LLM Providers

LLM Providers

Koog 0.8.0 supports 8 LLM providers with a unified executor API. Each provider has a factory function that creates a PromptExecutor and a companion object with model constants.

Provider Quick Reference

OpenAI

import ai.koog.agents.ext.simple.simpleOpenAIExecutor
import ai.koog.agents.ext.llm.OpenAIModels

val apiKey = System.getenv("OPENAI_API_KEY")
val executor = simpleOpenAIExecutor(apiKey)

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = OpenAIModels.Chat.GPT4o
)

Available Models

Anthropic

import ai.koog.agents.ext.simple.simpleAnthropicExecutor
import ai.koog.agents.ext.llm.AnthropicModels

val apiKey = System.getenv("ANTHROPIC_API_KEY")
val executor = simpleAnthropicExecutor(apiKey)

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = AnthropicModels.Opus_4_1
)

Available Models

Google AI

import ai.koog.agents.ext.simple.simpleGoogleAIExecutor
import ai.koog.agents.ext.llm.GoogleModels

val apiKey = System.getenv("GOOGLE_API_KEY")
val executor = simpleGoogleAIExecutor(apiKey)

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = GoogleModels.Gemini2_5Pro
)

Available Models

DeepSeek

import ai.koog.agents.ext.llm.deepseek.DeepSeekLLMClient
import ai.koog.agents.ext.llm.DeepSeekModels

val apiKey = System.getenv("DEEPSEEK_API_KEY")
val client = DeepSeekLLMClient(apiKey)

val agent = AIAgent(
    promptExecutor = client,
    llmModel = DeepSeekModels.DeepSeekChat
)

Available Models

OpenRouter

OpenRouter provides access to multiple LLM providers through a single API.

import ai.koog.agents.ext.simple.simpleOpenRouterExecutor
import ai.koog.agents.ext.llm.OpenRouterModels

val apiKey = System.getenv("OPENROUTER_API_KEY")
val executor = simpleOpenRouterExecutor(apiKey)

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = OpenRouterModels.GPT4o
)

Available Models

AWS Bedrock

import ai.koog.agents.ext.simple.simpleBedrockExecutorWithBearerToken
import ai.koog.agents.ext.llm.BedrockModels

val bearerToken = System.getenv("BEDROCK_API_KEY")
val executor = simpleBedrockExecutorWithBearerToken(bearerToken)

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = BedrockModels.AnthropicClaude4_5Sonnet
)

Available Models

Alternative: AWS Credentials

import ai.koog.agents.ext.simple.simpleBedrockExecutorWithCredentials

val executor = simpleBedrockExecutorWithCredentials(
    accessKeyId = System.getenv("AWS_ACCESS_KEY_ID"),
    secretAccessKey = System.getenv("AWS_SECRET_ACCESS_KEY"),
    region = "us-east-1"
)

Mistral AI

import ai.koog.agents.ext.simple.simpleMistralAIExecutor
import ai.koog.agents.ext.llm.MistralAIModels

val apiKey = System.getenv("MISTRAL_API_KEY")
val executor = simpleMistralAIExecutor(apiKey)

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = MistralAIModels.Chat.MistralMedium31
)

Available Models

Ollama (Local)

Ollama runs models locally — no API key required.

import ai.koog.agents.ext.simple.simpleOllamaAIExecutor
import ai.koog.agents.ext.llm.OllamaModels

// Default: http://localhost:11434
val executor = simpleOllamaAIExecutor()

// Custom endpoint
val executor = simpleOllamaAIExecutor(baseUrl = "http://192.168.1.100:11434")

val agent = AIAgent(
    promptExecutor = executor,
    llmModel = OllamaModels.Meta.LLAMA_3_2
)

Available Models

Multi-Provider Support

Use MultiLLMPromptExecutor to wrap multiple providers and switch between them:

import ai.koog.agents.ext.llm.MultiLLMPromptExecutor
import ai.koog.agents.ext.llm.LLMClient

val openAIClient: LLMClient = simpleOpenAIExecutor(openAiKey)
val anthropicClient: LLMClient = simpleAnthropicExecutor(anthropicKey)

val multiExecutor = MultiLLMPromptExecutor(openAIClient, anthropicClient)

// Switch providers at runtime
val agent = AIAgent(
    promptExecutor = multiExecutor,
    llmModel = OpenAIModels.Chat.GPT4o
)

// Later, switch to Anthropic
agent.switchModel(AnthropicModels.Sonnet_4)

Provider Selection Guide