Moonshot AI released Kimi Code Desktop on the afternoon of September 21, the official desktop client for Kimi Code. The macOS and Windows versions launched the same day, with the installer available at kimi.com/code. The macOS build supports both Apple silicon and Intel Macs.
Until now, Kimi Code lived mainly as a command-line tool in developers' terminals, while a number of third-party coding tools called the model behind it through an API. The desktop app wraps the same agent capabilities in a graphical interface: users describe a task in a chat box, and the agent reads and writes code, runs commands, and carries out the automation, listing each step on screen as it happens. Moonshot's own description is that it lets the agent "carry a task all the way from planning and execution through to delivery."
What the desktop version adds
The official launch post groups the new features into roughly three areas.
Project and configuration management. All projects sit in one window, and common settings can be adjusted through the graphical interface instead of hand-editing config files. Once signed into a Kimi account, subscription usage is visible directly in the client.
Built-in terminal, browser and Git view. Projects can be run and debugged directly inside the client, and web projects can be previewed in the built-in browser. A Git status panel brings code changes, review and PR-merge progress together in one place, so users don't need to switch windows to see which files changed or where a task is stuck.
Screenshot annotation. Users can draw a box around part of the screen, add a note, and hand it to the agent to fix — useful for front-end styling tweaks that are hard to describe in words.
On the model side, the client defaults to the official model included in the Kimi Coding Plan, but also lets users configure third-party model providers. Moonshot frames version 1.0 as a starting point, saying more features are still in development.
What the launch post leaves out
The official post doesn't go into detail on model or pricing. A tech outlet published hands-on coverage of the desktop client on September 18, three days earlier, reporting that it ships with K3 built in, can connect to third-party models such as DeepSeek, and comes as a 143MB installer. That coverage also described Computer Use (controlling the mouse to operate other applications), a plan mode, a goal mode, and a plugin called Kimi Datasource for querying financial, macroeconomic and corporate data. Subscriptions were reported at three tiers — 99, 199 and 699 yuan a month — with subscribers also getting Kimi membership benefits.
None of these details appear point-by-point in the September 21 official post, and there's no public explanation of how much usage each of the three pricing tiers includes; for now, Kimi's own plans page is the reference. One thing also doesn't line up: on the 11th of this month, Kimi Code rolled K2.8 Preview out to all users, with Moonshot describing its overall performance as close to K3. Which model each subscription tier actually calls inside the desktop app hasn't been officially addressed.
Coding agents are all moving to the desktop
Packing a command-line agent into a desktop client has been a common move across coding tools this year. Anthropic's Claude Code started as a terminal tool before shipping a desktop version with parallel multitasking; OpenAI's Codex also has a standalone desktop app. The approaches are similar: the terminal stays for power users and automation scripts, while the desktop version lowers the barrier so people who don't normally open a terminal can direct an agent to change code, with the progress of multiple projects and tasks laid out in one interface.
Kimi Code's bet here is model openness. Within the same client, users can switch between third-party models, letting developers in China pair domestic models like DeepSeek with Kimi's own model instead of installing a separate tool for each. Looking at the domestic market, there aren't many yuan-priced desktop coding agents built for Chinese-language users — ByteDance's TRAE and Alibaba's Tongyi Lingma take the IDE-or-plugin route, a different form factor from Kimi Code Desktop.
For people already using the command-line version, the desktop app doesn't change the underlying capability — the changes are in the interaction: task progress is now visualized, change review sits in one place, and a screenshot can serve directly as an instruction. Team users should take a closer look at configuration: the launch post doesn't say whether third-party model settings in the desktop version share the same setup as the command-line version, so it's worth testing the two separately before wiring either into a team workflow.
Sources: Kimi official launch post, IT Home, CocoLoop, Sina Technology; installer size, built-in model and subscription tiers checked against tech-media hands-on coverage.