Struct Agent 

pub struct Agent { /* private fields */ }

The ReAct loop: ask LLM, execute tools, repeat.

Agent orchestrates the interaction between an LLM provider, a tool registry, and a transcript. It drives a loop until the model stops calling tools or the budget (steps or transcript size) is exceeded.

Usage

let agent = Agent::builder()
    .llm(Arc::new(provider))
    .tools(registry)
    .system_prompt("You are a helpful assistant")
    .max_steps(32)
    .build()?;

let outcome = agent.run("Help me debug this file").await?;

Events

The agent emits StepEvents through an optional channel, allowing UI layers to observe progress without coupling to internals. See StepEvent for variants.

Isolation

Each Agent::run() call is independent. The transcript is maintained across steps within a single run but cleared between runs. Use seed_transcript() to carry state between runs if needed.

Implementations

impl Agent

pub fn builder() -> AgentBuilder

pub fn set_transcript(&mut self, transcript: Vec<Message>)

Restore transcript for multi-turn reuse. Call after run() returns to re-seed the agent with the conversation history.

pub fn set_events(&mut self, tx: Option<UnboundedSender<StepEvent>>)

Replace the events channel. Dropping the old sender closes its channel (letting any spawned printer task finish). Pass None to disable event emission entirely.

pub fn confirm_plan(&mut self)

Confirm a proposed plan, allowing execution to proceed.

pub fn reject_plan(&mut self, reason: &str)

Reject a proposed plan with a reason.

pub async fn run(&mut self, goal: impl Into<String>) -> Result<AgentOutcome>

pub fn transcript(&self) -> &[Message]