Kimi Code
A practical map of the new TypeScript-based CLI, its model choices, installation flow, and third-party endpoints.
Install the current CLI
The official scripts install a single binary. Open a new shell after the installer finishes.
irm https://code.kimi.com/kimi-code/install.ps1 | iexcurl -fsSL https://code.kimi.com/kimi-code/install.sh | bashRun kimi, then /login
Choose Kimi Code OAuth or a Moonshot AI Open Platform API key, then use /model to select the model lane.
The model lanes
Three IDs, with a real trade between frontier capability, coding focus, and output speed.
| Model ID | Underlying model | Use it for | Operational note |
|---|---|---|---|
k3 | Kimi K3 | Hard, long-horizon, visual, and large-context engineering | Up to 1M context; low, high, and max reasoning |
kimi-for-coding | Kimi K2.7 Code | Reliable day-to-day coding and repository tasks | Standard speed and quota use |
kimi-for-coding-highspeed | Kimi K2.7 Code | Fast iteration when latency is the bottleneck | About 5–6× faster, with 3× quota usage |
Switching models invalidates the current context cache. Start a fresh session before a serious side-by-side evaluation.
What the current CLI can do
Kimi Code can read and edit files, run shell commands, search and fetch web pages, use built-in subagents, connect to MCP servers, and apply lifecycle hooks around risky operations. It is a new TypeScript-based product rather than a small update to the legacy Python CLI.
Third-party tool access
The Kimi Code service documents an OpenAI-compatible base URL at https://api.kimi.com/coding/v1 and an Anthropic-compatible base URL at https://api.kimi.com/coding/. Those endpoints use Kimi Code membership credentials and model IDs, not the general API model names.
Windows requirement
On Windows, the official help center says Git for Windows should be installed before first launch because Kimi Code uses its bundled Git Bash shell environment.