ChineseModels for Kimi Code - Windows
Tested with Kimi Code 0.29.1 on 2026-10-22.
Requires Node.js 20+, Windows PowerShell 5.1+, Kimi Code and a ChineseModels key.

Extract this archive, inspect the scripts, then run:
  powershell -NoProfile -ExecutionPolicy Bypass -File .\install-kimi-windows.ps1

Enter your key at the hidden prompt. The native Kimi provider stores it in
~/.kimi-code/config.toml as plain text, like Kimi's built-in API-key setup.
KIMI_CODE_HOME is respected. To manage an environment variable instead,
replace api_key with api_key_env pointing to your variable; do not set both.

Existing providers, login, hooks, and default model are preserved. A timestamped
backup is saved before an existing config is changed. The managed ChineseModels
block is refreshed from your authenticated /v1/models list. Image/video models
are excluded. Rerun the installer to update it.

Run: kimi doctor config
Restart Kimi and type /model. Look for (ChineseModels) entries.
Or: kimi -m "chinesemodels/DeepSeek-V4-Flash"

Thinking metadata varies by model. Kimi forwards standard reasoning_effort;
this native installer does not normalize vendor thinking.type/enable_thinking
fields. A visible toggle is not proof that upstream thinking can be disabled.
The conservative 32768 context setting is not a claim about model capacity.
Model discovery and a text reply do not certify tool use. Test your workload.
Legacy kimi-cli uses another configuration schema and is not supported here.

Guide: https://chinesemodels.ai/documentation/kimi-code
