Struct AgentLoop 

pub struct AgentLoop { /* private fields */ }

Agent loop executor

Implementations

impl AgentLoop

pub fn new( llm_client: Arc<dyn LlmClient>, tool_executor: Arc<ToolExecutor>, tool_context: ToolContext, config: AgentConfig, ) -> Self

pub fn with_progress_tracker( self, tracker: Arc<RwLock<ProgressTracker>>, ) -> Self

Set the progress tracker for real-time tool/token usage tracking.

pub fn with_task_manager(self, manager: Arc<TaskManager>) -> Self

Set the task manager for centralized task lifecycle tracking.

pub fn with_tool_metrics(self, metrics: Arc<RwLock<ToolMetrics>>) -> Self

Set the tool metrics collector for this agent loop

pub fn with_queue(self, queue: Arc<SessionLaneQueue>) -> Self

Set the lane queue for priority-based tool execution.

When set, tools are routed through the lane queue which supports External task handling for multi-machine parallel processing.

pub fn detect_context_perception_intent( &self, prompt: &str, session_id: &str, workspace: &str, ) -> Option<PreContextPerceptionEvent>

Detect if context perception is needed based on user prompt.

Returns Some(PreContextPerceptionEvent) if the prompt suggests the model needs workspace knowledge (finding files, understanding code, etc.).

pub async fn execute( &self, history: &[Message], prompt: &str, event_tx: Option<Sender<AgentEvent>>, ) -> Result<AgentResult>

Execute the agent loop for a prompt

Takes the conversation history and a new user prompt. Returns the agent result and updated message history. When event_tx is provided, uses streaming LLM API for real-time text output.

pub async fn execute_from_messages( &self, messages: Vec<Message>, session_id: Option<&str>, event_tx: Option<Sender<AgentEvent>>, cancel_token: Option<&CancellationToken>, ) -> Result<AgentResult>

Execute the agent loop with pre-built messages (user message already included).

Used by send_with_attachments / stream_with_attachments where the user message contains multi-modal content and is already appended to the messages vec.

pub async fn execute_with_session( &self, history: &[Message], prompt: &str, session_id: Option<&str>, event_tx: Option<Sender<AgentEvent>>, cancel_token: Option<&CancellationToken>, ) -> Result<AgentResult>

Execute the agent loop for a prompt with session context

Takes the conversation history, user prompt, and optional session ID. When session_id is provided, context providers can use it for session-specific context.

pub async fn execute_streaming( &self, history: &[Message], prompt: &str, ) -> Result<(Receiver<AgentEvent>, JoinHandle<Result<AgentResult>>, CancellationToken)>

Execute with streaming events

pub async fn plan( &self, prompt: &str, _context: Option<&str>, ) -> Result<ExecutionPlan>

Create an execution plan for a prompt

Delegates to LlmPlanner for structured JSON plan generation, falling back to heuristic planning if the LLM call fails.

pub async fn execute_with_planning( &self, history: &[Message], prompt: &str, event_tx: Option<Sender<AgentEvent>>, pre_analysis: Option<PreAnalysis>, ) -> Result<AgentResult>

Execute with planning phase.

If pre_analysis is provided (from a single pre-analysis LLM call in execute_with_session), the goal and plan are already available and no additional LLM calls are needed for planning. Otherwise, falls back to calling extract_goal and plan individually.

pub async fn extract_goal(&self, prompt: &str) -> Result<AgentGoal>

Extract goal from prompt

Delegates to LlmPlanner for structured JSON goal extraction, falling back to heuristic logic if the LLM call fails.

pub async fn check_goal_achievement( &self, goal: &AgentGoal, current_state: &str, ) -> Result<bool>

Check if goal is achieved

Delegates to LlmPlanner for structured JSON achievement check, falling back to heuristic logic if the LLM call fails.

Trait Implementations

impl Clone for AgentLoop

fn clone(&self) -> AgentLoop

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.