OpenAI is the second paid provider — GPT-4o family models, generally cheaper per request than
Claude, a reasonable everyday default if you already have an OpenAI account.

1. Go to platform.openai.com/api-keys and sign in.
2. Create a new secret key (starts with sk-).
3. Add billing if you haven't already — OpenAI also requires a funded account.

Same two entry points as every provider: the first-run setup wizard, or **Settings → API
Keys** afterward. The wizard validates the key with a cheap request before continuing so you find
out immediately if it was rejected, rather than on your first real chat. Keys are encrypted at
rest and never leave your own install.

| Model | Label |
|---|---|
| gpt-4o | GPT-4o |
| gpt-4o-mini | GPT-4o mini |
| o1-mini | o1-mini |

As with every provider, the picker's custom model id field takes any OpenAI model id directly
— you're never limited to this shortlist.

OpenAI and NVIDIA share the same code path (lib/providers/openai.ts's OpenAICompat config) —
NVIDIA is really "OpenAI-compatible API, different base URL." That's also why a handful of
OpenAI-compatible quirks (like some backends rejecting the tools param) are handled with a
retry-without-tools fallback shared by both.

• NVIDIA (free tier) — same code path, no cost
• API key errors