mojentic-core/com.mojentic.agents/BaseAsyncLlmAgentWithMemory

BaseAsyncLlmAgentWithMemory

open class BaseAsyncLlmAgentWithMemory(broker: LlmBroker, model: String, val memory: SharedWorkingMemory, behaviour: String, val instructions: String, config: CompletionConfig = CompletionConfig()) : BaseAsyncLlmAgent

LLM-backed agent that remembers facts in a SharedWorkingMemory.

Before each turn the agent injects the current memory contents into the conversation as a system-style hint. After the turn, callers may use the returned response — for the Python reference the memory is updated from a structured-output field; the Kotlin port keeps the merge explicit so callers can pull the relevant block from a structured response themselves and pass it to mergeMemory (which is just a convenience wrapper around the memory's own merge).

Constructors

constructor(broker: LlmBroker, model: String, memory: SharedWorkingMemory, behaviour: String, instructions: String, config: CompletionConfig = CompletionConfig())

Properties

behaviour

var behaviour: String

the system-prompt persona / instructions.

broker

val broker: LlmBroker

the LlmBroker used for generation.

config

val config: CompletionConfig

completion-time settings (temperature, tool budget, etc.).

instructions

val instructions: String

memory

val memory: SharedWorkingMemory

model

val model: String

the LLM model identifier passed to the broker.

tools

val tools: List<LlmTool>

Immutable snapshot of currently registered tools.

Functions

addTool

fun addTool(tool: LlmTool)

generateResponse

open suspend override fun generateResponse(content: String): LlmGatewayResponse

Run a single LLM turn with content appended as a user message. Returns the full LlmGatewayResponse so subclasses can inspect content, tool calls, and structured output.

mergeMemory

suspend fun mergeMemory(delta: Map<String, JsonElement>)

receiveEvent

open suspend override fun receiveEvent(event: Event): List<Event>

Default implementation does nothing — override to emit follow-up events.