Command Code API provider for opencode. Use Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, MiniMax, Step, and other models through a single API key.

Forked from brent-weatherall/opencode-commandcode-provider — originally created by Brent Weatherall.

繁體中文說明請見 README.zh.md

• Auto-load catalog from local command-code — on startup, scrapes the model catalog (including reasoningEfforts) from your locally installed command-code npm package. Falls back to bundled models.json if the package is not found.
• Reasoning effort variants — models with reasoningEfforts automatically get variants (e.g. low / medium / high / xhigh / max). Upgrade command-code and restart to pick up new models and efforts — no manual sync needed.
• Shared catalog module — src/catalog.ts provides unified package resolution, bundle extraction, and model-building logic used by both the plugin and the sync command.
• No more dropped models — models without cost data now receive a conservative default cost instead of being silently dropped.
• Deduplicated display names — models sharing the same upstream display name (e.g. MiniMax M3 / MiniMax M3 Free) are automatically disambiguated.
• Windows compatibility — sync temp directory now uses os.tmpdir() instead of hard-coded /tmp.

opencode plugin @fanfan4204/commandcode-go-opencode-provider

This installs the provider and registers all available models automatically.

Tip: For the latest model catalog (including reasoning effort variants), install the command-code CLI globally:

npm install -g command-code

Without it, the plugin falls back to the bundled models.json which may be outdated.

Run /connect in opencode, search for Command Code, and enter your API key:

/connect

Run /models to pick from available models:

/models

If you prefer to configure manually, add this to your opencode.json:

{
  "plugin": ["@fanfan4204/commandcode-go-opencode-provider/server"],
  "provider": {
    "commandcode": {
      "npm": "@fanfan4204/commandcode-go-opencode-provider",
      "name": "Command Code",
      "env": ["COMMANDCODE_API_KEY"]
    }
  },
  "model": "commandcode/deepseek-v4-flash"
}

The plugin auto-registers models at startup. It prefers the model catalog (including reasoning effort variants) from a locally installed command-code package, and falls back to the bundled models.json if that package is not found. You only need the provider.commandcode block — no need to list individual models.

Optional overrides:

• Env: COMMANDCODE_PACKAGE_PATH — path to the command-code package root (or its dist/index.mjs)
• Config file ~/.config/opencode/commandcode-go-opencode-provider.json:

  {
    "commandCodePackagePath": "C:/Users/you/AppData/Roaming/npm/node_modules/command-code",
    "disableModelSync": false
  }

Set COMMANDCODE_API_KEY instead of using /connect:

COMMANDCODE_API_KEY=your-key opencode

The full model list is maintained in models.json as a fallback. Upgrade your local/global command-code install to pick up new models and reasoning efforts on the next OpenCode start. Run bun run sync to refresh the committed models.json from local command-code (or npm latest with --remote).

git clone https://github.com/FanFan4204/opencode-commandcode-provider.git
cd opencode-commandcode-provider
bun install

For local testing, create opencode.local.json (gitignored) with file:// paths:

{
  "plugin": ["file:///path/to/@fanfan4204/commandcode-go-opencode-provider/server"],
  "provider": {
    "commandcode": {
      "npm": "file:///path/to/commandcode-go-opencode-provider",
      "name": "Command Code (local)",
      "env": ["COMMANDCODE_API_KEY"]
    }
  }
}

Run opencode --config opencode.local.json to test with your local build.

bun run sync              # update models.json from local command-code (npm latest if missing)
bun run sync -- --remote  # force download latest command-code tarball from npm
bun run sync:global       # update models.json + write to ~/.config/opencode/opencode.jsonc

At runtime OpenCode does not need bun run sync if command-code is installed locally or globally — the plugin scrapes that package on startup.

MIT