The Dark Side of the Moon releases Kimi Work, supporting up to 300 agents collaborating.

According to Beating monitoring, the Kimi Work beta version—an all-purpose local agent client for knowledge workers—has been released together with the latest test version of Kimi’s Mac client, while the Windows client is coming soon. Users can dispatch files, use browsers, run scheduled tasks, and generate documents, spreadsheets, or presentation PPTs on their local devices through natural language, without having to configure their terminal environment.

Kimi Work’s development is deeply integrated with AI collaboration. Engineers from the Dark Side of the Moon completed the client build within one week using the local programming agent Kimi Code. Of more than 50,000 lines of effective code produced, the proportion autonomously generated by AI reached 92%. The fast client development relies on the long-duration task execution capability of the underlying Kimi K2.6 model, which supports 13 hours of continuous coding and can autonomously call tools more than 4,000 times in a single task.

On the core architecture, Kimi Work has transitioned from the previous command-line interface to a graphical user interface, enabling the shift from a programming agent to a general-purpose work agent. The client includes a built-in Kimi WebBridge solution that can control browsers, and integrates professional capabilities such as building websites and creating PPT presentations for the online version of Kimi Agent, as well as financial, scientific research, and legal databases. For complex tasks, Kimi Work supports multi-agent cluster collaboration: it can autonomously create teams of up to 300 sub-agents and process tasks in parallel based on the complexity of the task.

At the current testing stage, the main focus is to evaluate Kimi Work’s capabilities in tasks such as generating long deliverables, handling local files, and performing browser operations. In office scenarios, Kimi Work can read local product plans and design prototypes, log into subscribed databases via the browser, analyze more than 2,400 user comments to generate cross-border e-commerce market analysis reports, and finally convert them into presentation slides with a single click using PPT skills.