Kimi Work: The Ultimate Desktop AI Agent for Knowledge Work

Meet Kimi Work, a local AI agent that lives on your desktop with the ability to run 300 agents in parallel, operate your browser, and turn live data into finished reports. It's the AI desktop app built for real knowledge work, on both Mac and Windows.

11 min read2026-06-17
Kimi Work, the all-in-one desktop AI for your knowledge work

Most everyday work is not a single question with a single answer. It may include a chain of steps: gathering files, checking websites, cleaning data, and formatting the result. Ordinary chatbots stop at the answer and leave every step that follows to a person. Kimi Work closes that gap: Rather than living in a chat window, Kimi Work reads local files, operates the browser, runs code, and carries out scheduled tasks, then returns the finished deliverable instead of merely advice on how to produce it or simply generating a single file.

What is Kimi Work

Kimi Work is an AI desktop software built for knowledge workers, from developers to finance and research professionals. It can call on up to 300 agents at once and fuses the wider capabilities of Kimi into a single all-in-one application: it draws together the reasoning and long-document strength of the web product, the automated web operation of Kimi WebBridge, and the file-level and command-level execution of Kimi Code.

The tasks Kimi Desktop can handle range from a quick single task to a goal that runs continuously over hours. And rather than returning a one-shot answer, Kimi Work operates as a workflow loop: it assigns and executes sub-tasks across its agent swarm, verifies the results, and feeds them back into the next round of work. The entire process remains observable and open to human intervention.

The people Kimi Desktop serves span both technical and non-technical work. For developers, it handles coding tasks directly, from writing and reviewing code to running commands and managing local projects. Beyond engineering, it reaches into investment research, consulting, academia, university teaching, market research, and content operations, as well as any office professional who routinely works through web pages, local files, spreadsheets, and reports.

Unique Kimi Work features that empower your work

Kimi Work consolidates capabilities that are usually spread across several separate tools into one AI desktop application, and the features below define how that work actually gets done.

Goal Mode for auto-running objectives

Kimi Work’s Goal Mode shifts the AI from “completing a single step for you” to “continuously working toward a defined goal.” You specify the objective, acceptance criteria, and constraints, and Kimi Work keeps running tasks, gathering information, modifying code, producing reports, and verifying results while you sleep, attend meetings, or commute. You don't need to restart the conversation when you return. You can review progress, check the deliverables, adjust the direction if needed, and let it continue into the next round.

300-agent swarm supported parallel execution and multi-format output

To carry out a task, Kimi Work can launch a swarm of up to 300 agents from a single instruction, splitting one large task into hundreds of sub-tasks that run in parallel, with each agent handling a different part of the work before the separate results are assembled into a single coherent output. The finished work is not limited to plain text: when a task calls for it, Kimi Work can deliver polished, ready-to-share files in multiple formats. From code files, PDFs, PowerPoint decks, or interactive reports to Excel spreadsheets, Word documents, or even websites.

The desktop AI agent enables 300 agent supported parallel execution and multi-format output

Plugins that extend professional capabilities via simple requests

Plugins extend Kimi with external capabilities, built from several MCPs connected through skills. Once installed, Kimi Work can call it directly in conversation to query data, operate third-party services, and read and write documents. Through MCP, OAuth, or open-platform capabilities, users can connect third-party accounts they own or are authorized to use, keeping access boundaries clear and under control.

With Plugins, a single request retrieves real-time financial, corporate, and academic data, replacing manual searching and copy-paste. Office tools can be operated directly, turning repetitive document and collaboration tasks into one instruction. The coding plugin covers the full development cycle, handling code review, deployment, and database migration in conversation. Design work can be automated too: from posters and social images to slide pages, Kimi generates and edits design files directly, turning copy into publishable assets without the usual template hunting, layout, and resizing.

Native access to finance & academic databases

Kimi Work connects directly to professional data sources, so the figures behind an analysis are available without a separate subscription or any manual export. The built-in coverage spans global equities, futures, and indices for market and securities work, extends to World Bank macroeconomic data on national GDP, population, employment, and trade for wider context, and reaches into a deep academic layer of journals, papers, preprints, dissertations, and patents for research, with native access to academic databases.

With easy access to professional databases, Kimi Work can gather real data. A single request from users can move from raw figures all the way to a finished, data-driven analysis without ever leaving the app.

Kimi Desktop has native access to finance & academic databases

Browser automation supported by Kimi WebBridge

Supported by the browser extension Kimi WebBridge, Kimi Work operates a signed-in browser, navigating pages, clicking elements, completing forms, downloading files, and reading on-screen content much as a person would. Because it relies on existing sessions, repeated logins are unnecessary, and tasks that once required manual clicking can be handed off to the agent and carried through to completion.

Kimi Desktop can control browsers, supported by Kimi WebBridge

Round-the-clock scheduled task automation

Work in Kimi Work does not have to be started by hand, since tasks can be scheduled to run in the background at set times, whether that means a morning briefing, an end-of-day summary, or an overnight data job. The application keeps running while the machine is left unattended, so completed results are waiting when the next session begins.

Kimi Desktop enables round-the-clock scheduled task automation

From instruction to local file, in a single step

Once a local folder is mounted, Kimi Work writes its results straight back into it, carrying a request all the way from instruction to finished file in one pass, so nothing has to be downloaded, renamed, or swapped back into place by hand. That direct access to local files also drives one-click organization of an entire folder, sorting and tidying files in bulk so a cluttered directory is brought into order in a single step.

Combining the analysis of local documents, open web searches, professional datasets, and multiple agent skills, Kimi Work streamlines the traditionally lengthy workflow into one-click action, without sacrificing output quality.

Kimi Work vs Kimi (Web): What's the difference?

Both products are Kimi, and they are designed to complement each other. Kimi (the Web version) is the fastest route to ask, research, and generate from any device. Kimi Work is the desktop AI agent for deep, hands-on workflows that reach into local files and the browser to reduce manual work.

DimensionKimi (Web)Kimi Work (Desktop)
Form factorOpens in a browser, no install, any deviceNative Kimi Desktop application for macOS and Windows
Local file & system accessRequire manual uploading; no direct access to local files or the operating system.Able to read and write local files, run commands, and operate on local folders, spreadsheets, and reports.
Execution patternAgentic, multi-step execution within its cloud and browser environment, strong at research, reasoning, and tool use over online and uploaded content.The same agentic execution is extended to the local environment, its Goal Mode and Scheduled Tasks function, adding direct action on files, code, and system-level tasks.
Web operationSearches and reads the web.Automated web operation through Kimi WebBridge, acting on pages directly.
Coding capabilityGenerated in the cloudRuns Python and shell locally and processes local files, with data kept on-device

Kimi (Web) is an equally strong agent within its own environment, while Kimi Desktop brings that same agent capability to your local files and system. Kimi Work fits when the work needs to reach into your local environment, running longer, end-to-end workflows that act directly on local files, the browser, and system-level tasks.

How to use Kimi Work

Setting up the AI desktop app takes only a few minutes. For Kimi Desktop users, running one-time tasks, scheduled tasks, or run tasks in Goal Mode is all easy to master.

Before starting: Download Kimi Work (macOS and Windows)

Start by downloading the Kimi Work desktop app. We provide the installer for both macOS and Windows. Simply click the button below, and the system will download the right version for you.

How to run unscheduled tasks in Kimi Work

Launch Kimi Work and switch to Work mode, the agent workspace where tasks are actually carried out.

Give the desktop AI agent Kimi Work instructions

A project can be created for the job and linked to a folder, whether an existing one or a newly made directory, which then becomes the place Kimi Work reads from, writes to, and stores its files; for lighter, one-off tasks, naming a project and choosing a folder can be skipped altogether.

Create a project at Kimi Work, the AI desktop software

The level of autonomy is set in the same step, since any action that touches permissions prompts Kimi Work to ask first. Choosing “Full access” lets the agent run from start to finish without interruption and keeps the whole task hands-off, while “Ask permission” pauses at each sensitive step and waits for approval before proceeding.

Set permission access before the task is assigned

After setting the folder and the permission level (both are optional), describe your tasks and let Kimi Work execute them for you.

Example prompt:

Read every PDF in this folder, extract the key information, and build a one-page summary deck.

Kimi Work will display its plan and execution progress, then deliver finished output such as decks, sheets, and reports as you required into the workspace, or directly into the local folder.

Review the output at Kimi Work, the local AI agent

How to use Goal Mode in Kimi Work

Goal Mode lets you hand Kimi a goal and have it keep working toward that goal on its own. Instead of answering a single turn, Kimi saves the goal as a persistent state and, at the end of each round, decides what to do next: whether the goal is complete, blocked, paused, or ready to continue into another round. As long as your computer stays on, a goal can keep running across many rounds without you starting over each time.

When a task is a good fit for Goal Mode

Goal Mode works best when the task meets these conditions, and the more of them that apply, the better the fit:

  • The goal is clear and well defined.

  • The path to reach it is uncertain and needs to be worked out.

  • It takes several rounds of attempts to get there.

  • The result can be verified against some evidence.

A short report you can finish in one reply does not need Goal Mode. A goal that requires research, iteration, and checking does.

How to set up a goal

  1. State the goal you want completed in your prompt.

  2. Call Goal Mode with /goal. This is optional, but stating it explicitly helps Kimi plan and run the task more effectively.

  3. Add your requirements, constraints, and what counts as a finished deliverable. Kimi will then plan and execute on its own, round after round.

A reliable goal formula usually includes: Result + acceptance evidence + scope constraints + budget limits + what to do on failure

Prompt template:

I want to accomplish a goal: [your goal]. Please use /goal mode to keep working on it until you produce an executable, deliverable result. Requirements: 1. Start by breaking the goal down for me. 2. Proactively ask me for any information you need. 3. Move forward step by step. 4. At the end of each round, give me the current progress and the next step. 5. Deliver a final, ready-to-use result. My background/constraints: [add your situation] Acceptance criteria: [what counts as done]

Exmaple prompt:

/goal I want to complete a thesis topic-selection task: find a workable topic around "the impact of AI on educational equity." Please keep working until you have: settled on a topic, organized the core arguments, listed directions for literature, and produced a thesis outline. First, ask me about my major, the word-count requirement, and my advisor's preferences, then continue.
Use Goal mode to run tasks at Kimi Work

In other words, describe the outcome you want, how success will be verified, what the task should and should not cover, any limits on time or resources, and how Kimi should report back if it gets stuck. The clearer these are, the more reliably the goal runs to completion.

How to schedule tasks at Kimi Work

Kimi Work allows users to set tasks to be carried out as scheduled, whether the scheduled tasks are for one time or periodic. For work that needs to run on a regular basis, select "Schedule Tasks" from the left sidebar and click the "Create" button in the upper right corner.

Click "Scheduled Tasks" from the left sidebar and click the "create" button at Kimi Work

A new scheduled task can be added in one of two ways:

  • Add manually: Fill in the task name, describing what the agent should do, and set the execution clock time and frequency (Once/Daily/Weekly/Monthly); Then the tasks are scheduled successfully.

    Fill in the name and description of the task and set the execution time
  • Create from chat: Simply give instructions in natural language, and the task will be converted into a scheduled task directly. Once saved, the task runs automatically at the set time without any further input.

    Send a task and let Kimi Work create the scheduled task directly

How to use Plugins in Kimi Work to boost efficiency

  1. Click Plugins on the left side of the Kimi Work interface to open the plugin marketplace.

    Click Plugins on the left side of the Kimi Work interface to open the plugin marketplace.
  2. Browse Featured or use Search plugins in the top right to find the plugin you need.

    Browse Featured or use Search plugins in the top right to find the plugin you need.
  3. Install the plugins you need. Some plugins require OAuth authorization; complete the authorization before continuing.

    Install the plugins you need in Kimi Desktop
  4. Once installed, start a new conversation and type / in the input box to see your added plugins. Select one to use, such as “/Notion”.

Note: Installing a plugin means Kimi can use it, not that it will be called automatically in every round.

Practical use cases where Kimi Work excels

Moving fluidly across local files, the live web, and professional data sources, Kimi Work suits a wide range of scenarios, from software development and coding projects to financial investment, academic research, strategy consulting, marketing, and business analysis.

Finance and investment

The built-in coverage of financial databases enables Kimi Work to handle the full span of an investment workflow without the need to switch between terminals, data providers, or document tools.

For example, you can point Kimi Work at a folder of local holdings data and instruct it to read incoming research reports, cross-reference them against current positions and stated risk preferences, and deliver a daily briefing that flags concentration risks and surfaces potential rebalancing considerations, turning what would otherwise be several hours of manual aggregation into a standing automated process.

Example prompt:

Read all the earnings reports and broker notes in my Q2 folder, pull the latest price-to-earnings and price-to-book ratios for each company, flag any that have moved more than 10% from their 90-day average, and generate a morning briefing deck with a rebalancing watchlist.

Academic research

The 300-agent swarm can fan out across a large body of literature simultaneously, extracting findings and mapping relationships at a pace that sequential reading cannot match, while the native academic data coverage reaches across journals, preprints, dissertations, and patents in a single pass.

A researcher preparing a systematic review on a defined topic could have Kimi Work scan several hundred papers at once from local folders and academic databases, pull the methodology, sample size, and key conclusions from each, group the results by research approach, identify where the existing evidence converges or remains contested, and write that synthesis into a structured draft section which directly stored locally, ready for the researcher to refine.

Example prompt:

Scan every paper in my Literature folder, extract the methodology, sample size, and main findings from each, group them by research approach, identify the most contested claims across the set, and write a structured literature review draft of around 2,000 words with a citation summary table at the end.

Coding and developing

Kimi Desktop works at the file and command level rather than just describing changes, so it can read a repository, edit source files, run builds and tests, and act on the results. Development is also exactly the kind of problem Goal Mode handles well: the objective is clear, the path is uncertain, it takes many rounds, and the result can be verified against a build or test suite. Kimi Work keeps iterating toward that target while staying observable, so a long, multi-step engineering task can reach a tested, working state with far less hands-on time, and the output is the actual updated codebase rather than a snippet in a chat window.

A developer needs to migrate a mid-sized service from an outdated framework to the current one. The work is well defined in its goal but uncertain in its path: dependencies have to be untangled, dozens of files updated, tests kept green, and the whole thing verified before it can ship. Done by hand, this is the kind of task that stretches across days of focused work.

Example prompt:

Migrate this service to the new framework. Use the existing test suite and a clean build as acceptance evidence. Do not change the public API. If a migration step can't pass tests, stop and report the failing case and what's blocking it.

Strategy and consulting

WebBridge and the 300-agent swarm together allow Kimi Work to sweep across a large number of companies in one run, gathering live signals from public web pages and combining them with offline documents, which compresses competitive research that would ordinarily take a team several days into a single session.

A consultant scoping entry into a new market could task Kimi Work with profiling thirty competitors at once, extracting positioning statements, recent strategic moves, and available financial indicators for each, then assembling the findings into a finished landscape presentation with a side-by-side competitor matrix, ready to share with the client directly.

Example prompt:

Research these 25 companies in the Southeast Asian fintech market, pull their latest positioning, funding history, key products, and recent strategic moves from their websites and public filings, then build a competitive landscape presentation with a side-by-side matrix ranked by market presence.

Marketing operation

Scheduled automation and WebBridge allow Kimi Work to gather performance signals across multiple channels on a set cadence, merge them with locally stored historical records, and produce a consolidated report without any manual aggregation in between.

A marketing team running campaigns across several platforms could configure Kimi Work to pull the week's metrics every Friday afternoon, compare them against the prior four weeks held in a local spreadsheet, calculate channel-level efficiency and period-over-period shifts, and deliver a finished review deck so that the Monday planning meeting begins with a complete picture rather than a set of raw exports still waiting to be processed.

Example prompt:

Pull the last four weekly campaign reports from my Marketing folder, compare click-through rates, cost per acquisition, and conversion by channel against the four-week average, highlight the top and bottom performers, and produce a review deck with recommended budget adjustments for next week.

Data analysis

Direct folder access and on-device Python execution mean that Kimi Work can reach into locally stored datasets, clean and transform them, run the analysis, and write the results back into the same directory.

An operations team sitting on two years of monthly sales records spread across dozens of Excel files could instruct Kimi Work to consolidate and clean the full set, compute year-over-year and month-on-month performance broken down by region, product line, and sales channel, generate the corresponding charts, and package everything into an executive summary deck, completing in one unattended run what would otherwise occupy most of a working day.

Example prompt:

Merge all the monthly sales files in my 2026 folder, remove duplicate and incomplete rows, calculate year-over-year and month-on-month growth broken down by region and product category, plot the key trends as charts, and package everything into an executive summary deck for the Monday operations review.

Who is Kimi Work for?

Kimi Work is built for knowledge workers whose day depends on files, the web, and data rather than code:

  • Finance professionals, including analysts, investors, and researchers who need live market data and report-ready output

  • Researchers and academics working through long papers and large literature sets

  • Developers and engineers handling coding tasks directly, from writing and reviewing code to running commands and managing local projects

  • Consultants and strategists whose output centers on competitive research and decks

  • Marketers and operators converting messy data into performance reports

  • Anyone replacing manual desktop work, such as file organization, browser tasks, spreadsheets, and recurring reports

Conclusion

Kimi Work can run hundreds of agents at once, read local files, operate the browser, execute complex objectives automatically with Goal mode, and turn live data into finished decks and reports from a single desktop app. When real work lives outside the chat box, this is the AI agent built to complete it. Now Kimi Work is available for Mac and Windows, and is ready to take on the first long task right away.

FAQ

How to choose between Kimi Work and the Web version?
It is not a matter of one being more capable than the other, but of where they execute. Kimi (Web) is well-suited for research, reasoning, and generation across web sources and uploaded content, on any device. Kimi Work suits longer workflows that reach into local files, the browser, and scheduled tasks, and is the better choice when an AI agent needs to execute work end to end in your local environment.
Can I use the chat function in the desktop version?
Yes. The desktop version includes both a chat mode and an agent workspace, switchable from the sidebar, so a quick conversation and a full task handoff are both possible within one app.
Is Kimi Work free to use?
Yes. Kimi Work is available to free users, and the AI desktop app can be downloaded and run without a paid plan. New users are provided with limited free credits for initial trials, and free plan users can also set scheduled tasks (up to 2 tasks). For further use, more advanced paid plans are available.
Does Kimi Work run on both Mac and Windows?
Yes. Kimi Work ships as an AI Mac app and an AI Windows app, offering the same desktop AI software on both platforms.
Is data safe with Kimi Work?
Kimi Work runs locally and executes code on the local machine, so files remain on the computer. It also requests approval before making changes, which keeps the user in control.
Are coding skills required to use Kimi Work?
No. This AI agent desktop is driven entirely through natural language, with no commands, terminal, or API keys required, even though it can run code in the background.
Which data sources does Kimi Work connect to?
Kimi Work connects natively to data sources, supporting data from financial, economic, and academic fields such as the World Bank.
Can Kimi Work run tasks automatically on a schedule?
Yes. Tasks can run in the background on a schedule, such as morning briefings or end-of-day reports, allowing the agent to keep working while the user is away.
In Goal Mode, will Kimi Work just keep running on its own without limit?
No. A goal runs in rounds, and at the end of each round, Kimi Work decides whether the goal is complete, blocked, paused, or ready to continue. You can also set budget limits in your instruction, such as a maximum number of rounds, tokens, or time, and define what counts as “done” through acceptance criteria. The process stays observable and open to intervention throughout, so you can check progress, review intermediate results, and adjust the direction or stop it at any point. It does not run blindly toward an undefined target.
When I connect a third-party account to a plugin, what is Kimi allowed to access?
Only what you authorize. Plugins connect through MCP, OAuth, or open-platform capabilities, and you can only connect accounts you own or are authorized to use. The authorization scope is defined by the service you're connecting to, so Kimi acts within those granted permissions rather than gaining open-ended access to your account. This keeps the access boundary clear and under your control, and you can uninstall a plugin when you no longer need it.