Predefined Strategies

Predefined Strategies

Koog provides two built-in agent strategies that cover the most common agent patterns.

Overview

Chat Agent Strategy

A conversational agent that processes user input, calls tools when needed, and returns responses. Enforces tool usage over plain-text answers.

Dependency

import ai.koog.agents.ext.agent.chatAgentStrategy

Kotlin Setup

import ai.koog.agents.ext.agent.chatAgentStrategy
import ai.koog.agents.core.agent.AIAgent
import ai.koog.agents.core.tools.ToolRegistry
import ai.koog.agents.ext.simple.simpleOpenAIExecutor
import ai.koog.agents.ext.llm.OpenAIModels

val chatAgent = AIAgent(
    promptExecutor = simpleOpenAIExecutor(System.getenv("OPENAI_API_KEY")),
    llmModel = OpenAIModels.Chat.GPT4o,
    strategy = chatAgentStrategy(),
    toolRegistry = ToolRegistry {
        tool(searchTool)
        tool(weatherTool)
    }
)

val result = chatAgent.run("What's the weather like today?")

Java Setup

import ai.koog.agents.ext.agent.AIAgentStrategies;
import ai.koog.agents.core.agent.AIAgent;
import ai.koog.agents.core.tools.ToolRegistry;
import ai.koog.agents.ext.simple.SimpleOpenAIExecutorKt;
import ai.koog.agents.ext.llm.OpenAIModels;

ToolRegistry toolRegistry = ToolRegistry.builder()
    .tools(new SearchAndWeatherTools())
    .build();

AIAgent<String, String> chatAgent = AIAgent.builder()
    .promptExecutor(SimpleOpenAIExecutorKt.simpleOpenAIExecutor(
        System.getenv("OPENAI_API_KEY")))
    .llmModel(OpenAIModels.Chat.GPT4o)
    .graphStrategy(AIAgentStrategies.chatAgentStrategy())
    .toolRegistry(toolRegistry)
    .build();

String result = chatAgent.run("What's the weather like today?");

How It Works

User Input → LLM Processing → Tool Call? → Execute Tool → Loop Back
                                     ↓ No
                                  Response

1. User sends a message
2. LLM processes the message with available tools
3. If LLM requests tool calls → execute tools → send results back to LLM
4. Repeat until LLM produces a final text response
5. Provides feedback if the LLM tries to respond with plain text instead of using tools

Best Use Cases

• Conversational agents requiring tool access
• Assistants performing actions on user requests
• Chatbots accessing external systems or data
• Situations enforcing tool usage over plain text

ReAct Strategy (Reasoning and Acting)

Alternates between reasoning and execution stages in a loop: Reason → Act → Observe → repeat until complete.

Dependency

import ai.koog.agents.ext.agent.reActStrategy

Kotlin Setup

import ai.koog.agents.ext.agent.reActStrategy
import ai.koog.agents.core.agent.AIAgent
import ai.koog.agents.core.tools.ToolRegistry
import ai.koog.agents.ext.simple.simpleOpenAIExecutor
import ai.koog.agents.ext.llm.OpenAIModels

val reActAgent = AIAgent(
    promptExecutor = simpleOpenAIExecutor(apiKey),
    llmModel = OpenAIModels.Chat.GPT4o,
    strategy = reActStrategy(
        reasoningInterval = 1,
        name = "banking_agent"
    ),
    toolRegistry = ToolRegistry {
        tool(getTransactions)
        tool(calculateSum)
    }
)

val result = reActAgent.run("How much did I spend last month?")

Java Setup

AIAgent<String, String> reActAgent = AIAgent.<String, String>builder()
    .promptExecutor(SimpleOpenAIExecutorKt.simpleOpenAIExecutor(
        System.getenv("OPENAI_API_KEY")))
    .llmModel(OpenAIModels.Chat.GPT4o)
    .graphStrategy(AIAgentStrategies.reActStrategy(1, "banking_agent"))
    .toolRegistry(ToolRegistry.builder()
        .tools(new BankingTools())
        .build())
    .build();

String result = reActAgent.run("How much did I spend last month?");

Parameters

How It Works

User Question
    ↓
┌─── Reason: Plan next step ───┐
│                               │
├─── Act: Call a tool ──────────┤
│                               │
├─── Observe: Tool result ──────┤
│                               │
└─── (repeat until solved) ─────┘
    ↓
Final Answer

Banking Example Trace

User: "How much did I spend last month?"

1. Reason: I need to get transactions, filter deposits, then calculate the sum
2. Act: get_transactions(startDate: "2025-05-19", endDate: "2025-06-18")
3. Observe: Returns transaction records including positive (deposits) and negative (spending)
4. Reason: I need to remove the salary deposit and sum remaining transactions
5. Act: calculate_sum(amounts: [-100.00, -500.00, -200.00])
6. Observe: Result: -800.00
7. Final: "You spent $800.00 last month on groceries, rent, and utilities."

Best Use Cases

• Complex tasks requiring multistep reasoning
• Scenarios where the agent gathers information before answering
• Problems benefiting from decomposition into smaller steps
• Tasks combining analytical thinking with tool usage

When to Use Each

Key Differences