Kimi Work desktop AI agent launch

Kimi Work desktop AI agent launched by Moonshot AI is a downloadable desktop agent available for macOS and Windows. The Kimi Work desktop AI agent runs on the user’s machine, can access local files, and includes the WebBridge extension that lets agents drive real Chrome or Edge sessions locally. It supports an Agent Swarm of up to 300 sub-agents running in parallel. Source: Provided Content

Technical architecture and operational environment

Kimi Work operates on the user’s machine as a downloadable desktop agent for macOS and Windows and can access local files and interact with the computer. Browser interactions are managed through the Chrome DevTools Protocol, and logged-in sessions and cookies remain on the user’s machine rather than being transferred off-host. The WebBridge extension enables agents to drive real Chrome or Edge sessions locally, so browser automation runs within the same local environment as the agent. Because these components run on the host, browser session data and authentication state are handled on the same machine where Kimi Work executes.

Kimi Work exposes a local file layer that lets agents read mounted folders and interact with files available on the system for tasks that require local data access. Agents can run Python scripts in the background to process local data or perform computations without leaving the host environment. A Cron engine schedules tasks on daily, hourly, or conditional intervals and includes a Keep Computer Awake toggle intended to support overnight jobs. The operational environment also supports Agent Swarm execution, allowing up to 300 sub-agents to run in parallel on the same machine to carry out coordinated work.

AI model and processing capabilities

Kimi Work utilizes the Kimi K2.6 model, a sophisticated AI architecture with a trillion-parameter mixture-of-experts design. Released on April 20, this model is notable for its capacity to activate approximately 32 billion parameters per token processed. Furthermore, it features a substantial 256K-token context window, which allows for extensive input handling in tasks requiring a wide contextual understanding. These capabilities enable Kimi Work to efficiently manage complex computations and AI-driven tasks on the user’s machine, enhancing the productivity and versatility of the desktop AI agent.

Kimi Work combines local execution on the user’s machine with the ability to run multiple sub-agents concurrently to tackle coordinated tasks. The desktop agent provides local file access, scheduled background execution, and browser-control integration via the WebBridge extension to maintain sessions and operate against local resources. It also incorporates pre-integrated market data for several major markets and is built on a large-scale mixture-of-experts AI model to support complex workflows.