Kimi Work is the agent workspace inside Kimi’s desktop app. It can read and create local files, use browser tools, and carry out tasks that produce documents, spreadsheets, presentations, or other deliverables. You describe the outcome in ordinary language and review the work it produces.
Last verified: October 12, 2026.
The useful starting point is a small job with clear inputs and an output you can check. This guide covers the official download, supported computers, free access and paid plans, permission settings, and a practice workflow that turns meeting notes into an action log.
What can you do with Kimi Work?
According to the official Kimi Work overview, Work brings an AI agent into your local workspace. Typical jobs include organizing documents, analyzing files, combining research, and preparing reports. It is designed for knowledge workers who want to delegate a task without operating a terminal.
For example, an analyst might ask it to turn a folder of spreadsheets into a summary, while an operations manager might extract follow-up actions from meeting notes. Define the source files, the requested deliverable, and what must be checked before accepting it.
Kimi’s products overlap in their AI capabilities. The official product comparison makes the execution environment the useful distinction:
| Your main task | Where to start |
|---|---|
| Work directly with folders and files on your computer | Kimi Work in the desktop app |
| Ask questions, research, write, or generate content from online and uploaded material | Kimi on the web or mobile app |
| Develop software mainly in a terminal or code editor | Kimi Code |
The desktop app also includes Chat mode for quick conversations. The web version has Agent features too, so choose Work when direct access to the desktop workspace matters to the job.
Download Kimi Work: Windows and Mac requirements
Use the official Kimi download page and choose the installer for your computer. Check Work support separately from whether the general desktop app can be installed.
| Computer | Kimi Work support |
|---|---|
| Windows PC | Windows 10 or later |
| Mac with Apple silicon | macOS 12 or later |
| Intel-based Mac | The desktop app supports Kimi Chat only; Work mode is unavailable |
The operating-system requirements come from the Work overview; the download FAQ explains the Intel Mac restriction. A Linux Work installer is not listed on the official download page.
If you also need ordinary chat or mobile access, our Kimi app download guide explains the separate routes and device checks.
Install the app, sign in with your Kimi account, and switch to Work in the sidebar. Basic use follows the official getting-started workflow: connect a project folder, choose permissions, and describe the task. You do not need an API key for this desktop workflow.
Is Kimi Work free? Membership and credit costs
You can download Kimi Work and try it without a paid plan. Kimi’s official introduction says new users receive limited free credits for initial trials. It does not promise a fixed number of completed tasks or specify a recurring free-credit allowance. Check the allowance shown in your account before starting a large job.
For paid use, the international membership pricing lists the following new plans. These are Kimi memberships that cover eligible features, including Work.
| International plan | Monthly billing | Annual billing, paid upfront |
|---|---|---|
| Plus | US$19 per month | US$180 per year |
| Pro | US$39 per month | US$372 per year |
| Max | US$99 per month | US$948 per year |
| Ultra | US$199 per month | US$1,908 per year |
The annual totals correspond to US$15, US$31, US$79, and US$159 per month respectively, but require annual payment. China-region memberships and existing legacy subscriptions can have different names and terms; confirm the offer available to your account.
The international plan overview includes Agent Swarm in all four listed paid plans. Goal Mode starts with Pro. Higher tiers add capacity, but the best starting plan depends on the features you need and the usage you actually consume.
Under Kimi’s credit rules, Work shares a credit pool with other membership features. Consumption depends on the model, the amount of input and output, and tool use. Paid-plan credits refresh monthly, including for annual subscriptions; unused credits do not roll over.
A subscription therefore does not buy a guaranteed number of identical Work tasks. Review your usage after a representative task before relying on the same allowance for recurring work. API billing and third-party API services are separate from this membership workflow.
Choose permissions before connecting your files
Work can change files and run code, so choose its permission setting deliberately. The official Work FAQ describes three modes:
- Default: routine operations run automatically; sensitive operations, such as changing files or running code on them, require authorization.
- Manual approval: asks for authorization before acting.
- Fully automatic: proceeds without asking for further authorization.
For a first exercise, use Manual approval with copies of non-sensitive files. Read each requested action before approving it. A dedicated project folder helps organize the task, but should not be treated as a proven security boundary that prevents every other kind of access.
Does “local” mean offline or private by default?
“Local” describes Work’s access to files and tools on your computer. It does not establish that the AI model runs entirely offline or that every piece of task content stays on your device.
Kimi’s privacy policy covers user content such as prompts and files and describes processing for service delivery and improvement, including model training and optimization, subject to the applicable legal basis. Connected third-party services have their own policies.
The public documentation does not fully map which content from each local operation is sent for processing. For confidential work, check the terms applicable to your account and your organization’s rules before connecting files or services.
Before connecting work folders, review the Kimi safety and privacy checks for the service region and product you use.
How to use Kimi Work: a first project you can check
Use the fictional files below to practice turning a project brief and meeting notes into an action log and a short summary. This is an illustrative workflow, not a recorded Kimi Work test. The checks show what a correct result should preserve.
1. Prepare two small input files
Create a folder named Kimi-Work-Practice. In a plain-text editor, save the following as project-brief.txt:
Project: Harbor Studio website refresh Approved launch target: 2026-10-16 Required deliverables: - Revised homepage copy - A working contact form - An accessibility review A proposed launch change requires approval.
Save this second file in the same folder as meeting-notes.txt:
Planning meeting: 2026-10-12 Maya will send three homepage headline options by 2026-10-14. Jon will test the contact form by 2026-10-15. An accessibility review is needed. No owner or deadline was agreed. A participant proposed launching on 2026-10-19. The proposed date was not approved. No task progress updates were recorded.
Keep these exact filenames for the exercise. The example dates are part of the fictional source material. The task is to preserve their meaning, including the difference between an approved target and a proposal.
2. Connect the folder and start a task
- Open the desktop app and select Work.
- Create a project linked to
Kimi-Work-Practice, following the app’s folder-selection flow. - Choose Manual approval and start a new task in that project.
- Send the prompt below. Review any requested actions as they appear.
A project connects the task to a local folder. For this exercise, use a regular task; the small source set does not need parallel agents or an extended Goal session.
Read project-brief.txt and meeting-notes.txt in this workspace. Create output/action-log.csv with these columns: task,owner,due_date,status Include the three work items recorded in the meeting notes. Use YYYY-MM-DD dates. Leave an unknown owner or deadline blank. Set every status to "Not reported" because the notes contain no progress updates. Do not turn the proposed launch date into an action or an approved deadline. Create output/summary.md with: 1. The approved launch target. 2. The proposed launch change and its approval status. 3. The recorded actions. 4. Open questions about missing information. Use only the two input files. Do not browse, contact anyone, invent information, or modify the inputs. Create the output folder if needed and save both deliverables as actual files. State their paths and any unresolved questions. If either file cannot be saved, explain the obstacle.
The prompt defines the inputs, the output names, the CSV columns, and how missing information should be handled. It also separates producing an action log from contacting the people named in it.
3. Inspect the saved deliverables
Open output/action-log.csv in a spreadsheet app and output/summary.md in a text editor or Markdown viewer. Check the files themselves, including whether the CSV values appear in separate columns.
- Three action rows: headline options, contact-form testing, and the accessibility review.
- Recorded assignments: Maya has the headline task due on October 14; Jon has the form test due on October 15.
- Missing information stays missing: the accessibility review has no owner or deadline.
- No invented progress: all three statuses say
Not reported. - Approval is preserved: October 16 remains the approved launch target; October 19 remains an unapproved proposal.
- Originals are preserved: the two input files still contain the text you supplied.
These checks are derived from the fictional inputs, not from measured product performance. A well-formatted document can still contain an invented owner, an incorrect date, or a proposal presented as a decision.
4. Request a separate revision if needed
If the output fails a check, identify the discrepancy and ask for a separate revision so you can compare both versions:
Recheck output/action-log.csv and output/summary.md against project-brief.txt and meeting-notes.txt. Verify every action, owner, deadline, progress status, and approval claim against the source text. The approved launch target is 2026-10-16. The date 2026-10-19 is an unapproved proposal. Leave missing owners and deadlines blank. Keep progress statuses as "Not reported". Correct any unsupported claims. Save corrected copies as output/action-log-revised.csv and output/summary-revised.md. Preserve the input files and first outputs. Explain the discrepancies you corrected and identify anything still unresolved.
For your own work, keep the same four ingredients: named sources, a specific deliverable, limits on changes, and checks you can perform independently.
When to add browser automation, Swarm or Goal mode
Once a small task works, add capabilities that solve a specific problem in the next job. Each extra data source or execution mode adds something else to configure and verify.
Use browser tools and plugins for external information
Kimi Browser Extension, formerly WebBridge, lets an agent open pages, click, fill forms, and extract information. The official setup provides Chrome and Edge installations and a companion capability installed through Kimi Work’s plugin marketplace.
A useful first browser assignment is to collect facts from a short list of public pages and save the source URLs beside the findings. Specify which sites to use and whether the task is limited to reading. Review any proposed submission or account change separately.
The Plugin Center also connects external apps and specialist data sources. Install the relevant plugin, complete any required account authorization, and select it through the slash menu when needed. Availability in the catalog does not establish that every connected service is free or already authorized.
Browser control and desktop-app control are separate capabilities. Kimi lists Computer Use as a plugin for interacting with desktop applications; installing a browser extension alone does not establish that broader setup.
Use Swarm for divisible work and Goal mode for iteration
Agent Swarm divides a larger assignment among cooperating agents. It is useful when parts can be tackled independently, such as reviewing separate sets of documents before combining their findings. The number of agents inside a Swarm is different from your plan’s limit on simultaneous tasks.
For a small file exercise, begin with an ordinary task. Move to Swarm when the assignment has useful independent parts, then check that the combined result uses consistent definitions and does not duplicate findings. See the Work capabilities and your plan’s included features.
Goal mode keeps iterating toward a defined outcome. Kimi’s Goal documentation describes tasks with a clear goal, an uncertain path, multiple attempts, and verifiable results. It can suit a document-conversion job that needs several rounds of checking and correction.
To start, switch to Work, open the + menu beside the input box, and choose Goal. State the deliverable, the checks required for completion, and the conditions that should stop the task or bring a question back to you. You can interrupt or redirect it while it runs.
Keep repeatable work in Skills, Widgets and Dashboard
The getting-started guide explains how project instructions in AGENTS.md can carry requirements across tasks. Use this for rules that should remain consistent, such as preserving source files or marking missing information. Skills can package reusable methods for recurring work.
Widgets present interactive content inside a conversation. A Dashboard keeps selected widgets together beyond that conversation. For example, you could turn a reviewed action log into a project view. Check its connected source and refresh mechanism before treating displayed information as current.
Can Kimi Work keep running when you are away?
Desktop scheduled tasks need the computer awake and the Kimi Work app open. The Work FAQ says triggers missed while the computer sleeps, shuts down, or the app is closed are not replayed automatically when it opens again.
A schedule sets when a task starts; Goal mode governs continued work toward an outcome. Neither should be treated as a promise that a desktop task will keep running after its computer becomes unavailable.
You can create a recurring task through the scheduling interface or describe the schedule in conversation. Specify the time, frequency, inputs, output location, and what to do with an existing report. Confirm the saved schedule and run it once before relying on it. Kimi’s scheduling help distinguishes local Work tasks from cloud tasks, which do not require an open desktop client.
Can you control it from your phone?
Yes. Kimi’s Remote Control guide documents starting and monitoring desktop tasks from the mobile app. Enable phone control on the computer, sign in to the same account on your phone, and approve the device connection. The computer must remain powered on and online.
This is remote access to desktop execution. Kimi’s download FAQ says Work conversations remain on their originating desktop, while Chat conversations sync across devices. Plan around the computer that holds the files and project.
Troubleshooting Kimi Work
Start with the smallest check that explains the problem. Preserve the inputs and any useful output before repeating a task.
| Problem | What to check or do |
|---|---|
| Work mode is missing | Confirm you installed the desktop client for a Work-supported device. Intel Mac Chat availability does not include Work. |
| A source file cannot be found | Check the project folder, exact filename, and requested permissions. Make sure the file is actually present in the selected location. |
| A browser or app plugin cannot work | Check installation and connection status, the signed-in account, and the authorization required by that service. |
| New tasks are blocked by credits | Check the shared credit balance and its next refresh date. Usage in another Kimi feature can affect Work’s remaining allowance. |
| A task is interrupted | Check whether the app is running and the connection is available. Inspect saved files before retrying so you can avoid duplicate or conflicting outputs. |
| The saved result is wrong | Identify the unsupported claim or failed check. Request a separate corrected version and compare it with the original sources. |
| A scheduled report is missing | Check the task’s enabled state and run history. If a local trigger was missed, use the scheduling controls to run the task when the computer and app are available. |
For a recurring application problem, record the app version, operating system, task steps, and any displayed error. Compare the behavior with the official release notes before attempting the same large task again. Kimi still describes Work as beta, so version-specific fixes can matter.
For sign-in, loading, or recurring app failures beyond a single Work task, follow the broader Kimi troubleshooting steps before resetting or reinstalling.