Fx with Nbility

This guide covers the Vercel Labs Fx coding agent. Named connections use OpenAI Chat Completions and must be defined in user-level settings.

Configure the connectionCopy link

Merge into ~/.fx/settings.json:

{
  "provider": "nbility",
  "providers": {
    "nbility": {
      "protocol": "openai-chat-completions",
      "base_url": "https://api.nbility.ai/v1",
      "auth": { "type": "bearer", "env": "NBILITY_API_KEY" },
      "model_metadata": {
        "YOUR_MODEL_ID": {
          "context_window": 8192,
          "max_output_tokens": 1024,
          "supports_tool_use": true
        }
      }
    }
  },
  "models": { "nbility": "YOUR_MODEL_ID" }
}

Replace both model IDs consistently. Adjust the illustrative limits to the actual model and enable tool support only when supported. A known context window enables automatic compaction.

Launch and verifyCopy link

export NBILITY_API_KEY="YOUR_API_KEY"
fx provider nbility
fx status --json
fx ask "Reply with one short sentence."

Status should show the selected endpoint and model. Verify the request in Usage logs.

Configuration boundariesCopy link

Project .fx.json cannot define or select custom connections. Keep the connection at the top level of ~/.fx/settings.json. The client appends /chat/completions to base_url.

Custom connections do not automatically provide Vercel AI Gateway search, credit checks, or vision fallback. Set supports_vision only for a vision-capable model. Fx’s execution permission settings remain separate.

ReferencesCopy link

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