agents-core/ai.koog.agents.core.environment

Package-level declarations

Types

AIAgentEnvironment provides a mechanism for AI agents to interface with an external environment. It offers methods for tool execution, error reporting, and sending termination messages.

ContextualAgentEnvironment

class ContextualAgentEnvironment(environment: AIAgentEnvironment, context: AIAgentContext) : AIAgentEnvironment

GenericAgentEnvironment

class GenericAgentEnvironment(agentId: String, logger: KLogger, toolRegistry: ToolRegistry) : AIAgentEnvironment

Represents base agent environment with generic abstractions.

ReceivedToolResult

@Serializable

data class ReceivedToolResult(val id: String?, val tool: String, val toolArgs: JsonObject, val toolDescription: String?, val content: String, val resultKind: ToolResultKind, val result: JsonElement?)

Represents the result or response received from a tool operation.

SafeTool

data class SafeTool<TArgs, TResult>(tool: Tool<TArgs, TResult>, environment: AIAgentEnvironment, clock: Clock)

A wrapper class designed to safely execute a tool within a given AI agent environment. It provides mechanisms for handling tool execution results and differentiating between success and failure cases.

SafeToolFromCallable

data class SafeToolFromCallable<TResult>(val toolFunction: KFunction<TResult>, environment: AIAgentEnvironment, clock: Clock)

A wrapper class that creates a safer abstraction for executing a tool function alongside an associated environment. The tool function is backed by a callable in the form of a KFunction and interacts with the AIAgentEnvironment to perform operations and exchange data.

TerminationTool

object TerminationTool

Provides constants related to the termination tool functionality in an AI agent environment.

ToolResultKind

@Serializable

sealed class ToolResultKind

Represents the possible result types for a tool operation.

Functions

executeTools

suspend fun AIAgentEnvironment.executeTools(toolCalls: List<Message.Tool.Call>): List<ReceivedToolResult>

Executes a batch of tool calls within the AI agent environment and processes their results.

result

fun PromptBuilder.ToolMessageBuilder.result(result: ReceivedToolResult)

Adds a tool result to the prompt.

toSafeResult

fun <TResult> ReceivedToolResult.toSafeResult(tool: Tool<*, TResult>): SafeTool.Result<TResult>

Converts a ReceivedToolResult instance into a SafeTool.Result for safer result handling.