AzureOpenAIModelProvider

Namespace AgentPrism · Assembly AgentPrism.Azure.dll

The IModelProvider implementation for Azure OpenAI.

public sealed class AzureOpenAIModelProvider : IModelProvider, IModelProviderHealthCheck, IModelProviderConfigurationDiagnostics

Inheritance

object ← AzureOpenAIModelProvider

Implements

IModelProvider, IModelProviderHealthCheck, IModelProviderConfigurationDiagnostics

Inherited Members

object.GetType(), object.ToString(), object.Equals(object?), object.Equals(object?, object?), object.ReferenceEquals(object?, object?), object.GetHashCode()

Remarks

The provider does not reject a deployment name absent from the catalog. Deployment names are chosen by whoever sets up the Azure resource, and AgentPrism’s configuration is not expected to be updated when a new deployment is opened; when the catalog is non-empty, only an informational log entry is left.

Circuit-breaker and content-filter detection live outside this type, at the ModelProviderRegistry level; this package gets both for free. When Azure’s own content filter cuts a response short, ContentFilterDetectingChatClient records it as content_filtered — there is no extra code in this package for it.

Constructors

AzureOpenAIModelProvider(string, AzureOpenAIChatClientFactory, IReadOnlyList<ModelDescriptor>, ILogger<AzureOpenAIModelProvider>?, AzureOpenAIProviderOptions?, EgressSocketGuard?)

Creates a new provider.

public AzureOpenAIModelProvider(string name, AzureOpenAIChatClientFactory chatClientFactory, IReadOnlyList<ModelDescriptor> models, ILogger<AzureOpenAIModelProvider>? logger = null, AzureOpenAIProviderOptions? healthCheckOptions = null, EgressSocketGuard? egressGuard = null)

Parameters

name string

The provider name. ModelBinding.Provider in agent definitions matches this value.

chatClientFactory AzureOpenAIChatClientFactory

The chat client factory.

models IReadOnlyList<ModelDescriptor>

The deployments this provider offers.

logger ILogger<AzureOpenAIModelProvider>?

The logger.

healthCheckOptions AzureOpenAIProviderOptions?

When given, AzureOpenAIModelProvider.CheckHealthAsync calls the GET {endpoint}/openai/models endpoint using this option’s address and credential. When null, health status always returns ModelProviderHealthStatus.Unknown.

egressGuard EgressSocketGuard?

The outbound network guard. Attached only to a client built from a tenant-supplied endpoint override; when null, such an override is not guarded.

Exceptions

ArgumentNullException

One of the required dependencies is null.

ArgumentException

name is empty.

Properties

Models

The models this provider offers.

public IReadOnlyList<ModelDescriptor> Models { get; }

Property Value

IReadOnlyList<ModelDescriptor>

Name

The provider name. ModelBinding.Provider matches this value. Comparison is case-insensitive.

public string Name { get; }

Property Value

string

Methods

CheckHealthAsync(CancellationToken)

Checks the provider’s reachability.

public ValueTask<ModelProviderHealth> CheckHealthAsync(CancellationToken cancellationToken = default)

Parameters

cancellationToken CancellationToken

The cancellation token.

Returns

ValueTask<ModelProviderHealth>

The check result.

CreateChatClient(ModelBinding, ModelProviderCredential?)

Produces a raw chat client for the given binding.

public IChatClient CreateChatClient(ModelBinding binding, ModelProviderCredential? credential = null)

Parameters

binding ModelBinding

The model binding.

credential ModelProviderCredential?

A resolved per-tenant credential (BYOK). When null, the provider’s own setup-time credential is used and behavior is identical to a setup without per-tenant credentials. When given, the provider builds (or reuses a cached) client using ModelProviderCredential.ApiKey and, if present, ModelProviderCredential.Endpoint, instead of its setup-time credential.

Returns

IChatClient

The provider-specific client. Decorators specific to the provider (example: Anthropic’s settings decorator) may be added here.

Remarks

Do not build the common pipeline here. UseFunctionInvocation, UseOpenTelemetry, the content guard, the circuit breaker, and extra resolution are added by ModelProviderRegistry.CreateChatClient. Before that shared pipeline, every provider package built the tool-call loop inside itself; the result was that no ring the registry wraps around could see the loop’s turns — a tool result entered the model uninspected.

If the loop is also built here, two nested FunctionInvokingChatClient instances form: the inner one resolves tools, the outer one never sees any call. The damage is not functional but measurable (double wrapping, a misleading span tree).

GetConfigurationDiagnostic()

Returns the resolution status of the configuration key required by this provider.

public ConfigurationDiagnostic? GetConfigurationDiagnostic()

Returns

ConfigurationDiagnostic?

null when the provider was not configured with an options object.