Built-in Providers
Built-in providers are the ready-made catalog the AI Gateway ships with: OpenAI, Anthropic, and Google today. Unlike a Custom Provider, you don't register these yourself. You just bring your own API key for the ones you want to use.
Why the base URLs point at OpenAI-compatible endpoints
Since your client is always speaking OpenAI's wire format to the gateway, each built-in provider is configured to forward to whichever endpoint of the real provider also speaks OpenAI format, rather than that provider's native API.
| Provider | Base URL | Notes |
|---|---|---|
| OpenAI | https://api.openai.com/v1 |
Native. Already speaks OpenAI format. |
| Anthropic | https://api.anthropic.com/v1 |
Anthropic's official OpenAI-compatible layer. See the caveat below before relying on it for advanced features. |
https://generativelanguage.googleapis.com/v1beta/openai |
Google's official OpenAI-compatible endpoint. Uses Authorization: Bearer <gemini-api-key>, the same shape as the other two, even though Gemini's native API normally uses a query-parameter key. |
Anthropic's OpenAI-compatible layer has real limits
Anthropic's OpenAI-compatible endpoint isn't their full-featured native Messages API. Strict function-calling schemas aren't enforced, there's no audio input, no prompt caching, and temperature is capped at 1.0. It's fine for basic chat completions, but if you need the full Claude feature set, use Anthropic's native API or SDK, instead of using the OpenAI-compatible endpoint.
Keeping the catalog current
SocketXP maintains the list of supported built-in providers and models on your behalf.
There's nothing for you to configure here beyond registering a BYOK key: if a provider or
model you need isn't listed yet, reach out to [email protected].
Model ids follow the provider:creator/model convention, for example
google:google/gemini-3.7-flash. That naming is specific to this gateway's own catalog,
not something borrowed from any provider's real API. When you call a model, you don't
have to type the full form — the shorter creator/model form, or just the bare model
name on its own, both work just as well, as long as the name isn't ambiguous across more
than one provider.
Catalog freshness
The built-in model catalog is only as current as whoever operates the gateway keeps
it. It isn't automatically refreshed from each provider's own release feed, so if
you're relying on a specific model name, it's worth confirming with [email protected] and it's still current
rather than assuming it always will be.
Registering a BYOK key
Go to AI Providers → BYOK Keys in the SocketXP Web Portal and add your API key for the provider you want to use. The key is shown back to you masked, with only the last 4 characters visible, from then on.
You can register more than one key for the same provider, for instance a personal key
alongside a team key. Keep in mind that the gateway always routes using the oldest key on
file for that provider; there's no per-request key selection or load balancing across
keys today (will be supported in the future, please write to: [email protected]). If you need to rotate a key, add the new one and delete the old one.