit. - The subscriber web Summarizer requires
KAGI_SESSION_TOKENand uses the authenticatedGET /mother/summary_labs?...stream path instead of the public/api/v0/summarizeendpoint. - Live verification on March 18, 2026 showed that direct HTTP bootstrap can mint
translate_sessionfrom the sameKAGI_SESSION_TOKENby reading theSet-Cookieheader fromhttps://translate.kagi.com/. - After bootstrap, the CLI uses normal Rust HTTP requests for
/api/detect,/api/translate,/api/alternative-translations,/api/text-alignments,/api/translation-suggestions, and/api/word-insights. - Assistant requires
KAGI_SESSION_TOKENand currently targets/assistant/prompt,/assistant/thread_list,/assistant/thread_open,/assistant/thread_delete, and/assistant/<thread_id>/download. - Quick Answer requires
KAGI_SESSION_TOKENand currently targetsPOST /mother/context?q=...withAccept: application/vnd.kagi.stream. - News uses
https://news.kagi.com/api/...JSON endpoints and does not require auth.