Kimi Work is a desktop AI agent for knowledge workers who want software to carry out multi-step tasks on their own computer. It is a separate product experience from the Kimi K3 model page: Kimi Work is the local Work mode, while Kimi K3 is the model and broader Kimi product family. This review uses Moonshot AI’s English help documentation and release notes. We did not run an independent benchmark or claim first-hand performance results.
What Kimi Work is
Kimi describes Work as a general-purpose local Agent inside the Kimi desktop client for macOS and Windows. Instead of opening a terminal or configuring a development environment, a user describes a goal in natural language. The agent can break the goal into steps, run tasks in parallel, use a browser, create or organize folders, and deliver documents, spreadsheets, slide decks, or other files on the local machine. Kimi says the Work mode is powered by the Kimi Code kernel and Kimi K3 model, but the intended user is a knowledge worker rather than a software engineer.
The product has a Work and Chat split. Work includes a dashboard, projects, plugins, skills, scheduled tasks, WebBridge, and a file-oriented workbench. The overview also documents widgets, persistent dashboards, browser operation, and Agent Swarm, which can assemble up to 300 sub-agents for more complex jobs. Those are product claims from Kimi’s documentation, not independent measurements of speed or reliability.
How the workflow is different
The practical distinction is the execution surface. A normal chat assistant may give instructions, while Kimi Work is designed to act on files and browser tasks after the user grants permission. Its documented permission levels are Default, Manual approval, and Fully automatic. In Default mode, routine operations can run automatically while sensitive actions such as modifying, overwriting, or running code on local files require authorization. That control model makes it possible to use the agent for repeatable office work without giving every task unlimited access.
For a fair evaluation, start with a reversible task: collect a small set of public documents, summarize them into a spreadsheet, and ask the agent to show the source files and generated output. Then test a second task that uses a browser or a plugin. Record which steps required approval, what files were created, and whether the final deliverable is easy to inspect. This is a better decision signal than a generic chatbot leaderboard for a local agent.
Who should use it
- Good fit: analysts, researchers, marketers, operations teams, and founders who regularly move information between files, web pages, reports, spreadsheets, and presentations.
- Good fit: users who want a desktop workbench, scheduled tasks, reusable skills, or plugin connections to services such as Canva, Notion, and WPS.
- Possible fit: developers who want a single desktop surface that can hand technical work to the Kimi Code kernel, but the dedicated Kimi Code workflow may be more direct.
See our Kimi K3 review for the model and general Kimi product context, and use the practical guides when you need a repeatable evaluation checklist.
Limitations and who should skip it
Kimi Work launched in June 2026 and remains in beta according to the official overview. The release notes show frequent iteration, including fixes for crashes, login state, browser failures, and Windows behavior. That pace is useful for early adopters but means menus, permissions, supported file types, and stability can change between versions. It is not a fit for teams that need a mature, fixed enterprise workflow, guaranteed uptime, or a documented compliance contract before connecting local files and third-party services.
Do not infer that “local Agent” means every action stays offline. The product can use web search, browser operation, plugins, and professional databases. Before using confidential material, confirm the current privacy policy, account controls, plugin permissions, and where each connector processes data. Our comparison hub can help you compare this execution model with browser-first or chat-first tools.
Kimi Work vs Kimi K3
Kimi K3 is the model and product family page; Kimi Work is the desktop agent built around that model and the Kimi Code kernel. Choose Kimi Work when the outcome is a local deliverable and you need file handling, browser actions, scheduled tasks, or permission checkpoints. Choose the Kimi K3 experience when you mainly need conversation, long-context analysis, or model access without a desktop automation layer. They are related, but they should not be treated as two independent model benchmarks.
Our editorial verdict
Kimi Work is a credible new entry for the desktop-agent category, especially for people who want an AI assistant to complete a chain of office tasks rather than only draft text. Our positive view is based on the documented workflow surface, permission controls, and fast release cadence—not on a fabricated speed score or hands-on test. The main reason to wait is beta risk: verify the exact version, permissions, integrations, and data handling on the day you adopt it. We will revisit this page on the next review date or sooner if Kimi changes the product’s availability or privacy model.
Primary sources and review method
Official sources: Kimi Work Overview and Kimi Work Release Notes. We used the official product description, permission model, system requirements, beta status, and dated release notes. No independent benchmark, screenshot, or first-hand user-experience claim is presented as evidence.
September 2026 release check: inspectability and handoff improvements
Kimi’s official release notes show a concentrated September desktop cycle. Version 3.2.9 added share-to-Kimi and open-with-Kimi flows for supported files, a refreshed desktop pet, a PDF index panel, and a power option that keeps background work running after the screen turns off. Version 3.2.10 temporarily removed the Windows share-target companion to address a launch crash, then 3.2.11 restored the share capability and added a detail panel that exposes sub-agent execution progress.
These are product facts from Kimi’s release notes, not results from an AI Tools Manual hands-on test. The practical signal is that Kimi Work is becoming more inspectable and more integrated with the desktop file system, while the product remains in beta and changes quickly.
Checks before relying on a release
- Confirm that the installed version is 3.2.11 or later and that the Windows shortcut still points to the current app.
- Review the three permission modes before allowing file edits, code execution, browser actions, or scheduled work.
- Test share/open-with behavior on a non-sensitive folder and confirm the file preview and PDF index work for your formats.
- Use the execution detail panel to review sub-agent actions, but do not treat visibility as proof that every output is correct.
- Recheck beta availability, operating-system requirements, plugins, privacy, and data-handling terms before a team rollout.
For the model context, see our Kimi K3 review. For a broader workflow choice, compare Kimi with ChatGPT. The primary evidence is Kimi’s official release notes and official overview.
September 26, 2026 release check: Kimi Work 3.2.14 and precise document editing
Kimi’s official release notes now list version 3.2.14 (September 24) as the latest desktop build. The release adds phone-based remote-control improvements: users can preview Office and HTML files on a phone and upload attachments from a phone to Kimi Work. It also adds Side Chat, text quoting, version history for Office files, and browser annotations that can be quoted into the input box. The notes say Keep Awake is now on-demand to reduce background power use, and include a Windows taskbar restore fix. These are documented product changes, not results from an AI Tools Manual hands-on test.
Version 3.2.12 (September 22) is also important for document workflows: users can select text in PPT, Excel, Word, or Markdown files and ask for precise edits; the previous attachment-size limit was removed; and the archived-tasks tab was rebuilt. Version 3.2.11 added a detail panel for sub-agent execution, while 3.2.7 added a one-step website deployment plugin. Taken together, the recent cycle makes Kimi Work more inspectable and more useful for a phone-to-desktop handoff, but it also expands the number of places where a user must review files, permissions, and history.
Who should use the new workflow
This is a good fit for analysts, operations teams, and researchers who move Office files between a phone and a desktop and want an approval checkpoint before edits. It is less suitable for regulated or confidential work until the team confirms retention, connector permissions, version-history behavior, and whether remote-control sessions can reach the intended machine. Our decision rule is to start with a reversible copy, inspect the change history, and only then allow edits to the working file.
Checks before relying on the update
- Confirm the installed build is 3.2.14 or later and check the Windows/macOS support notes.
- Use a non-sensitive Office file to test phone preview, attachment upload, precise editing, and version rollback.
- Keep remote control and Keep Awake off unless the task needs them; verify who can access the session.
- Review the execution detail panel and file history, but do not treat visibility as proof that an edit is correct.
The primary evidence is Kimi’s official release notes. For the product’s permission model, see the Kimi Work overview. Compare the desktop-agent workflow with our Kimi Work vs Replit comparison and use the AI privacy checklist before connecting business files. These sources support the documented features; we have not independently benchmarked speed, accuracy, stability, or data handling.
Kimi Work 3.2.14 improves the remote desktop workflow
Moonshot’s official Kimi Work release notes list version 3.2.14 on September 24, 2026. The update lets users preview Office and HTML files directly on a phone, upload attachments from a phone to Kimi Work, open a Side Chat that inherits the current session context, quote selected text into a conversation, manage version history for Office files, and annotate web pages before quoting them into the input box. “Keep Awake” is now on demand, which the release notes say reduces background power consumption.
Version 3.2.12, released two days earlier, added precise edits to selected content in PPT, Excel, Word, and Markdown files and removed the original size limit for input-box attachments. These are workflow changes rather than a new foundation model, but they matter for people using Kimi Work as a desktop agent across documents, a phone, and a browser.
Best fit and limits
Kimi Work is a strong fit for users who want a local desktop workspace with remote control, file previews, and agent-assisted editing. The new version-history and quoting features make it easier to trace changes and keep context, but they do not replace a repository, document-management policy, or backup strategy. Remote control also expands the security boundary: a lost phone, over-broad permission, or ambiguous prompt can expose or change desktop work.
The release notes do not establish universal support for every file type, mobile operating system, account tier, or regional download path. Teams should confirm supported formats, login state, local file permissions, power settings, and where attachments and conversation data are stored before using confidential material.
Our decision rule
Use 3.2.14 for bounded document work first: preview a file, quote a small section, make one reversible edit, and inspect version history before approving a larger change. Keep remote control off when it is unnecessary, and maintain a separate backup for important files. Read the Kimi Work profile and the Kimi Work vs. Replit comparison when deciding between a desktop agent and a cloud app builder.
Evidence disclosure
Facts above come from Kimi’s official release notes. AI Tools Manual did not independently install 3.2.14 or verify mobile previews, remote control, file-size behavior, or data handling. Confirm the current installer, supported platforms, account permissions, and privacy terms before approval.
September 30, 2026: Kimi Work 3.2.15 adds Notion and Feishu co-editing
Moonshot AI’s English release notes list Kimi Work 3.2.15 on September 30. The release adds a path to open Notion and Feishu documents from the home-page sidebar, displays plugin-created documents as artifact cards, and lets users open documents in the built-in browser for comment editing and page questions. The notes do not specify plan eligibility, rollout geography, supported permission scopes, or whether every connected workspace can use each operation. Confirm those details inside the client and the linked service’s authorization screen before adopting it for shared documents.
Workflow and reliability changes
The same release collapses the Agent execution process by default so the final reply is shown first, while the user can expand to inspect details. Projects in the left sidebar can now be reordered by dragging, and Kimi Work prompts a user before signing out while tasks are running. The release also says it fixed an issue where concurrent PowerPoint editing could corrupt files, improved launch speed, and addressed several other bugs. This is useful for people moving between documents and long-running desktop tasks, but a release note is not a guarantee that every collaboration edge case has been eliminated.
Who benefits, and what to verify
The update is most relevant to knowledge workers who already use Notion or Feishu and want an AI assistant to work with documents inside a desktop-agent workflow. It is less compelling for teams that require a documented enterprise permission matrix, auditable change history for every edit, or a confirmed compatibility list before enabling access. Because the visible Agent process is now collapsed by default, reviewers handling consequential work should expand and inspect the execution details rather than treating a polished final reply as a complete audit trail. For local-versus-cloud trade-offs, read Kimi Work vs. Replit and our AI-agent rollout checklist.
Evidence disclosure — verify before approval
The primary source is Moonshot AI’s Kimi Work release notes. AI Tools Manual did not independently test document editing, concurrent PowerPoint behavior, launch speed, or the security and privacy of connected workspaces. The source is the vendor’s English changelog; it does not establish broader product availability or a service-level guarantee.
Reliability note for version 3.2.15: Moonshot’s release notes also say the update fixes a concurrent PowerPoint editing issue that could corrupt files, a Windows console flash issue, and unclear messaging when opening a missing file. Teams that edit decks in parallel should confirm they are on 3.2.15 or later and keep a recoverable copy; the release notes do not identify every affected scenario or describe recovery for files already damaged. We have not independently reproduced the bug or verified the fix. Primary source: Kimi Work release notes.
October 12, 2026 release: PDF preview and background browser tasks
Kimi Work version 3.2.16 improves the PDF preview engine, including zoom, citation controls, and text-selection gestures. The release notes also say the built-in browser is no longer interrupted while tasks run in the background. These are workflow and reliability refinements; they do not establish a new model capability or an independent performance result.
For people reviewing cited PDFs or leaving a browser task running while the agent works elsewhere, the changes may remove small interruptions. After updating, check citation selection and the background-browser behavior in your own workflow before depending on them for a deadline. Moonshot’s notes do not describe every affected file or guarantee that all browser tasks complete unattended.
Editorial assessment: this is a useful maintenance release for document-heavy work, not a reason by itself to grant Kimi Work broader file or browser access. For model-specific context, see our Kimi K3 review. Primary source: Kimi Work release notes.
October 21–22, 2026: Kimi Work adds enterprise plugin and access controls
Kimi Work release notes 3.2.18 and 3.2.19 describe several enterprise-facing changes. Version 3.2.18 lets enterprise users create custom plugins and upload them to the plugin marketplace, adds workspace switching with role labels, and improves visibility into tool-call results. Version 3.2.19 adds network diagnostics and a clearer restricted-access page when an enterprise IP allowlist blocks entry, with options to retry or switch back to a personal workspace.
Who is affected: administrators managing more than one enterprise workspace, and teams that plan to distribute custom plugins. The release notes establish that these controls exist; they do not, by themselves, answer how marketplace plugins are reviewed, what each role can change, how audit events are retained, or how personal and enterprise data boundaries behave.
Editorial assessment: the changes improve operational clarity, but a plugin marketplace is also a software-supply-chain decision. Before enabling uploads, document who can publish, how code and credentials are reviewed, how plugins are revoked, and whether the marketplace is internal or shared. Test the allowlist flow with a non-admin account and verify that switching workspaces does not expose the wrong files or tools. We did not independently test these controls.
Primary source: Kimi Work release notes. Related: our human-review workflow guide. Next review: November 23, 2026. Human review: confirm role permissions, plugin review and rollback, allowlist behavior, and workspace data boundaries.