agents-features-memory/ai.koog.agents.local.memory.feature/AgentMemory

AgentMemory

class AgentMemory(agentMemory: AgentMemoryProvider, llm: AIAgentLLMContext, scopesProfile: MemoryScopesProfile)(source)

Memory implementation for AI agents that provides persistent storage and retrieval of facts.

The AgentMemory feature enables agents to:

Store information (facts) for later retrieval
Organize memory by concepts, subjects, and scopes
Share knowledge between different agents based on scope
Extract facts from conversation history
Load relevant facts into the agent's context

This class serves as the main interface for memory operations within an agent, combining the memory provider with the agent's LLM context to seamlessly integrate memory capabilities into the agent's workflow.

To install the AgentMemory feature in your agent:

val agent = AIAgents(
    strategy = myStrategy,
    promptExecutor = myExecutor
) {
    // Install memory feature with custom configuration
    install(AgentMemory) {
        // Configure memory provider (required)
        memoryProvider = LocalFileMemoryProvider(
            config = LocalMemoryConfig("my-agent-memory"),
            storage = SimpleStorage(JVMFileSystemProvider),
            root = Path("memory/data")
        )

        // Configure scope names (optional)
        featureName = "code-assistant"
        productName = "my-ide"
        organizationName = "my-company"
    }
}

Example usage within an agent node:

val rememberUserPreference by node {
    withMemory {
        // Save a fact about user preference
        agentMemory.save(
            fact = SingleFact(
                concept = Concept("preferred-language", "User's preferred programming language"),
                value = "Kotlin"
            ),
            subject = MemorySubjects.User,
            scope = MemoryScope.Product("my-ide")
        )
    }
}

Constructors

AgentMemory

constructor(agentMemory: AgentMemoryProvider, llm: AIAgentLLMContext, scopesProfile: MemoryScopesProfile)

Types

Config

class Config : FeatureConfig

Configuration for the AgentMemory feature.

Feature

object Feature : AIAgentFeature<AgentMemory.Config, AgentMemory>

Feature companion object that allows installing the AgentMemory feature in an agent.

Functions

loadAllFactsToAgent

suspend fun loadAllFactsToAgent(scopes: List<MemoryScopeType> = MemoryScopeType.entries, subjects: List<MemorySubject> = MemorySubject.registeredSubjects)

Loads all available facts from memory and adds them to the LLM chat history.

loadFactsToAgent

suspend fun loadFactsToAgent(concept: Concept, scopes: List<MemoryScopeType> = MemoryScopeType.entries, subjects: List<MemorySubject> = MemorySubject.registeredSubjects)

Loads facts about a specific concept from memory and adds them to the LLM chat history.

saveFactsFromHistory

suspend fun saveFactsFromHistory(concept: Concept, subject: MemorySubject, scope: MemoryScope, preserveQuestionsInLLMChat: Boolean = false)

Extracts and saves facts from the LLM chat history based on the provided concept.

AgentMemory

See also

Constructors

Types

Functions