Moonshot AI · Training
KimiBot
Nothing this token fetches reaches a surface anyone will see. Moonshot describes it as crawling "content potentially used to train Kimi's foundation models" (the hedge in "potentially" is the vendor's own), and what it collects becomes neither a queryable index nor an answer someone is waiting on, which is precisely why it needs its own name to be governable at all. For an operator that means its visits are invisible downstream: no impression, no citation and no referral will ever trace back to one. The `kimibot.json` range list does not narrow that down either, since it is Moonshot's single list republished at a third URL: a matching address rules out an impostor while saying nothing about whether training, indexing or a live user request brought the crawler to your door.
- Operated by
- Moonshot AI
- Purpose
- Training
- robots.txt token
- KimiBot
- Verification
- IP verified
Collects pages into a corpus used to train models.
How to verify KimiBot
Moonshot AI publishes the IP ranges KimiBot crawls from, and we fetch that list on a schedule. A request claiming to be KimiBot can therefore be checked against 4 published ranges: one that does not match is not this crawler.
Published ranges last confirmed by us on .
User agent
Moonshot AI publishes this user agent for KimiBot. Match on the KimiBot product token rather than the whole string: vendors revise the surrounding version and URL fragments without notice.
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; KimiBot/1.0; +https://www.kimi.com/policies/kimi-crawlersrobots.txt for KimiBot
Yes, and the vendor frames the directive as carrying meaning past mere access control: "Disallowing this bot signals that your content should not be used for model training." The same four patterns apply (full block, path scoping, `Crawl-delay`, per-token stanzas), and this is the token used in every worked example on the policy page. One matching caveat before you write the rule: the published Kimi user-agent strings put the `compatible;` clause outside the parenthesised comment rather than inside it, so any parser that reads the token from within the first set of parentheses will miss all three bots; substring-match the bare token instead. Moonshot also publishes a mailbox, `kimibot@moonshot.ai`, and asks for the domain and logs when reporting misbehaviour.
Block
User-agent: KimiBot
Disallow: /Allow
User-agent: KimiBot
Allow: /robots.txt is a request, not an enforcement mechanism. It is honoured by convention, and a crawler that ignores it is stopped at your edge, not in a text file.
What blocking KimiBot costs you
Close to nothing in traffic terms, and the vendor's own page effectively concedes it: this is the only one of its three sections with no stated downside, describing the effect purely as a training-consent signal. Disallowing it leaves your standing in Kimi search results and your availability for on-demand summarising untouched, because those belong to the other two tokens. That makes this the cleanest separation on offer anywhere in the directory (refuse training use, keep both discovery surfaces, in a single stanza), and the only thing given up is whatever indirect benefit follows from your text informing a future model. Keep one counter-intuitive point in view while judging the vendor: publishing a per-token policy with distinct purposes and machine-readable ranges at all, even ranges that turn out to be one list served three times, is more than most far larger vendors in this directory manage.