Struct AgentLoop 

pub struct AgentLoop { /* private fields */ }

The core ReAct agent loop.

Drives an LLM through iterative reasoning and tool execution until the model produces a final text answer or the iteration limit is reached.

The loop does not own the LLM client; instead, run() takes &dyn LlmClient so the same loop can be reused with different backends.

Implementations

impl AgentLoop

pub fn builder() -> AgentLoopBuilder

Create a builder with default settings.

pub fn channel_config(&self) -> &ChannelConfig

Returns the channel configuration for this agent loop.

Used by AgentSession to create bounded channels with the configured capacities and policies.

pub async fn run( &self, llm: &dyn LlmClient, messages: Vec<Message>, on_event: impl Fn(AgentEvent) + Send + Sync, ) -> Result<AgentResult>

Run the agent loop to completion.

Iteratively calls the LLM, executes any requested tools, and feeds tool results back into the conversation until the LLM produces a final text response or max_iterations is exceeded.

The on_event callback is invoked for each notable event (tool started, tool completed, text chunk). Pass |_| {} for a no-op.

pub async fn run_with_ops( &self, llm: &dyn LlmClient, messages: Vec<Message>, ops_rx: Option<Receiver<AgentOp>>, on_event: impl Fn(AgentEvent) + Send + Sync, ) -> Result<AgentResult>

Run with an optional interactive operations channel.

Pass ops_rx: None for behavior equivalent to run.

pub async fn run_streaming( &self, llm: &dyn LlmClient, messages: Vec<Message>, on_event: impl Fn(AgentEvent) + Send + Sync, ) -> Result<AgentResult>

Streaming variant of run.

Uses LlmClient::chat_stream so that on_event receives AgentEvent::TextChunk for each text delta as it arrives, plus AgentEvent::TextDone with the full accumulated text and AgentEvent::IterationStarted at each reasoning iteration.

Tool execution is identical to the non-streaming path.