Azure OpenAI

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": "azure-gpt",
  "adapterKey": "azureopenai",
  "upstreamModel": "gpt-5.4",
  "credentials": { "apiKey": "...", "baseUrl": "https://my-resource.openai.azure.com" }
}

Behavior and limitations

• Credentials: apiKey, baseUrl.
• baseUrl must be HTTPS, and can be either the bare resource endpoint (https://my-resource.openai.azure.com) or the full /openai/v1 base. Legacy /openai/deployments/... URLs and query parameters are rejected. Version selectors belong in credentials.apiVersion, not in baseUrl.
• Your Azure deployment name must equal the catalog model id (upstreamModel) — Azure resolves the model by deployment name, not by a separate model parameter.
• Audio transcription catalog ids include gpt-transcribe, gpt-4o-transcribe, and gpt-4o-mini-transcribe. They use the classic deployment-based Audio API by default because some Azure resources return DeploymentNotFound from the v1 preview route for otherwise valid deployments. The classic route defaults to API version 2024-06-01, does not support streaming, and accepts credentials.apiVersion when the deployment requires another version. Custom deployment names require an inline catalogEntry.
• To opt into /openai/v1/audio/transcriptions?api-version=preview, set transportOverrides["audio.transcribe"] to audio_transcriptions. The v1 transport sends the Azure deployment name as multipart model.
• Default transport: Responses API for every public text endpoint, including Chat Completions and Messages; embeddings for embeddings. Set transportOverrides["text.generate"] to chat_completions to opt into Azure Chat Completions. The public endpoint does not select the upstream transport. No native image transport today.
• Strict function tools are emitted on both Chat Completions and Responses. Azure requires parallel_tool_calls: false for this mode, so the adapter supplies that default and rejects an explicit true rather than weakening the guarantee.
• Azure OpenAI and Azure Foundry are two independent adapters and catalogs, even though both live under the Azure umbrella — use this one for models you deploy through Azure OpenAI resources specifically.
• Catalog: src/adapters/azureopenai/catalog.json.

Next steps

• Azure Foundry — the other Azure adapter, for models Azure itself sells.
• Creating deployments — the full request shape.