@pandada8/opencode-axonhub

OpenCode v2 plugin that discovers AxonHub models from /v1/models and /v1/models?include=all, merges both responses, and exposes them as the axonhub provider through the catalog.

It targets the opencode v2 plugin format: the module default-exports a definition with an id and a setup(ctx) function (see @opencode/plugin). Models are cached at ~/.cache/opencode/axonhub-models.json for one day. If no base URL or API key is configured, no models are registered.

Usage

Add the plugin and an axonhub provider entry to opencode.jsonc. Provide the endpoint and API key as plugin options, provider options, or environment variables.

{
  "plugin": [["@pandada8/opencode-axonhub", { "baseURL": "https://your-axonhub.example.com", "apiKey": "ah-..." }]],
  "provider": {
    "axonhub": {
      "options": {
        "baseURL": "https://your-axonhub.example.com"
      },
      "models": {}
    }
  }
}

Alternatively, set the key via environment so it never lands in config:

export AXONHUB_BASE_URL="https://your-axonhub.example.com"
export AXONHUB_API_KEY="ah-..."

Resolution order for the endpoint/key: plugin options (baseURL/apiKey) -> configured provider options (provider.axonhub.options.baseURL / apiKey) -> environment (AXONHUB_BASE_URL, AXONHUB_API_KEY). If neither is found, the plugin registers nothing.

Routing

Each model is routed to the vendor protocol AxonHub expects through its per-model SDK package and endpoint:

• OpenAI models -> @ai-sdk/openai against /v1
• Gemini/Google -> @ai-sdk/google against /gemini/v1beta
• everything else -> @ai-sdk/anthropic against /anthropic/v1

Because opencode v2 stores the endpoint per model (there is no per-model api.url), the plugin sets each model's settings.baseURL accordingly, while exposing all of them under the single axonhub provider.

Notes

• Model enrichment from OpenCode's ~/.cache/opencode/models.json and experimental.modes (*-fast variants) are not yet ported to the v2 API.
• Requires opencode v2 (tested against 2.0.18).