@lunaluva/codex

Configure a Lunaluva endpoint for local AI coding clients.

npm install -g @lunaluva/codex
lunaluva
lunaluva test

Default endpoint:

https://codex.lunaluva.com/v1

Default model:

gpt-5.5

The API key is stored with the platform-native secret mechanism: macOS Keychain on Mac and a Windows DPAPI-protected local secret file on Windows. The generated shell files live in ~/.codex-lunaluva/ with mode 0600.

Before changing shell, PowerShell, or Codex configuration files, the CLI saves the original file state under ~/.codex-lunaluva/backups/. lunaluva remove restores those originals so a rollback returns to the pre-Lunaluva state while keeping the API key in the local secret store for the next apply.

Commands

lunaluva login --key sk-... --clients all
lunaluva
lunaluva test
lunaluva models
lunaluva remove

Supported client targets:

• global: writes OpenAI-compatible and Anthropic-compatible environment exports and sources them from ~/.zshrc.
• codex: writes a managed Codex provider block into ~/.codex/config.toml, points it at the Lunaluva base URL, disables Responses websocket transport for Lunaluva compatibility, and stores the API key in Codex's local auth.json API-key login mode. It also backs up and narrows ~/.codex/models_cache.json so Codex Desktop's model menu matches the Lunaluva models exposed by 月渡 AI.
• claude: writes Claude Code-style environment exports and sources them from ~/.zshrc.
• cursor: writes the values needed for Cursor settings to a local helper file. Cursor does not expose a stable public CLI config API, so this package does not edit Cursor internals.

lunaluva remove removes the managed shell/Codex blocks and generated files, but keeps the saved API key. Use --delete-key only when you also want to forget the local secret-store entry.

lunaluva test sends hello to the selected model through /v1/chat/completions; this is a stronger end-to-end check than listing models.

lunaluva models pulls the currently available model list from the configured Lunaluva endpoint. In the interactive menu, option 5 lets you choose one of those models and writes it back to Codex and other selected local clients as the default model; for Codex it also refreshes the local model menu cache.

额度、余额和用量明细请在月渡 AI 后台查看；本地配置助手不再提供该菜单项。