Crate lmm_agent 

lmm-agent - equation-based autonomous agent framework.

This crate lets you compose, build, and run autonomous agents that reason through the lmm equation engine and optionally enrich their knowledge via DuckDuckGo (no LLM provider required).

Quick Start

use lmm_agent::prelude::*;

#[derive(Debug, Default, Auto)]
pub struct MyAgent {
    persona: Cow<'static, str>,
    behavior:  Cow<'static, str>,
    status:    Status,
    agent:     LmmAgent,
    memory:    Vec<Message>,
}

#[async_trait]
impl Executor for MyAgent {
    async fn execute<'a>(
        &'a mut self, tasks: &'a mut Task,
        execute: bool, browse: bool, max_tries: u64,
    ) -> Result<()> {
        let prompt = self.agent.persona().to_string();
        let response = self.generate(&prompt).await?;
        self.agent.add_message(Message {
            role:    "assistant".into(),
            content: response.into(),
        });
        Ok(())
    }
}

#[tokio::main]
async fn main() {
    let agent = MyAgent {
        persona: "Research Agent".into(),
        behavior: "Research Rust async patterns.".into(),
        agent: LmmAgent::new("Research Agent".into(), "Research Rust async patterns.".into()),
        ..Default::default()
    };

    AutoAgent::default()
        .with(agents![agent])
        .build()
        .unwrap()
        .run()
        .await
        .unwrap();
}

Modules

agent

LmmAgent - the core agent struct.

cognition

cognition - closed-loop agent cognition.

error

Error types for lmm-agent.

prelude

Prelude

runtime

AutoAgent - the async orchestrator.

traits

Traits module.

types

Agent domain types.

Macros

agents

Wraps any AgentFunctions implementor in the type-erased pointer form expected by AutoAgent::with.