Kimi Code is Moonshot AI's official agent for desktop, the command line and IDEs, built around Kimi K3 with up to 1M tokens of context. It reads and writes code, runs commands, searches files and the web, and decides what to do next from the results.
Already have an account? Sign in
Kimi CodeAbout 10 minutes: install Kimi Code CLI → create a key in Portal → add the Codelux provider and models to config.toml → start and switch models.
The install script doesn't need Node.js; the npm route needs Node.js 22.19.0+. A true-colour terminal such as Kitty or Ghostty is recommended.
curl -fsSL https://code.kimi.com/kimi-code/install.sh | bashirm https://code.kimi.com/kimi-code/install.ps1 | iexInstall Git for Windows before first launch. If Git Bash is in a custom location, set KIMI_SHELL_PATH to the absolute path of bash.exe.
brew install kimi-codenpm install -g @moonshot-ai/kimi-codeOpen a new terminal and run:
kimi --versionUpgrade later with kimi upgrade.
/login only supports Kimi's own accounts. To use Codelux, edit ~/.kimi-code/config.toml or add it interactively with /provider.
sk-) and the API base URL. The key is shown in full only once, so keep it somewhere safe.The config file is created the first time you run kimi. Open it and add:
| OS | File |
|---|---|
| macOS / Linux | ~/.kimi-code/config.toml |
| Windows | %USERPROFILE%\.kimi-code\config.toml |
default_model = "kimi-k3"
[providers.codelux]
type = "openai"
base_url = "https://codelux.io/v1"
api_key_env = "CODELUX_API_KEY"
[models."kimi-k3"]
provider = "codelux"
model = "kimi-k3"
max_context_size = 200000
[models."claude-sonnet-5"]
provider = "codelux"
model = "claude-sonnet-5"
max_context_size = 200000
[models."glm-5.2"]
provider = "codelux"
model = "glm-5.2"
max_context_size = 128000[providers.codelux] with type = "openai" uses the Chat Completions API; base_url is the Codelux API base URL with /v1.api_key_env names the environment variable holding your key. Kimi Code does not read shell variables unless you name them here.[models."<alias>"] entry needs provider, model (the model ID in Portal → AI Hub) and max_context_size. Quote aliases that contain a dot.default_model must be one of the aliases you defined.Then set the environment variable:
echo 'export CODELUX_API_KEY=sk-your-codelux-key' >> ~/.zshrc
source ~/.zshrcsetx CODELUX_API_KEY "sk-your-codelux-key"Prefer not to edit TOML? Start kimi, run /provider, choose Add New Platform and fill in the provider type (OpenAI-compatible), API base URL and key.
/provider manager. · Kimi Code docs (MIT)Start Kimi Code in your project, confirm the model comes from Codelux, then switch as needed.
cd your-project
kimiThe Model line in the welcome box should show your default_model. Run /status for the version, model and working directory. After editing config.toml, run /reload to apply it without restarting.
Run /model and choose from the models defined in config.toml. Add more [models."<alias>"] entries to get more choices.
| Action | What it does |
|---|---|
/init | Create AGENTS.md for the project |
Shift + Tab | Toggle Plan mode |
kimi -p "..." | Run a single instruction |
kimi -c | Resume the previous session |
/compact | Compress the conversation to free context |
See each request's model, tokens and charge in Portal → Logs.
The variable named in api_key_env is empty in this terminal. Set it (or use api_key) and reopen the terminal.
default_model must match an alias defined under [models], not a raw model ID that isn't configured.
For type = "openai", base_url must end with /v1.
CC-Switch doesn't manage Kimi Code at the moment; configure it in config.toml or with /provider.