Providers Overview

A provider is the LLM backend your agent talks to. Every agent needs exactly one provider configured — either at build time, or switched per-run via AgentRunConfig.

Available Providers

Provider Key	NuGet Package	Builder Method	Notes
openai	HPD-Agent.Providers.OpenAI	.WithOpenAI(...)	GPT-4o, o1, audio
azure-openai	HPD-Agent.Providers.OpenAI	.WithAzureOpenAI(...)	Traditional Azure OpenAI endpoints
anthropic	HPD-Agent.Providers.Anthropic	.WithAnthropic(...)	Claude models
google-ai	HPD-Agent.Providers.GoogleAI	.WithGoogleAI(...)	Gemini models
azure-ai	HPD-Agent.Providers.AzureAI	.WithAzureAI(...)	Azure AI Foundry / Projects
azure-ai-inference	HPD-Agent.Providers.AzureAIInference	.WithAzureAIInference(...)	Azure AI Inference endpoint
ollama	HPD-Agent.Providers.Ollama	.WithOllama(...)	Local models via Ollama
mistral	HPD-Agent.Providers.Mistral	.WithMistral(...)	Mistral AI models
bedrock	HPD-Agent.Providers.Bedrock	.WithBedrock(...)	AWS Bedrock
huggingface	HPD-Agent.Providers.HuggingFace	.WithHuggingFace(...)	HuggingFace Inference API
openrouter	HPD-Agent.Providers.OpenRouter	.WithProvider("openrouter", ...)	OpenRouter (100+ models)
onnx-runtime	HPD-Agent.Providers.OnnxRuntime	.WithOnnxRuntime(...)	Local ONNX models

---

Two Ways to Configure a Provider

1. Builder methods — code-first, IntelliSense-driven

Use the provider-specific builder method when you're writing config in code. You get full IntelliSense on every option, typed validation at BuildAsync, and a clientFactory to wrap the underlying IChatClient.

var agent = await new AgentBuilder()
    .WithOpenAI(
        model: "gpt-4o",
        configure: opts =>
        {
            opts.Temperature = 0.7f;
            opts.MaxOutputTokenCount = 4096;
        })
    .BuildAsync();

For basic setup where you don't need provider-specific options, .WithProvider works too:

var agent = await new AgentBuilder()
    .WithProvider("openai", "gpt-4o")
    .BuildAsync();

Both require the provider's NuGet package — the package's [ModuleInitializer] registers the provider automatically on assembly load.

2. JSON config — file-driven, fully serializable

Use AgentConfig + ProviderOptionsJson when the config lives outside your code — an agent.json file, environment config, FFI, or any scenario where you want a single file to describe the complete agent.

agent.json:

{
    "name": "MyAgent",
    "systemInstructions": "You are a helpful assistant.",
    "provider": {
        "providerKey": "openai",
        "modelName": "gpt-4o",
        "apiKey": "sk-...",
        "providerOptionsJson": "{\"temperature\":0.7,\"maxOutputTokenCount\":4096}"
    }
}

var agent = await AgentConfig.BuildFromFileAsync("agent.json");

ProviderOptionsJson is a JSON string containing the provider-specific options. The keys are camelCase — they match the [JsonPropertyName] attributes on each provider's config class. See each provider's page for the full list of keys.

---

Switching Providers at Runtime

Both approaches above configure the default provider used when no override is specified. You can switch provider, model, or API key on a per-RunAsync basis without rebuilding the agent:

// Default run — uses the agent's configured provider
await foreach (var evt in agent.RunAsync("Hello")) { }

// This run — switches to a different model
await foreach (var evt in agent.RunAsync("Hello", runConfig: new AgentRunConfig
{
    ProviderKey = "anthropic",
    ModelId = "claude-opus-4-6"
})) { }

All providers referenced at runtime must have their NuGet package referenced in the project (the [ModuleInitializer] registers them at load time).

→ See Run Config for all runtime override options.

---

Provider Pages

Each provider page covers:

• Installation and authentication
• Builder method reference
• Full JSON config reference (ProviderOptionsJson keys)
• Provider-specific features and examples

→ OpenAI · Anthropic · Google AI · Azure AI · Azure OpenAI · Azure AI Inference · Ollama · Mistral · Bedrock · HuggingFace · ONNX Runtime