Providers and vendors

The generated catalog is the source of truth for provider rows. Docs can name the adapter families, but the catalog decides what is available in a build. Use:

GET /v1/models
POST /v1/models/search

Each model row includes a provider and public model ID. The engine resolves that row, computes the provider wire model with upstreamModelId ?? modelId, and calls the matching adapter.

Adapter families

The API includes adapters for providers and OpenAI-compatible families such as:

Provider or family	Used for
OpenAI	Chat, responses, embeddings, images, audio, video.
Azure OpenAI	Chat, embeddings, images.
Gemini and Vertex	Text, embeddings, images, speech, video, native Google wire formats.
Alibaba	Text, embeddings, images, speech, transcription, video.
Cloudflare	Workers AI model rows, including text, image, speech, transcription.
Fireworks	OpenAI-compatible text and selected media routes.
ASI Cloud	OpenAI-compatible text rows.
Deepgram, ElevenLabs, Cartesia	Speech and transcription.
Roboflow	Vision and computer-vision outputs such as detections and segmentation.

Provider support comes from source data and adapter code. If the catalog row exists and the adapter supports that modality, the route can execute it.

Metering External OpenAI-compatible clients

Overview

External Use

Families

Modalities

Guides

Adapter families

​Adapter families

​Related

Adapter families

Related