Manual Setup

The following section describes how to use the Arc Framework in any application.

For some examples, goto https://github.com/eclipse-lmos/arc/tree/main/examples/src/main/kotlin

Defining Agents

The following example shows how to use the agents function to define Arc Agents.

The agents function provides many default values that can be overridden to customize the behavior of the framework.

See the section on Loading LLM Clients for more information on how to load the correct LLM clients.

fun main() = runBlocking {
    // Set the OpenAI API key as a system property or environment variable.
    System.setProperty("OPENAI_API_KEY", "****")

    val agents = agents(
        functions = {
            function(name = "get_weather", description = "Returns the current weather.") {
                "the weather is sunny!"
            }
            // add more functions here
        }
    ) {
        agent {
            name = "MyAgent"
            model { "gpt-4o" }
            tools { +"get_weather" }
            prompt {
                """
                You are a weather assistant. Help the user with their questions about the weather.
                """
            }
        }
        // add more agents here
    }

    val reply = agents.getChatAgent("MyAgent").ask("What is the weather like?").getOrNull()
    println(reply)
}

The agents function provides many defaults that can be overridden to customize the behavior of the framework.

Loading LLM Clients

LLM Clients are loaded automatically based on the libraries on the classpath and the variables set.

The following variables are used to load the LLM Clients:

Name	Description	Values	Required
ARC_CLIENT	The client library to use.	azure, openai, ollama, gemini, bedrock	yes
ARC_MODEL_ALIAS	An alias for the model name. If set, this value must be reference on the Agent DSL.		no
ARC_MODEL	The name of the model.		only if ARC_MODEL_ALIAS is set.
ARC_AI_URL	The endpoint of the model.		not required for openai nor ollama.
ARC_AI_KEY	The api key to access the model.
ARC_AI_ACCESS_KEY	The access key to access the model.		usually required for bedrock.
ARC_AI_ACCESS_SECRET	The access secret to access the model.		usually required for bedrock.

OPENAI_API_KEY can also be used to create an OpenAI client.

Multiple clients can be defined by appending an index to the variable name, for example: ARC_CLIENT[0], ARC_CLIENT[1], etc.. Max number of clients is 10.

Variables are resolved in the following order:

1. System properties
2. Environment variables
3. home/.arc/arc.properties

Dependencies

The following shows the dependencies required to use the Arc Framework in your project.

Base dependencies:

    implementation("org.eclipse.lmos:arc-agents:$arcVersion")

Azure / Open AI

    implementation("org.eclipse.lmos:arc-azure-client:$arcVersion")

Gemini

    implementation("org.eclipse.lmos:arc-langchain4j-client:$arcVersion")
    implementation("dev.langchain4j:langchain4j-google-ai-gemini:$langchain4jVersion")

Ollama

    implementation("org.eclipse.lmos:arc-langchain4j-client:$arcVersion")
    implementation("dev.langchain4j:langchain4j-ollama:$langchain4jVersion")

Bedrock

    implementation("org.eclipse.lmos:arc-langchain4j-client:$arcVersion")
    implementation("dev.langchain4j:langchain4j-bedrock:$langchain4jVersion")