What it does
7 features
Kimi · Chat assistant with agent mode and coding agent suite
Product and access
This profile covers two parts of Kimi. According to the help centre, OK Computer was the first agent mode (launched 26 September 2025); the help centre now speaks of Kimi Agent. According to the product comparison, Kimi Code is the coding suite for developers. The comparison lists Kimi Work (local desktop agent) and Kimi Claw (cloud automation) as separate products. The older Python repository kimi-cli is archived and points to Kimi Code CLI.
Core features
7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.
OK Computer agent mode
The agent plans tasks from research to delivery and calls tools repeatedly. Entry is through kimi.com/agent or the K3 model in the app.
Source 3Documents, analyses and web pages
According to the help centre, the agent produces code projects, data analyses with charts, office documents such as Word, PDF and PPT, and deployed web pages, among other outputs.
Source 3Agent Swarm
According to the help centre, Agent Swarm can coordinate up to 300 sub-agents and support more than 4,000 tool calls.
Source 3Kimi Code CLI
Terminal agent for software tasks with plan mode, sessions and confirmation before file changes and shell commands. You sign in with a Kimi Code login or a Kimi Platform API key.
Source 6Subagents, skills and hooks
According to the README there are coder, explore and plan subagents in isolated contexts, skills and MCP servers from a marketplace or GitHub, and hooks that run local commands at defined points.
Source 7Editor integration through ACP
With kimi acp, Kimi Code can be used in Zed and JetBrains according to the README.
Source 7
Show all 7 features
Several Kimi products
The product comparison separates Kimi (web and app), Kimi Work (local desktop agent), Kimi Code (CLI and VS Code extension) and Kimi Claw (cloud automation).
Source 2
Usable AI models
Open choice
In web and app you use Kimi models, with K3 depending on your plan. In the Kimi Code CLI you can also add Anthropic, OpenAI and Google with your own key, or an OpenAI-compatible endpoint, for example for DeepSeek or Qwen. The docs do not cover local model servers.
Model families named
- Moonshot (Kimi)
- Anthropic (Claude)
- OpenAI (GPT)
- Google (Gemini)
- DeepSeek
- Alibaba (Qwen)
- Other models can be connected freely
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Yes
- Local models
- Not stated
- Depends on
- Plan
The rating covers the generally available main access. Documented · Sources accessed on
Kimi Code CLI: Providers and models · Kimi Code CLI: Getting started · Kimi: Agent overview · Kimi Code: Benefits
Does it fit my task?
2 practical tasks · 5 requirements and limits
A good fit if …
- you want to hand research and documents to agent mode as a task and collect them later
- you want a coding agent in the terminal with an MIT-licensed CLI and confirmation before changes
- you want to use Kimi models and accept the provider's data position for your data
Less suitable if …
- you must have data processed exclusively within the EU
- you rely on a free plan or prices in US dollars
- you need scheduled runs in agent mode, which the reviewed pages do not document
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Research as a presentation
- Your input
- A topic and wanted sources in agent mode on kimi.com/agent.
- Possible result
- The agent plans and researches and, according to the help centre, delivers results as PPT, Word or PDF, among other formats.
- What to check
- Check figures and sources against the originals; a run usually takes 5 to 20 minutes.
Code change with Kimi Code CLI
- Your input
- A repository in the terminal and a limited task, in plan mode first if needed.
- Possible result
- Changes to files and shell commands only after your confirmation.
- What to check
- Check the diff and tests before you accept.
Give a research job to agent mode and receive the result as a document or presentation.
Your possible prompt
Compare three providers for [task]. Use their official documentation. Show differences, sources, and open questions in a table.
Adjust the task, data and permissions.
Requirements and limits
Requirement
Plan for Kimi Code
Kimi Code is usable from Plus on the new plans and from Andante on the older ones. The Go plan has no coding quota.
Source 10Limitation
Context, duration and file output
The agent has a 128K-token context window, single tasks usually take 5 to 20 minutes, and in the general agent scenario usually only one file is output at a time; multiple files need Agent Swarm.
Source 4Limitation
Data storage in China and training
The privacy policy for users in mainland China stores personal data in the People’s Republic of China and does not provide it abroad without separate consent. Strictly de-identified inputs and outputs are used to optimise models; opting out goes through customer service.
Source 13Limitation
Older kimi-cli archived
The Python repository kimi-cli is no longer maintained according to its page and points to Kimi Code CLI. Existing installations are no longer supported.
Source 8Not yet confirmed
US dollar prices and plan for agent mode
The membership page names five China plans with yuan prices and no free offer; agent credits are a shared monthly pool according to the page. US dollar prices and offers outside China could not be confirmed.
Source 9
Can I use it?
Availability still unclear
Europe & regional availability
Help centre and two privacy policies reviewed. No page names a country list or availability in Europe.
Reviewed scope: Kimi (web, app, agent) and Kimi Code; different privacy policies apply by region
Reviewed on:
Availability still unclear
Europe: no availability statement
No reviewed page names availability or a plan offer in Europe. The privacy policy on kimi.com applies to users in mainland China according to the page; it refers users outside that region to kimi.ai, which was not reviewed.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: The reviewed sources name no country release for this.
Official source · www.kimi.com · Official source · www.kimi.com
Availability still unclear
Data processing and storage
Under the policy for mainland China, personal data is stored in the People's Republic of China; the help centre names storage on servers in mainland China. The Moonshot AI PTE. LTD. policy (Kimi Claw and Kimi) instead says data may be stored outside the country of residence and names no location. No reviewed page says where data is processed.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland
Documented reason: The pages differ by company and region; which one applies to you is open.
Official source · www.kimi.com · Official source · www.kimi.com · Official source · www.kimi.com
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Kimi account with membership; Kimi Code also with an API key
Sources reviewed:
Show all 3 points
Documented
Account used
Kimi Code CLI signs in with a Kimi Code login (device code) or a Kimi Platform API key; /logout clears the credentials. Other providers are entered in ~/.kimi-code/config.toml.
Evidence
Memory and conversation continuity
0 of 6 criteria documented
Kimi Code CLI stores configuration, sessions and logs locally. The reviewed pages describe no memory across sessions.
Kimi Code CLI locally; Kimi web and app with an account
Sources accessed:
Documentation by criterion
StoragePartly documented
Storage and scope
The CLI stores configuration, session records, logs and the update cache under ~/.kimi-code/ by default; KIMI_CODE_HOME changes the location. The help centre generally names storage on servers in mainland China, without naming a domain or product.
SelectionNot publicly established
Selection and timing
Not described: what the agent keeps across sessions or when.
Edit & deletePartly documented
Inspect, edit and delete
With /sessions you browse and resume sessions. According to the privacy policy, you delete chat history with the delete-conversation button in the history and the account in settings.
Access boundariesNot publicly established
Agent and channel boundaries
Not described: how Kimi Code, agent mode and chat share memory. The privacy policy names web, app, Kimi Work and Kimi Code as covered products.
RetrievalPartly documented
Retrieval
Sessions can be resumed with /sessions and copied with /fork. No automatic search across earlier sessions is described.
New conversationsPartly documented
Session reset and compaction
/compact compresses the context manually to free tokens. Agent mode has a 128K-token context window; the pages say nothing on resetting.
Cost and effort
5 plans · Prices as of 5 November 2026
Prices and included usage
Pricing checked
The membership help page names the China plans Go, Plus, Pro, Max and Ultra at ¥49, ¥99, ¥199, ¥699 and ¥1,399 per month. It names no free plan. Prices in US dollars could not be confirmed.
| Plan | Price | Included |
|---|---|---|
| Go | ¥49 per month (China) | Agent credits at base level, one parallel agent task, no Agent Swarm. |
| Plus | ¥99 per month (China) | Double agent credits, two parallel tasks, Agent Swarm. |
| Pro | ¥199 per month (China) | Four times the agent credits, two parallel tasks, Agent Swarm. |
| Max | ¥699 per month (China) | 14 times the agent credits, four parallel tasks, Agent Swarm. |
| Ultra | ¥1,399 per month (China) | 28 times the agent credits, four parallel tasks, Agent Swarm. |
- The page explicitly names these as the current China plans. Prices outside China were not confirmed.
- Kimi Code access depends on the plan (from Plus, or Andante on older plans); credits are a shared monthly pool according to the help page.
Setup
For agent mode, open kimi.com/agent or choose the K3 or K3 Swarm model in the app. You install Kimi Code CLI by script and sign in with /login.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Agent mode in the cloud and Kimi Code CLI locally
Sources reviewed:
Editorial assessment
The CLI asks before risky actions and stores locally. For agent mode, the context limit, duration and the privacy position by region matter; before use, check which company and policy apply to you.
Show all 6 points
Documented
Files and working state
The CLI stores sessions locally under ~/.kimi-code/. The reviewed pages do not say how long cloud agent runs are kept.
EvidenceDocumented
Sign-ins
Sign-in by device code or API key; /logout clears the credentials. The page names no time limits.
EvidenceDocumented with conditions
Starting further runs
Agent tasks usually take 5 to 20 minutes; you can leave the page, the task continues in the background and the system notifies you when it is done. Schedules belong to separate pages and products that were not reviewed.
EvidenceDocumented
Human intervention
The CLI asks before file changes and shell commands; read-only operations run automatically by default. Shift-Tab toggles Plan mode.
EvidenceOpen in reviewed sources
Detecting errors
The reviewed pages say nothing on error handling and retries. Agent mode can be interrupted with Stop output.
EvidenceDocumented with conditions
Resuming after failures
You resume CLI sessions with /sessions or copy them with /fork. For cloud agent runs the pages describe no resuming.
Evidence
Evidence and open questions
14 sources · 3 open points
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Europe: Availability still unclear
Requirements and limits: not yet confirmed
Sources and evidence
Provider help centre, product pages, privacy policies and repositories reviewed. Some pricing pages could not be read, so US dollar prices are missing. Example tasks are editorial; hands-on tests are pending.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
The privacy policy for mainland China names Beijing Moonshot AI as operator; another policy names Moonshot AI PTE. LTD.
- Operator
- Moonshot AI
- Organization
- Provider
- Website
- www.kimi.com
- Repository
- MoonshotAI/kimi-code
- GitHub stars
- 8.143 · Reviewed
- License
- MIT
Changes to this profile 1
Kimi from Moonshot AI with the OK Computer agent mode for research and documents, plus Kimi Code as a coding agent for terminal, VS Code and desktop.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.