On September 21, 2026, Moon Dark Side announced via its official WeChat public account that Kimi Code Desktop 1.0 will be launched simultaneously with macOS (Apple and Intel chips) and Windows versions, available for download on kimi.com/code. This means Kimi Code has evolved from an open-source CLI into an official desktop client, and competition among programming agents has extended from model capability to "form wars": as models converge, the key to victory becomes whose workflow is smoother.
From the terminal to the desktop, it's not just the interface that has changed
The desktop fully retains Kimi Code's Agent capabilities: reading and writing code, running commands, and completing automated tasks through conversation. At the same time, it consolidates operations previously scattered in terminal commands and configuration files into a single window: centralized project management, visual adjustments of commonly used configurations, and clear visibility of every step of Agent task execution. After logging into a Kimi account, official models can be used directly, and third-party model vendors are also supported; The workspace is organized by folder, with tabs, pinning, and global search to ensure multiple projects run parallel without confusion. Existing local CLI tasks appear directly on the desktop, ensuring no gap in migration for existing users. For developers' actual workflows, the desktop also includes built-in terminal, browser, and Git status viewing, supporting direct debugging projects, reviewing code changes, and associating PR status.
The four working modes are truly differentiated
The Plan mode is suitable for tasks with wide impact, issuing execution plans first and confirming them before acting; The Goal mode sets clear objectives for the Agent, advancing multiple rounds and continuously checking around them; In the Swarm mode, the main Agent breaks tasks into sub-tasks, scheduling the Subagents to divide labor and summarize results; The Tower mode (experimental) allows multiple Agents to advance in parallel around the same goal. The latter two models turn multi-agent collaboration into product features, rather than just a demo concept. The four modes together cover a spectrum from "think carefully before acting" to "parallel division of labor to accelerate progress," allowing developers to choose the appropriate working method based on task risk and scale.
Viewed within the competitive landscape
Competitors like Cursor and Claude Code have long proven that the success of programming agents is not just about models. Kimi Code Desktop's opportunity lies in its backing by the Kimi Coding Plan subscription system and compatibility with third-party model providers, making it more user-friendly for domestic developers; Its shortcomings are also obvious: the GUI of plugins and skills is still incomplete, and veterans heavily dependent on terminals may find this unnecessary. For more Kimi product updates, see the Kimi special section.
