Azure Foundry

{
  "publicModel": "azure-deepseek",
  "adapterKey": "azurefoundry",
  "upstreamModel": "DeepSeek-V3.1",
  "credentials": { "apiKey": "...", "baseUrl": "https://my-foundry-resource.services.ai.azure.com" }
}

• Credentials: apiKey, baseUrl.
• Use this adapter — not Azure OpenAI — for models Azure itself sells through Foundry, such as DeepSeek-V3.1 or grok-4.3. Their upstream transport is Chat Completions; Bifrost still renders every public text contract (Chat Completions, Responses, Messages) over it.
• Default transport: Chat Completions for text. No native image/embedding transport.
• Two independent providers, two independent catalogs: a model available on both Azure OpenAI and Azure Foundry needs a separate catalog entry (and deployment) per adapter.
• Catalog: src/adapters/azurefoundry/catalog.json.

What to read next

• Azure OpenAI — the other Azure adapter.
• Creating deployments — the full request shape.