Crush with Nbility

Crush connects through an openai-compat provider. Add it alongside existing providers to retain your other services.

Configure the providerCopy link

Set export NBILITY_API_KEY="YOUR_API_KEY" in the launch terminal and edit ~/.config/crush/crush.json. If you use an XDG configuration directory, use the path Crush actually loads.

{
  "providers": {
    "nbility": {
      "name": "Nbility",
      "type": "openai-compat",
      "base_url": "https://api.nbility.ai/v1",
      "api_key": "$NBILITY_API_KEY",
      "models": [
        {
          "id": "YOUR_MODEL_ID",
          "name": "Nbility model",
          "context_window": 8192,
          "default_max_tokens": 1024
        }
      ]
    }
  }
}

Copy an ID from Models. The values 8192 and 1024 are illustrative limits, not model specifications. Adjust them to the actual context and output limits before longer tasks.

Select the modelCopy link

Start Crush and select the Nbility entry in its model picker. If you configure both large and small model roles, review the provider for each role.

Verify and troubleshootCopy link

Test plain text first and check Usage logs, then try file or terminal tools.

• Missing provider: check JSON syntax, configuration location, and your release’s schema.
• Authentication failure: ensure the launching process can read NBILITY_API_KEY.
• Frequent compaction: review context_window; never raise it beyond the actual model limit.
• 404: end base_url at /v1, without /chat/completions.

ReferencesCopy link

Adapted and rewritten for Nbility from aicentos’s Crush guide, licensed under CC BY 4.0. Its account, plan, and proprietary service instructions do not apply here. See the client’s official documentation.