Model and provider

This page covers model-provider execution, selected through the model: constructor argument or UseModel(...). The provider/model string selects the provider and model. Configure the connection before testing a response.

Local CLI agents use different configuration: Codex CLI and Claude CLI require an explicit CLI model in their configuration callbacks. See the Claude guide for callback API availability. Both use the host account's CLI authentication rather than an agent apiKey argument. The provider registry and generation defaults below apply to model execution.

Configure your first connection

For local development, initialize user secrets in the ASP.NET Core project:

dotnet user-secrets init
dotnet user-secrets set "OpenAI:ApiKey" "YOUR_API_KEY"

Use the key from configuration when creating the agent:

using Runiq.AI.Agents;

var agent = new Agent(
    id: "support-agent",
    name: "Support Agent",
    instructions: "Help users understand the application. State uncertainty clearly.",
    model: "openai/gpt-5",
    apiKey: builder.Configuration["OpenAI:ApiKey"]);

Place this after creating builder and register the definition with options.AddAgent(agent). The model is an example; use one available in your provider account. User secrets are for development; use your deployment's secret configuration in production.

Understand the model reference

In openai/gpt-5, openai selects the provider strategy and gpt-5 is passed to its client. Runiq validates the provider name, but model availability and supported parameters depend on the provider.

For a locally running Ollama server, use a model you have installed:

model: "ollama/llama3.2"

The default Ollama connection does not require an API key.

Provider reference

These are the defaults defined by Runiq's provider registry.

Override connection options

ProviderOptions belongs to Runiq.AI.Core.Configuration. Pass it as the agent's provider argument:

using Runiq.AI.Core.Configuration;

var provider = new ProviderOptions
{
    Url = "https://api.openai.com/v1",
    Timeout = TimeSpan.FromSeconds(60)
};

Url overrides the default endpoint. Timeout handling depends on the provider client implementation; the option should not be treated as a universal deadline. Pass a cancellation token when running the agent.

Tune response behavior

These are Runiq's accepted configuration values. Their effect depends on the selected provider and model; they do not guarantee identical behavior across providers.

Check connection failures

If execution reports ApiKeyMissing, confirm the key is present in the running host's configuration. For rejected model requests, check the model identifier, endpoint, credentials, and provider response.

Next: Run the registered agent →