← Research

KagiKagi

Kagi

订阅制Subscription 自建索引Custom index 无广告无追踪No ads or tracking FastGPTFastGPT 隐私向Privacy-focused partial lifecycle: active 核验Checked 2026-10-22

Kagi 是「付费换隐私」的自建索引搜索引擎：无广告无追踪是主轴，AI 答案层（FastGPT/Teclis）是附加——它与「AI 搜索」的顺序正好相反。

A self-built-index search engine where you pay for privacy: no ads and no tracking is the main axis, the AI answer layer (FastGPT/Teclis) is the add-on — the exact reverse order of "AI search".

适合与不适合When to use it, when not to

适合：付费换无追踪体验、「搜索+全文读」答案、对追踪敏感用户。不适合：免费用户、中文检索主打、依赖国内支付面。

Users willing to pay for ad-free/tracking-free search, "search + full-text reading" style answers (the Teclis line), privacy-sensitive users. Not for: free-first users (no free tier as the main axis), mainland direct-access sensitivity (direct 200 reachable but payment is overseas-style), Chinese retrieval as the main axis.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-22

出品属地Origin: 国外出品foreign-made　Kagi（美国，创始人团队叙事为前 Google/Uber 工程师）。Kagi (US; founder-team narrative: ex-Google/ex-Uber engineers).
国内使用Direct access: 大陆可直连direct OK　kagi.com 大陆直连实测 200（2026-10-22）。kagi.com measured 200 direct from mainland (2026-10-22).
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-22 检索）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (searched 2026-10-22); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 自建搜索索引（Kagi 自建索引是核心叙事）+ FastGPT 等 LLM 答案层（可切换多家模型，官方产品面）；「搜索是本体、AI 是附加」的顺序与多数 AI 搜索相反。
A self-built search index (Kagi's self-built index is the core narrative) + FastGPT-style LLM answer layers (multiple models switchable, official product surface); the "search is the substance, AI is the add-on" ordering is the reverse of most AI search products.
运行位置Where it runs: 网页（kagi.com）+ 浏览器扩展 + Mini Bar（官方产品线）；Kagi FastGPT/FastGPT 类问答在云端；搜索索引自建。
Web (kagi.com) + browser extension + Mini Bar (official product line); Kagi FastGPT/FastGPT-style Q&A on the cloud; the search index self-built.
本地文件Local files: 搜索历史可选本地化/零留存模式（官方隐私叙事：不追踪不画像）；未见官方声明的其他本机数据面。
Search history can optionally be localized / zero-retention mode (official privacy narrative: no tracking, no profiling); no other local data surfaces officially stated seen.
关机后的任务Tasks after shutdown: 搜索即时返回；未见官方任务队列或「关机续跑」类后台任务面（未逐条核验）。
Search returns instantly; no official task queue or "keep-running-after-shutdown" style background task surface seen (not verified item by item).
工具与扩展Tools & extensions: 无广告搜索（招牌）、Teclis/Exa 类「先搜后 LLM 读全文」答案线（官方产品面）、 Universal Summarizer（URL/视频/文档摘要）、翻译；隐私承诺是产品主轴之一（无追踪叙事）。
Ad-free search (the signature), Teclis/Exa-style "search first, then LLM reads full texts" answer lines (official product surface), Universal Summarizer (URL/video/document summarization), translation; the privacy promise is one of the product's main axes (no-tracking narrative).
上下文与记忆Context & memory: 无对话记忆主打——搜索按会话制组织；跨会话记忆与画像未核验到官方声明（隐私姿态反而强调不追踪）。
No conversation-memory selling point — search is organized per session; cross-session memory and profiling not verified to an official statement (the privacy posture instead stresses not tracking).
权限与限制Permissions & limits: 订阅制（D9：价格不采）；「不追踪、不卖数据」是官方隐私叙事—— 条款级落点未逐条核验；付费换隐私是它的产品逻辑。
Subscription-first (D9: pricing not collected); "no tracking, no selling data" is the official privacy narrative — terms-level landing not verified item by item; paying for privacy is its product logic.
适合什么任务Best for: 适合：愿意付费换无广告/无追踪搜索体验、要「搜索+全文读」式答案（Teclis 线）、对追踪敏感的用户。不适合：免费用户（无免费档主打）、大陆直连敏感（直连 200 可达但支付面按海外口径）、要中文检索主打。
Fits: users willing to pay for an ad-free/untracked search experience, wanting "search + full-text reading" style answers (Teclis line), tracking-sensitive users. Does not fit: free users (no free tier as the lead), mainland-direct-sensitive (direct 200 reachable but payments follow overseas conventions), wanting Chinese retrieval as the mainstay.

头号误解Top misconception

Kagi 是「搜索引擎+AI 附加」不是「AI 搜索」——索引自建，答案层是附加；对比口径别错位Kagi is "search engine + AI add-on", not "AI search" — the index is self-built and the answer layer is the add-on; keep the comparison frame straight
订阅制是产品姿态——没有免费档兜底，「先试试」的成本结构与其他家不同（D9：价格不采）Subscription is the product posture — no free-tier fallback, so the "just try it" cost structure differs from the others (D9: no pricing relayed)
kagi.com 大陆直连实测 200（2026-10-22）——可达，但账号/支付按海外口径kagi.com measured 200 direct from the mainland (2026-10-22) — reachable, but account/payment follow the overseas route

未知项清单Known unknowns

自建索引规模与更新频率（未核验到官方声明）
各答案线当前可切换模型清单（未逐条清点）
隐私承诺的条款级落点（未逐条核验）
中文查询的实际检索质量（未做实测）

Self-built index size and update frequency (no official statement verified)
The currently switchable-model list for each answer line (not inventoried item by item)
The terms-level landing of the privacy promises (not verified item by item)
Actual retrieval quality for Chinese queries (not measured)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-22，内容更新日content updated on 2026-10-22。

类型Type	名称Name	链接Link
official	Kagi · 官网（产品与能力叙事）Kagi · official site (product and capability narrative)	https://kagi.com
official	Kagi · Privacy（官方隐私页，2026-10-22 直抓核验）Kagi · Privacy (official privacy page, direct-fetch verified 2026-10-22)	https://kagi.com/privacy
changelog	Kagi · Blog（官方更新线，changelog）Kagi · Blog (official update line, changelog)	https://blog.kagi.com

实测记录Measured runs

本档未跑检索质量评测；结论只覆盖产品能力与来源披露。

实测建议（三条，都指向官方可读页没给的东西）：

自建索引 vs Google/Bing 的长尾查询覆盖差
Teclis「先搜后读全文」答案与 Perplexity 的质量对比
零留存模式的实际生效验证（历史行为观察）

This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to creation products).

Suggested measurements (three, all pointing at things the official readable pages do not give):

Long-tail query coverage gap: self-built index vs Google/Bing
Teclis "search then read full text" answers vs Perplexity's quality
Actual verification that zero-retention mode takes effect (history-behavior observation)