Kimi Code
Operating Kimi Code on workers — sign-in, prompt mode and its limits, several repositories, and the retired Python kimi-cli.
Install
npm install -g @moonshot-ai/kimi-codeThe worker looks for kimi on the worker user's PATH.
Sign-in
Start kimi once as the worker's user and use /login (or run kimi login). Without a login a run ends with
"No model configured", which the worker reports as authentication required.
How it runs
kimi --output-format stream-json [--model <alias>] [--add-dir <dir>] -p <prompt>
- Prompt mode cannot be combined with
--yoloor--auto("Cannot combine --prompt with --auto"), so neither is passed. --modeltakes a model alias from Kimi's own configuration.- The older Python package
kimi-clialso installs akimicommand, which now only prints that it is no longer maintained. Install Kimi Code instead; its installer renames the old command. - Projects with several repositories reach it: the other repositories are passed with
--add-dir.
Add-on models
Kimi Code cannot use add-on models or providers configured on the worker. When it reaches its limit, the task moves to another compatible agent, or waits, as the fallback policy says.
Sessions
Resume is not claimed for Kimi Code: no resume command could be verified without an account. After a crash, recovery starts a fresh session from the checkpoint.
What was verified
Checked against Kimi Code 2.1.1 on 2026-10-06: its --help, and a real run without credentials through the
worker's session runtime. A task with the vendor's account has not been run.