Contents
After Kimi finishes, does this happen?
The long read and the code fix were done in one go with Kimi. And yet, here is what happens next.
You got the key points. What the meeting decided is buried in some other note.
The fix is committed. Nobody has written the review request or told the client.
Owners and deadlines were agreed out loud. The one checking is always you.
Kimi gets the reading and writing done fast. Whether the work moves is decided after it finishes.
What you really wanted after Kimi finishes
Same meeting. Same Kimi. The moment it finished, you wanted this.
Reading and code stay with Kimi. The agreed fixes, owners and dates are already lined up.
The review request and the client update are ready, in your words.
SIMY flags work waiting on a reply or a review before your next meeting.
"Release this week?" That one question is all that comes back.
This starts from the next meeting after you connect your meetings and chat to SIMY. Keep using Kimi for what it does best.
Where Kimi's answer stops
Kimi is great at reading and writing what you ask for. What comes after, nobody picks up.
- QuestionYou ask
- Long readThe whole document
- Code / draftKimi Code too
- Owners and dates
- Review request
- Progress update
- Follow-up check
* Flow shown for illustration only
Kimi chat, the API and Kimi Code all stop at the same place. You can get answers as often as you like. But you ask every time, and you check whether it got done.
SIMY connects the rest. Keep Kimi, hand off what comes after
SIMY fills the four blanks. Only one square comes back to you: the decision.
- Question
- Long read
- Code / draft
- Owners and datesPicked up from meetings
- Review requestPosted once approved
- Progress updateDrafted in Gmail
- DecisionOne question back
* Flow shown for illustration only
- Picks up work from meetings and chat
- Does it your way
- Returns only decisions
SIMY finds the follow-up work in your meeting records and chats on its own. It learns the steps and checks you repeat from your conversations, does the work the same way, and lists what's done in My Actions.
- Sign up for SIMY and connect ChatGPT (Codex)SIMY's cloud runs use your own ChatGPT account. Codex usage falls within your ChatGPT plan.
- Install the SIMY desktop appIt can use Claude Code on your computer to do the work. Repeat work in your Claude Code and Codex chats becomes "Suggestions from SIMY." Chat text is not stored on our servers. Download here
- Connect your meeting recordsConnect Plaud, Notion and more on the integrations screen. Paste the key points you got from Kimi into a meeting or chat, and SIMY picks them up too.
* SIMY screen shown for illustration only
How SIMY relates to Kimi: SIMY does not connect to Kimi at this time. SIMY itself runs on your ChatGPT (Codex) account and on Claude Code on your computer. Keep using Kimi for reading, drafting and code. Bring the results into your meetings and chats, and SIMY takes it from there.
One meeting, side by side: Kimi's summary and SIMY's next steps
A sample from a 45-minute design review about reworking the checkout screen, summarized with Kimi. Kimi on the left, SIMY on the right.
- Decision
- Replace the card entry form with the new payment SDK. Keep the old screen running alongside for one month.
- Impact
- Three places: the order confirmation API, the receipt email and refunds in the admin console.
- Open question
- Release this weekend, or after next week's team meeting?
- To-dos
- Issue test environment keys. Tell the client the switchover date.
Replace the SDK: Alex, Oct 6. Check refunds: Sam, Oct 7. Issue test keys: Infra, Oct 3.
"Here's the PR replacing the payment SDK. Please look closely at the impact on refunds." Posted to chat once you approve.
Candidate switchover dates, and the old screen staying for a month, in your usual wording. Saved as a Gmail draft.
Release this weekend, or push to next week? Only what needs your decision comes back to you.
Kimi reads and writes. SIMY moves the work. Whether it goes out is still your call.
By industry: how people use Kimi, and what SIMY delivers
After you read and write with Kimi, here is what SIMY prepares. Solo or as a team, the pattern is the same.
Dev teams Kimi: Kimi Code, code generation
- Design review→Agreed fixes and owners
- After the code is written→A review request
- End of the week→Progress and pending decisions
Research Kimi: reading long documents
- Reading a report→Points adopted and their owners
- Findings meeting→Follow-up research request and due date
- Before the deadline→Reminders for work awaiting replies
Consulting and planning Kimi: first-draft proposals
- Client meeting→To-do list and a thank-you email draft
- Internal planning meeting→Chosen idea, owners and deadlines
- Day before the pitch→Pending items and one decision
Leadership meetings and 1:1s Kimi: summarizing decks, framing issues
- Leadership meeting→Decisions and owners, shared with the right people
- 1:1→Progress and pending decisions before you have to ask
- End of the week→A weekly report across meetings
Get started: keep Kimi, 3 steps
- Use Kimi as usualReading, drafting, code. Kimi chat, the API or Kimi Code all work.
- Sign up for SIMY, connect ChatGPT and install simyCloud runs use your ChatGPT (Codex) account. How to connect is shown above. If you use Claude Code, add the SIMY desktop app too. Download here
- Finish one meetingOwners and deadlines, request drafts and update drafts arrive. Decide and send. That's it.
How sign-up works: choose a plan → verify your email → pay by card. There is no free plan. See pricing. Get the desktop app from the download page.
Nothing changes in how you use Kimi. SIMY doesn't replace Kimi. It is a separate system that handles the work after Kimi finishes writing.
What is Kimi, and what is Kimi K3: usage, API, Kimi Code and data (for a closer look)
Open the sections below when you want to check something about Kimi itself. Model names and specs are as of October 2026.
What is Kimi: which country is it from (Moonshot AI)An AI built by Moonshot AI in Beijing, China
Who makes Kimi
Kimi is a chat AI and a series of large language models (LLMs) built by Moonshot AI, an AI company in Beijing, China. Kimi for China is run by the Beijing entity, while the international API (Kimi Open Platform) is provided by MOONSHOT AI PTE. LTD., a Singapore entity.
Three ways to use it
- Kimi chat: talk to it on kimi.com or the mobile app. Free to start, with paid memberships.
- API: build it into your own systems and dev tools through Kimi Open Platform. You pay for what you use.
- Open weights: run the model weights published on Hugging Face on your own servers. License terms apply.
What sets Kimi apart
It handles long context in one go and comes with a full set of coding tools. The latest Kimi K3 handles a 1M-token context and accepts images and video as well as text. Moonshot AI also publishes Kimi Code CLI, which runs in the terminal.
Searching for "Kimi" also brings up other things with the same name, such as F1 driver Andrea Kimi Antonelli. This page is about Moonshot AI's Kimi.
How to use Kimi chat: web and appFree to start. A membership raises your limits
- Open kimi.com. It works in the browser. There is also a mobile app.
- Log in. With an account, your chat history is saved.
- Start asking. Attach PDFs or images and ask things like "Put the key points in a table" or "Explain this diagram."
- Hand over long documents at once. Reading several files together to compare or summarize them is where Kimi shines.
Paid memberships
Kimi's paid membership is credit-based, and the official help lists four monthly tiers (as of October 2026). Every tier includes Kimi K3 and an allowance for Kimi Code. Prices vary by region and currency, so check the official membership page.
Tips for work
- Say what you want before it reads: state the goal first, like "From this contract, pull out only the termination terms and penalties." Answers stay steady even on long documents.
- Ask where it found it: before using numbers or terms from a summary, ask "Which page says that?"
- Use one chat per project: it keeps answers from being pulled by earlier documents.
Turning those points into who does what by when, and chasing it, happens outside Kimi. If you do that yourself every time, see how SIMY connects.
What Kimi can do: uses at workLong documents, many files, code. Start where it's strongest
- Reading long documents: give it contracts, specs or research reports and pull out the terms and issues.
- Comparing documents: hand over several quotes or proposals and get the differences in a table.
- Understanding images and video: show it screenshots, diagrams or short videos and have it explain them.
- Coding: from Kimi Code or your editor, have it read code, fix it and add tests.
- Drafting: first drafts of reports, emails and meeting notes.
What to try first
Pick one long document you have and ask, "List three things this document needs us to decide." It's a quick way to see where Kimi is strong. Once comfortable, move on to comparing documents and coding.
In every case, Kimi gives you an answer or a draft. The owners and deadlines decided in meetings, and the messages to people involved, are picked up by SIMY from your meetings and chats and moved forward.
What is Kimi K3: how it differs from K2, and the main modelsAs of October 2026. Names and generations change often
Kimi K3 is Kimi's latest generation of models, released in July 2026 (as of October 2026). It has 2.8 trillion total parameters and uses only 104 billion of them per step: a MoE (mixture of experts, which switches between smaller models with different strengths). It handles a 1M-token context (a token is a rough unit of text length) and accepts text, images and video.
Scroll the table sideways →
| Model | Role | License |
|---|---|---|
| Kimi K3 | The latest flagship. Long context, image and video input, agent-style work | Kimi K3 License (custom) |
| kimi-k2.7-code | A K2-series model built for coding. Cheaper per token on the API than K3 ($0.95 input / $4.00 output per 1M tokens, excluding tax) | Via the API |
| Kimi K2.6 and earlier | The previous generation. Still used via the API and as open weights | Modified MIT |
K3 or K2?
- K3: when you want it to read long documents, handle images or video, or take on hard tasks.
- K2 series: when you want to run lots of code generation cheaply. Open models up to K2.6 also come with looser license terms than K3.
An easy way to remember: Kimi's model names change every few months. Remember them by role: "the latest flagship (K3)" and "the cheap one for coding (k2.7-code)."
Kimi Code CLI: a coding agent in your terminalOpen source (MIT). Works within membership allowances too
Kimi Code CLI is a coding agent that Moonshot AI publishes under the MIT license. It runs in the terminal, reads and fixes code, and runs commands. It supports MCP (a common way to connect external tools and data), and through ACP (a way for editors to call agents) you can use it from editors like Zed and JetBrains. The older Python-based kimi-cli has been archived and replaced by it.
- Install: use the official install script, or install via npm. Instructions are on the GitHub page.
- Log in: launch it, run
/login, and choose a Kimi Code account or a Kimi Open Platform API key. Members can use their plan's allowance. - Good for: developers who want to try the Claude Code or Codex way of working with Kimi's models.
Getting started
- Launch Kimi Code in the folder you want to work in. First, choose how to log in.
- Say what you want in plain language. Start from the goal: "Explain how this repository is structured" or "Fix the failing tests."
- Review changes before accepting them. Check file edits and commands before you allow them.
After you write code with Kimi Code, SIMY picks up the fixes and assignments agreed in the review meeting and lists them for you.
Kimi's API: Anthropic-compatible, so Claude Code can use itOpenAI- and Anthropic-compatible. Priced in US dollars per 1M tokens
- Where to start: issue an API key on Kimi Open Platform (platform.kimi.ai). The provider is the Singapore entity.
- Compatibility: there are OpenAI-compatible and Anthropic-compatible endpoints. Point the OpenAI SDK, or tools that speak the Anthropic format, at Kimi instead.
- Using it from Claude Code: change Claude Code's endpoint (
ANTHROPIC_BASE_URL) and API key to Kimi's, and you can call Kimi's models from the Claude Code screen. The steps are in the official docs. - Pricing: as of October 2026, Kimi K3 costs $3.00 input and $15.00 output per 1M tokens (excluding tax). Cache reads, which reuse the same prompt prefix, are cheap at $0.30; cache writes are charged separately depending on how long they are kept. Check the official pricing page for current rates.
Getting started with the API
- Create a Kimi Open Platform account. Set up top-ups or a payment method in the console.
- Issue an API key. Keep it in an environment variable, never directly in code or shared documents.
- Point your usual tool at it. Pick the OpenAI or Anthropic format, then swap the URL and model name.
- Start small and watch usage. Long contexts use many tokens. Check daily usage before scaling up to production volume.
How SIMY relates: SIMY does not connect Kimi API keys at this time. SIMY's cloud runs use your ChatGPT (Codex) account.
Open weights and the license: K3 comes with conditionsDon't lump it in with "open source"
Kimi K3's model weights are published on Hugging Face, but under a custom "Kimi K3 License." You can use, modify and redistribute them, with conditions that depend on the size of your business.
- Businesses offering the model: if you offer K3 as a service and your revenue over the past 12 months exceeds $20 million, you need a separate commercial agreement with Moonshot AI.
- Large products: products with more than 100 million monthly users or more than $20 million in monthly revenue must display "Kimi K3" in the interface.
- K2.6 and earlier: Modified MIT license, with looser terms than K3.
For most companies, these terms are unlikely to be a big constraint for internal use. If you build it into a product, read the full license text.
Can you run Kimi locally? On Ollama, it's a cloud tag"Available on Ollama" doesn't always mean "runs on your PC"
As of October 2026, kimi-k3, kimi-k2.7-code and kimi-k2.6 in Ollama's official library all carry the "cloud" tag. That means they run in Ollama's cloud, not on your machine. Your input is sent to Ollama's servers.
- Running it on your own servers: you can run the published weights with vLLM or SGLang. K3 is a 2.8-trillion-parameter model meant for servers with multiple high-end GPUs.
- Running it on a personal PC: Kimi's latest models aren't realistic. For a Chinese model that runs locally, look at small and mid-size Qwen models.
If the goal is "keep data in-house," cloud-tagged models don't qualify. Run the weights on your own servers, or choose a model small enough to run on your machine.
Deciding for company use: where data is stored and trainingWhere your data goes depends on how you use it
Scroll the table sideways →
| How you use it | Operator / provider | Where data is stored | Used for training |
|---|---|---|---|
| Kimi chat (per the official help in Chinese) | Beijing entity | Servers in China | May be anonymized and used for training. Can be stopped by asking support (5–7 business days) |
| API (Kimi Open Platform) | MOONSHOT AI PTE. LTD. (Singapore) | Servers in Singapore | States that input may be used for "model optimization." Check the policy text for details |
| Running open weights yourself | Your company | Only inside your own servers | Never leaves |
- Check internal rules first: some companies have generative AI rules that set which services you can use and what you may enter.
- Using the chat from outside China: check which terms apply and where data is stored on the screens after you log in and in the help pages. The table above summarizes the official explanations as written.
- Default to keeping secrets out: with any chat AI, think before entering customer data or unpublished figures.
- Check client requirements too: some clients set contract terms about which countries data may be stored in.
Data handling is based on Kimi's data usage explanation and the Kimi Open Platform privacy policy as of October 1, 2026. How SIMY handles data is published on our Security page and in our Privacy Policy.
Kimi in English and other languages: how far it goesYou can ask in your language, but no supported languages are officially listed
You can talk to Kimi chat in English and other languages and get answers back in the same language. However, as of October 2026, the K3 model card doesn't list supported languages. Quality in any given language isn't officially guaranteed, so test it on real work documents.
- Easy to try: summarizing documents in your language, summarizing Chinese documents in your language, asking for code explanations.
- Be careful with: subtle tone and register, local proper nouns, and statements about laws or business practices. Check them before you use them.
To start small, give it a report you wrote yourself and ask for a summary and edits to the wording. Because you know the original, it's easy to judge the quality.
You can split the work: Kimi reads the long document, and SIMY handles the owners and deadlines decided in the meeting and the update you share with your team.
How it compares with ChatGPT, Claude and other Chinese LLMsNot better or worse. Each is strong in different places
Scroll the table sideways →
| Item | Kimi | Qwen | MiniMax | DeepSeek | SIMY |
|---|---|---|---|---|---|
| Developer | Moonshot AI (Beijing) | Alibaba Group (Hangzhou) | MiniMax (Shanghai) | DeepSeek (Hangzhou) | AwakApp Inc. |
| Strengths | Long context, coding | Many languages, local use | Cheap API, video and audio | Reasoning and a cheap API | Moving the work after the answer |
| Own coding CLI | Kimi Code | Qwen Code | MiniMax Code | None | Can use Claude Code on your PC to do the work |
| Starting point | You ask | You ask | You ask | You ask | SIMY picks it up from meetings and chat |
How to choose
- Reading long documents in one go: Kimi K3. It's on the pricier side among Chinese models, close to top international models.
- Keeping API costs down: DeepSeek or MiniMax.
- Running on your own PC: small Qwen or GLM models.
Using them together: any AI, Kimi included, can produce the answer or the draft. If you chase the owners, deadlines, requests and updates yourself every time, that's where SIMY comes in. See our comparison of Chinese LLMs for the differences, and the Doubao guide for ByteDance's models.
Frequently asked questions
What is Kimi?
Kimi is a chat AI and a series of large language models (LLMs) built by Moonshot AI in China. You can use it three ways: chat on kimi.com or the app, the API, or the published model weights.
Which country is Kimi from?
China. It is built by Moonshot AI in Beijing. Kimi for China is run by the Beijing entity, and the international API is provided by MOONSHOT AI PTE. LTD., a Singapore entity.
What is Kimi K3?
Kimi K3 is Moonshot AI's latest generation of models, released in July 2026 (as of October 2026). It handles a 1M-token context, accepts images and video as well as text, and its weights are published under the custom Kimi K3 License.
How is Kimi K3 different from Kimi K2?
K3 is the latest flagship, with long context and image and video input, published under the custom Kimi K3 License. The K2 series is the previous generation, and the coding model k2.7-code is priced lower than K3 on the API.
Is Kimi free?
Chat on kimi.com and the app is free to start. Paid memberships raise your limits, and the API is pay-as-you-go.
What is Kimi Code?
It is a coding agent that runs in the terminal, published by Moonshot AI under the MIT license. You log in with a Kimi Code account or an API key, and members can use it within their plan's allowance.
Can I use Kimi's models from Claude Code?
Yes. Kimi's API has an Anthropic-compatible endpoint, and the official docs explain how to swap Claude Code's endpoint and API key.
Can I run Kimi locally on my own PC?
Running the latest Kimi K3 on a personal PC isn't realistic. Kimi in Ollama's official library carries the cloud tag and runs in Ollama's cloud. To run it in-house, operate the published weights on high-end GPU servers.
Where is what I enter into Kimi stored?
According to the official explanations, Kimi chat (per the help in Chinese) stores data on servers in China, and the international API stores data on servers in Singapore. If you handle confidential information, check your internal rules and the full text of each policy.
Does Kimi work in English?
You can ask in English and get answers in English. However, the official model card doesn't list supported languages, so check the quality on your work documents before relying on it.
Does SIMY run on Kimi?
No. SIMY does not connect to Kimi at this time; it runs on your ChatGPT (Codex) account and on Claude Code on your PC. You can bring the key points and code you made with Kimi straight into your meetings and chats.
Do I have to ask SIMY every time for the work after Kimi?
No. SIMY picks up decisions, owners and deadlines from your meeting records and chats on its own. It learns your way of working from your conversations and returns only what needs your decision.
Sources
We checked Kimi's models, pricing structure, license and data handling against the official sources below (checked October 1, 2026). Model names and specs change often, so see each official page for the latest.