On June 3, Moonshot AI announced the public beta launch of Kimi Work, a general-purpose local Agent designed for knowledge workers, which will be released alongside the latest beta versions of the Kimi Mac and Windows clients.
According to the announcement, users simply need to describe their objectives in natural language, and Kimi Work can break down tasks, execute them in parallel, call tools, use a web browser, create and organize folders, and deliver work products such as documents, spreadsheets, and presentations on the user's computer.
The core of Kimi Work is Kimi Code, a local Coding Agent used daily by hundreds of thousands of programmers and all Kimi team members. It provides foundational local Agent capabilities, including installing and using Skills and running scheduled tasks.
Kimi Work inherits the professional Skills from the online Kimi Agent version, such as website building and presentation creation, along with specialized databases for finance, scientific research, and law. It also features the built-in Kimi WebBridge solution, which enables it to use a web browser just like a human user.
Most importantly, it supports Agent clustering, allowing it to autonomously create a team of up to 300 sub-Agents based on task complexity to handle more intricate and time-consuming assignments.
From Coding Agent to Working Agent
Kimi Work essentially represents a shift from TUI to GUI and from a Coding Agent to a Working Agent. It migrates the Agent capabilities already validated by Kimi Code in engineering scenarios to the desktop interface familiar to general knowledge workers.
Users no longer need to open a terminal, write commands, or configure environments. They simply describe their goal in natural language, and Kimi Work can handle task decomposition, parallel execution, tool invocation, web browsing, folder management, and the delivery of documents, spreadsheets, and presentations on the computer.
Interestingly, Kimi Work itself was developed with significant participation from Kimi Code. During the development process, Kimi engineers used tools like Kimi Code to complete the beta versions of the Mac and Windows clients within one week.
The development process cumulatively produced over 50,000 lines of effective code, with 92% of it autonomously generated by AI. This achievement is underpinned by the long-range task execution capability of the Kimi K2.6 model, which supports 13 hours of continuous coding, parallel collaboration among 300 sub-Agents, and over 4,000 autonomous tool calls.
From Vibe Coding to Vibe Working
While Kimi Code serves developers who write code, Kimi Work serves knowledge workers bogged down by tedious daily tasks. The former frees developers from repetitive implementation, while the latter liberates knowledge workers from information transfer, file organization, data analysis, and report generation.
Let's explore what Kimi Work can help you accomplish:
Financial Scenario
Instruct Kimi Work to research Berkshire Hathaway's portfolio holdings over the past decade, summarize the investment strategy, and generate a Buffett investment Skill. Then, install this Skill and use it to connect to Kimi's professional financial data sources to analyze whether your portfolio and investment strategy align with Buffett's value investing principles.
Workflow: Research → Generate Skill → Install Skill → Use Skill and call Kimi's professional financial data sources for analysis.
Research Scenario
Task Kimi Work to act as an urban computing researcher, helping to clean your local simulation dataset, summarize core characteristic patterns, and create academic charts. Then, use this cleaned data and charts to assist in completing your research paper.
Workflow: Local Data Cleaning → Chart Creation → Paper Completion Assistance.
Office Scenario
Direct Kimi Work to read product proposals, industrial design prototypes, and Go-To-Market plans from folders on your computer. It can then utilize the Kimi WebBridge capability to log into subscribed data dashboard websites via a browser for research.
By conducting in-depth research on the overall market, overseas legal regulations, and competitor business landscapes, and combining this with an analysis of over 2,400 user comments from a proprietary dashboard, it can generate a 128-page cross-border e-commerce market analysis report.
Finally, instruct Kimi Work to call a dedicated PPT Skill to transform the report content and local design prototypes into a presentation with a single click.
Workflow: Local File Reading → Browser Access to Subscribed Databases → In-depth Report Generation → Calling PPT Skill → Client Proposal Output.
Comments