OpenAI-compatible

Use this adapter for an API that implements OpenAI's Chat Completions (or compatible Images/Embeddings) contract but isn't one of the named providers — self-hosted vLLM/SGLang deployments, aggregators, or a provider Bifrost doesn't have a dedicated adapter for yet.

Configure a deployment

Send this body to POST /admin/deployments with an operator credential. Replace the example credentials and choose a model available to your provider account. See Creating deployments for the shared configuration fields.

{
  "publicModel": "my-model",
  "adapterKey": "openaicompatible",
  "upstreamModel": "some-vendor/some-model",
  "credentials": { "apiKey": "...", "baseUrl": "https://api.vendor.example/v1" }
}

Behavior and limitations

• Credentials: apiKey, baseUrl (both required — there's no built-in default host).
• Default transport: Chat Completions for text, images for image generate/edit, videos for OpenAI-shaped video APIs, and embeddings for embeddings.
• Compatibility goal: this adapter covers the common OpenAI-shaped surface and stays extensible through catalogEntry, transportOverrides, and credentials.headers. It does not promise perfect behavior for every aggregator quirk or proprietary OpenAI feature.
• Strict tools: declare capabilities.strictTools: true only when the target itself guarantees schema-adherent function arguments. The adapter forwards each tool's strict flag natively and rejects strict requests when the catalog does not make that guarantee.
• Since the gateway can't know this model's real capabilities, catalogEntry is required — declare operations, capabilities, and parameters by hand. See Model catalog and the custom-model example in Creating deployments.
• Aggregators and hosted compatible APIs use this same adapter with an explicit baseUrl. If an operation needs a provider-specific transport shape, set transportOverrides explicitly, such as image.generate: "chat_completions" or video.generate: "videos_async".
• If the target speaks vLLM/SGLang-style thinking controls (chat_template_kwargs), declare reasoning.kind: "chat_template_flag" in the custom catalog entry — see Reasoning.

Next steps

• Creating deployments — the full custom-model example.
• Model catalog — the catalog entry a custom model must declare.