What is Kimi

Kimi is the AI assistant built by Moonshot AI. Many people first met it as the assistant that could read very long documents in one go. It has since grown into a broader platform for knowledge work and agentic coding. You can still chat and search the web, but you can also hand it a whole task and let it plan the steps, call tools and deliver a report, a slide deck or a working website.

Kimi runs on Moonshot’s own Kimi model family. According to the Kimi API platform documentation, the current flagship is Kimi K3, aimed at long-horizon coding and end-to-end knowledge work, with a 1M-token context window and native vision. Alongside it are Kimi K2.7 Code for programming and Kimi K2.6, which accepts text, image and video input and can switch between thinking and non-thinking modes.

Everyday users reach Kimi through kimi.com and the mobile app. Developers can call the same models through the API platform or use the Kimi Code programming tool.

In one line, Kimi’s evolution has been from reading to doing. Early users mostly had it read long documents and answer questions. Now the focus is on delivering results: you set the goal, and it breaks the job down, searches, uses tools and assembles the output with as little hand-holding as possible. You can see this in the product itself. Alongside chat, the web sidebar lists slides, deep research, documents, scheduled tasks and plugins, each tied to a kind of finished work.

Key features

Chat and search. The basic mode is conversation. Kimi can search the web and organise what it finds, and you can upload files and images for it to read. A memory space lets it remember your preferences and background so you repeat yourself less.

Agent mode and agent swarms. For complex jobs, switch to Agent mode and let Kimi break the work down and execute it with tools. Kimi also offers an agent swarm, where several agents work on parts of a large task in parallel.

Deep research. Given a research question, Kimi runs several rounds of searching, reading and summarising, then produces a structured report. It suits market research and literature reviews where a lot of checking is needed.

Slides, docs and sheets. The PPT assistant builds presentations from a topic or existing material, with template choice and automatic layout. Document and spreadsheet agents handle writing, editing and data work.

Build apps. Describe what you want and Kimi builds a web app, deploys it for sharing and keeps iterating with you. The site also shows it building playable multiplayer and 3D games.

Kimi Work. A workspace for ongoing work, with a goal mode, plugin centre, widgets and boards, for handing Kimi a long-running task to push forward.

Scheduled tasks, plugins and skills. Set tasks to run on a schedule, and install or create plugins and skills to extend what Kimi can do. There is also a browser extension.

Kimi Code and the open API. Kimi Code includes a command-line tool for programming. The Kimi API is compatible with both OpenAI and Anthropic formats, so existing SDKs work, and it can serve as the backend model for coding agents such as Claude Code.

How to use it

  1. Open the web app or mobile app. Go to kimi.com and sign in, or install the Kimi app.
  2. Ask directly. Describe your question and attach files or images if needed. Specific questions get better answers.
  3. Give complex tasks to Agent mode. For example: “Research the LLM APIs of three major cloud providers, write a comparison report, then turn it into a 10-slide deck.” Choose Agent mode or deep research and let it work through the steps.
  4. Use the dedicated tools. The sidebar links straight to slides, deep research and documents, and Projects keep related conversations together.
  5. Integrate as a developer. Create an API key on the Kimi API platform (platform.kimi.com), point base_url at Kimi’s endpoint and call the models with the OpenAI or Anthropic SDK. The docs recommend storing the key in an environment variable rather than in code.

A few API details are worth knowing. The platform offers three interfaces: Chat Completions, Responses and Messages (Anthropic-compatible). Rate limits are measured four ways (concurrency, requests per minute, tokens per minute and tokens per day) and apply per user, not per key. Input plus output must fit within the chosen model’s maximum context. As a rule of thumb from the docs, one token is roughly 1.5 to 2 Chinese characters in ordinary text.

Use cases

  • Digesting long material: summarise, question and compare papers, contracts and financial reports.
  • Research and reports: market studies, competitor analysis and literature reviews with deep research.
  • Office output: first drafts of slide decks, cleaned-up documents and spreadsheet analysis.
  • Small websites: portfolios, landing pages, event pages and simple games.
  • Programming: use Kimi Code or plug Kimi models into your coding agent.

FAQ

Are Kimi and the Kimi models the same thing? Not quite. Kimi is the user-facing assistant. Kimi K3, K2.7 Code and K2.6 are the models behind it, which developers can call directly through the API.

How does Agent mode differ from normal chat? Normal chat is one question, one answer. Agent mode splits a task into steps, searches, uses tools and combines the results. It fits multi-step work and takes longer.

Is Kimi free? There is a free tier and paid memberships. Check the membership section of the help centre for current tiers and billing.

Can businesses use it? Yes. Kimi Business adds member management and shared workspaces, and companies can also work with the sales team on API agreements.

Official website