koog-ktor/ai.koog.ktor

Package-level declarations

Types

class Koog(val application: Application, val promptExecutor: PromptExecutor, val agentConfig: KoogAgentsConfig.AgentConfig, val agentFeatures: List<GraphAIAgent.FeatureContext.() -> Unit>, job: CompletableJob)

Represents an instance of Koog with configuration for prompt execution, language model, tool management, agent setup, and features.

KoogAgentsConfig

class KoogAgentsConfig(scope: CoroutineScope)

Configuration class for setting up a Koog agents server. Provides options to configure LLM connections, agent tools, features, and other related settings.

McpToolsConfig

class McpToolsConfig(agentConfig: KoogAgentsConfig.AgentConfig)

Configuration class for MCPTools that manages the integration of various tool registries into the system. Provides methods to process tools using different transport mechanisms.

Functions

aiAgent

suspend fun <Result> RoutingContext.aiAgent(runMode: ToolCalls = ToolCalls.SINGLE_RUN_SEQUENTIAL, model: LLModel, block: suspend (agent: AIAgent<String, String>) -> Result): Result

A default aiAgent is an agent that runs using singleRunStrategy, by default, it relies on sequential ToolCalls. Inside the block lambda you can use the agent to perform tasks, and calculate a result, such as GraphAIAgent.run.

inline suspend fun <Input, Output> RoutingContext.aiAgent(strategy: AIAgentGraphStrategy<Input, Output>, model: LLModel, input: Input): Output

Creates an agent using aiAgent, and immediately runs it given the input. When the agent is completed it provides the final Output.

inline suspend fun <Input, Output> RoutingContext.aiAgent(strategy: AIAgentGraphStrategy<Input, Output>, model: LLModel, tools: ToolRegistry = ToolRegistry.EMPTY): AIAgent<Input, Output>

inline suspend fun <Input, Output, Result> RoutingContext.aiAgent(strategy: AIAgentGraphStrategy<Input, Output>, model: LLModel, block: suspend (agent: AIAgent<Input, Output>) -> Result): Result

suspend fun <Input, Output> RoutingContext.aiAgent(inputType: KType, outputType: KType, strategy: AIAgentGraphStrategy<Input, Output>, model: LLModel, tools: ToolRegistry = ToolRegistry.EMPTY): AIAgent<Input, Output>

Creates an AI agent using the provided AI agent strategy within the specified route.

suspend fun RoutingContext.aiAgent(input: String, model: LLModel, runMode: ToolCalls = ToolCalls.SINGLE_RUN_SEQUENTIAL): String

A default aiAgent is an agent that runs using singleRunStrategy, by default, it relies on sequential ToolCalls. It takes an input, and when the agent finishes running provides a final result String.

llm

fun RoutingContext.llm(): PromptExecutor

Retrieve the configured llm, or PromptExecutor instance from the underlying Koog plugin.

mcp

fun KoogAgentsConfig.AgentConfig.mcp(configure: McpToolsConfig.() -> Unit)

Configures the MCP (Modular Configuration Protocol) tools for the agent with the provided configuration block.