Choose a provider
Every built-in adapter: credentials, default transport, and where to find its catalog.
Choose an adapter for the provider and operation you need, then create a deployment with its
adapterKey, upstream model, and credentials. The adapter handles protocol translation; its
catalog supplies known model capabilities and pricing.
Choose an adapter
The transport column describes the upstream protocol. Your application still uses AI Gateway's public API. OpenRouter currently supports reranking only.
| Provider | Adapter key | Credentials | Default transport |
|---|---|---|---|
| OpenAI | openai | apiKey | Responses API |
| Azure OpenAI | azureopenai | apiKey, baseUrl | Responses API |
| Azure Foundry | azurefoundry | apiKey, baseUrl | Chat Completions |
| Google AI Studio | googleaistudio | apiKey | generateContent |
| Anthropic | anthropic | apiKey | Messages API |
| DeepSeek | deepseek | apiKey | Chat Completions |
| MiniMax | minimax | apiKey | Chat Completions |
| Moonshot | moonshot | apiKey | Chat Completions |
| Z.AI | zai | apiKey | Chat Completions |
| Vercel AI Gateway | vercel | apiKey | Responses API |
| OpenRouter | openrouter | apiKey | Reranking |
| OpenAI-compatible | openaicompatible | apiKey, baseUrl | Chat Completions |
Configure the connection
Adapters with a built-in upstream host do not require baseUrl. openaicompatible, Azure OpenAI,
and Azure Foundry require baseUrl because the operator chooses the target endpoint.
GET /admin/operations returns this same adapter list live, including required credential keys and
which operations/transports each adapter actually implements.
Next steps
- Creating deployments — the request shape shared by every provider.
- Model catalog — how a provider's models are declared.
- First deployment — the minimal end-to-end example.