Struct AgentLoop 

pub struct AgentLoop { /* private fields */ }

Implementations

impl AgentLoop

pub fn new( agent_id: AgentId, llm_client: Arc<dyn LlmClient>, tools: ToolRegistry, system_prompt: String, max_iterations: usize, ) -> Self

pub fn with_max_context_tokens(self, tokens: usize) -> Self

Set the maximum context window size in tokens.

pub fn with_compaction_strategy(self, strategy: CompactionStrategy) -> Self

Set the compaction strategy to use.

pub fn with_agent_name(self, name: impl Into<String>) -> Self

Set a human-readable name for this agent (used in events).

pub fn with_messages( agent_id: AgentId, llm_client: Arc<dyn LlmClient>, tools: ToolRegistry, messages: Vec<ChatMessage>, max_iterations: usize, ) -> Self

Create an AgentLoop with existing conversation history (for multi-turn).

pub fn set_event_sink(&mut self, tx: UnboundedSender<AgentEvent>)

pub fn set_background_rx(&mut self, rx: UnboundedReceiver<BackgroundResult>)

Set the receiver for background agent results. Results arriving on this channel are injected as user messages before each LLM call, mirroring Claude Code’s background agent notification.

pub fn messages(&self) -> &[ChatMessage]

Get a clone of the current conversation messages.

pub async fn run( &mut self, initial_user_message: String, ) -> SdkResult<AgentLoopResult>