Trait AgentLoopPort 

pub trait AgentLoopPort: Send + Sync {
    // Required method
    fn run<'life0, 'async_trait>(
        &'life0 self,
        messages: Vec<AgentMessage>,
        config: AgentConfig,
        tx: Sender<AgentEvent>,
    ) -> Pin<Box<dyn Future<Output = Result<AgentRunOutput, AgentError>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;
}

Port: drives the full backend agentic loop.

Usage

use tokio::sync::mpsc;
use gglib_core::ports::AgentLoopPort;
use gglib_core::domain::{AgentConfig, AgentEvent, AgentMessage};

async fn run_loop(agent: &dyn AgentLoopPort) {
    let (tx, mut rx) = mpsc::channel::<AgentEvent>(64);

    // Spawn a task to consume the event stream (e.g. SSE or logging).
    tokio::spawn(async move {
        while let Some(event) = rx.recv().await {
            println!("{:?}", event);
        }
        // rx.recv() returns None when tx is dropped (loop ended).
    });

    let messages = vec![AgentMessage::User { content: "Hello".into() }];
    let output = agent.run(messages, AgentConfig::default(), tx).await?;
    println!("Final: {}", output.answer);
    // `history` contains the full accumulated message list including all
    // assistant and tool-result messages appended during the loop — safe
    // to pass directly as the `messages` argument for the next turn.
}

Channel ownership and stream termination

events is taken by value. When run returns (whether Ok or Err) the Sender is dropped, which closes the channel and signals None to the Receiver. Axum SSE handlers and CLI consumers can rely on this to know the stream has ended without needing an explicit sentinel event.

Required Methods

fn run<'life0, 'async_trait>( &'life0 self, messages: Vec<AgentMessage>, config: AgentConfig, tx: Sender<AgentEvent>, ) -> Pin<Box<dyn Future<Output = Result<AgentRunOutput, AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Execute the agentic loop and return the final answer.

Parameters

• messages — The initial conversation history (system prompt + user message at minimum).
• config — Loop control parameters (iteration limits, timeouts, etc.).
• tx — Async channel over which the loop streams AgentEvents. Taken by value; dropped on completion to close the SSE stream.

Returns

• Ok(AgentRunOutput) — The final answer and full accumulated message history (safe to pass back as messages on the next turn).
• Err(AgentError) — A fatal loop-level failure (max iterations reached, loop detection, stagnation, or internal error). No partial history is returned on failure; the caller’s existing history is left intact.

Implementors