Connections

Caution

Remember to always store secrets like api-key in UserSecrets, Environment Variables, Key Vaults, or similar; NEVER directly in the code!

What are connections?

AgentFactories and EmbeddingFactories need credentials to authenticate against the underlying services.

Often, such credentials are a single api-key (or an endpoint + api-key) that can be given to a factory:

//Sample: Creating an OpenAI Agent Factory
OpenAIAgentFactory agentFactory = new OpenAIAgentFactory("<api-key>");

Advanced/Provider Specific options

In more advanced cases, you might need additional options for the connection, such as timeouts, endpoints, in which case you make a provider-specific Connection object:

//Sample: Creating an OpenAI Connection
OpenAIConnection connection = new OpenAIConnection
{
    ApiKey = "<api-key>",
    NetworkTimeout = TimeSpan.FromMinutes(5), //Set call timeout
});

//Pass connection to Factory instead of simple api-key
OpenAIAgentFactory agentFactory = new OpenAIAgentFactory(connection);

All Options on connections

Here is a list of all Properties on the various connection objects

Property	Description & Notes	Supported Providers
ApiKey	The API-Key to authenticate against the LLM - GitHub uses AccessToken instead - AzureOpenAI offer an RBAC alternative - AmazonBedrock uses a bearer token applied via AWS_BEARER_TOKEN_BEDROCK (if not already defined)	Most
AccessToken	GitHub Models access token (fine-grained PAT with Models access)	GitHub
Region	AWS region for Amazon Bedrock Runtime (RegionEndpoint)	AmazonBedrock
NetworkTimeout	Timeout Settings (if your LLM Calls take longer than the default HTTP Timeout)	All
DefaultClientType	Control if Agents created should use ChatClient or Responses API protocol if not specified specifically by the Agent - Default is ChatClient	AzureOpenAI, OpenAI, OpenRouter, XAI
Endpoint	Control the Endpoint the LLM Talk to - Mandatory for AzureOpenAI; for the rest, this is optional - Can be used in OpenAI to implement 3rd party OpenAI providers	AzureOpenAI, OpenAI, OpenRouter, XAI, Anthropic
AdditionalOpenAIClientOptions	Option to set additional OpenAIClientOptions - Used for not out of the box scenarios for this toolkit	OpenAI, OpenRouter, XAI
Credentials (TokenCredential)	Azure Role Based Access Control - Use Standard Azure RBAC like example AzureCliCredential	AzureOpenAI
AutoCorrectFoundryEndpoint	Autocorrect Foundry project URLs like https://[name].services.ai.azure.com/api/projects/[project] to https://[name].services.ai.azure.com (Default = true)	AzureOpenAI
AdditionalAzureOpenAIClientOptions	Option to set additional AzureOpenAIClientOptions - Used for not out of the box scenarios for this toolkit	AzureOpenAI
AdditionalAzureAIInferenceClientOptions	Option to set additional AzureAIInferenceClientOptions - Used for not out of the box scenarios for this toolkit	GitHub
VertexAI	If set, uses Vertex AI APIs instead of Gemini API	Google
Credential (ICredential)	Vertex-only credentials	Google
Project	Vertex-only project id	Google
Location	Vertex-only location	Google
HttpOptions	Advanced Google HTTP options	Google