AgentRunKit

Open your Package.swift file and add AgentRunKit as a dependency pointing to the GitHub repository. Then add it to your target's dependency list.

// Package.swift
.package(url: "https://github.com/Tom-Ryder/AgentRunKit.git", from: "4.3.0")

In the targets section of Package.swift, add AgentRunKit (and optionally AgentRunKitMLX for on-device inference or AgentRunKitFoundationModels for Apple Foundation Models) to your target's dependencies.

.target(
    name: "YourApp",
    dependencies: [
        "AgentRunKit",
        // Optional: "AgentRunKitMLX",
        // Optional: "AgentRunKitFoundationModels"
    ]
)

Initialize a provider client with your API key. AgentRunKit supports seven providers. Set OPENAI_API_KEY (or the equivalent for your provider) in your environment or pass it directly to the client constructor.

import AgentRunKit

let client = AnthropicClient(apiKey: ProcessInfo.processInfo.environment["ANTHROPIC_API_KEY"] ?? "")

Create tools using the Tool generic type. AgentRunKit validates parameters at compile time against a Codable struct, generating JSON schemas automatically. This eliminates runtime schema errors.

struct WeatherParams: Codable {
    let city: String
}

let weatherTool = try Tool<WeatherParams, String, EmptyContext>(
    name: "get_weather",
    description: "Get the current weather for a city"
) { params, _ in
    return "72°F and sunny in \(params.city)"
}

Instantiate an Agent with your client and tools, then call run() with a user message. The agent handles tool calls, streaming, and multi-turn conversations automatically.

let agent = Agent(client: client, tools: [weatherTool])
let result = try await agent.run(
    userMessage: "What's the weather in San Francisco?",
    context: EmptyContext()
)
print(result)

For privacy-sensitive workloads on Apple Silicon, swap in the MLX client instead of a cloud provider. Add the AgentRunKitMLX module and initialize the MLX client with your local model path.

import AgentRunKitMLX

let client = MLXClient(modelPath: "/path/to/local/model")
let agent = Agent(client: client, tools: [weatherTool])