Kimi is a long-context AI assistant built by Moonshot AI. Where most chatbots lose the thread after a few thousand words, Kimi is designed to hold an entire novel, codebase, or research dossier in working memory and answer questions about it without forgetting earlier sections. That single capability — a context window that scales to roughly two hundred thousand tokens in the consumer product and higher in enterprise tiers — is the reason Kimi has become a serious tool for students, researchers, lawyers, developers, and analysts who cannot fit their work into the smaller windows of competing chatbots. If you cannot explain what a tool is in one sentence, you cannot use it deliberately. This lesson gives you that sentence and the mental model behind it.

Kimi is a browser-based AI assistant available at kimi.com (and via mobile apps on iOS and Android) developed by Moonshot AI, a Beijing-based startup founded in 2023. It runs on Moonshot's own family of large language models called Kimi, which were trained with an emphasis on long-document comprehension, retrieval, and Chinese-English bilingual reasoning. Three capabilities define the Kimi experience. First, long context: the free product accepts up to two hundred thousand tokens of input — roughly the text of a four-hundred-page book — and the model can cite back specific passages from that input. Second, integrated web search: when you ask a question about recent events or facts Kimi does not know, it can issue live web searches, read the results, and synthesize an answer with citations. Third, file understanding: you can upload PDFs, Word documents, Excel spreadsheets, code files, slide decks, and images, and Kimi will read, summarize, extract data from, and answer questions about their contents. These three pillars — long context, live search, and file understanding — make Kimi a working tool rather than a chatbot.

Under the hood, Kimi takes your text and any attached files or search results, encodes them into its context window, and then generates a response token by token. The model attends over the entire context simultaneously, so a question about page three hundred of an uploaded document can reference a detail on page one. When Kimi decides it needs fresh information, it issues a web search behind the scenes, reads the top results, and incorporates them with citations. When you upload a file, Moonshot's servers extract text and structure from it before sending to the model, so a PDF with tables becomes structured information Kimi can reason about. The result is a single interface where you can paste a contract, ask for risky clauses, request citations, and get an answer that quotes the contract directly.

1. Open kimi.com

2. Start a new chat

3. Ask a factual question

4. Try a follow-up

5. Check the source

A graduate student preparing for comprehensive exams uploaded a thirty-page academic paper, a fifty-page literature review, and her own forty-page draft thesis into a single Kimi conversation. She then asked Kimi to identify which arguments in her draft were already addressed in the literature review, which were novel, and which were contradicted by the paper. Kimi produced a structured critique referencing specific pages of all three documents. The same task in a chatbot with a smaller context window would have required splitting the documents into chunks and stitching the answers together manually — a process that took her peer an entire weekend.