Crate dragen 

Dragen - CodeAct-style agent framework

Dragen provides a simple framework for building AI agents that execute Python code in a secure sandbox. It uses the CodeAct pattern where the LLM writes Python code to accomplish tasks, with tools exposed as Python functions.

Quick Start

use dragen::{Agent, AgentConfig};
use littrs::tool;

/// Get the current weather for a city.
///
/// Args:
///     city: The city name
#[tool]
fn get_weather(city: String) -> String {
    format!("Weather in {}: Sunny, 22°C", city)
}

#[tokio::main]
async fn main() {
    let mut agent = Agent::new(AgentConfig::new("gpt-4o"));
    agent.register(get_weather::Tool);

    let result = agent.run("What's the weather in Paris?").await.unwrap();
    println!("{}", result);
}

Sharing Data Between Agents

Use Context to pass data between agents without manual Arc/Mutex management:

use dragen::{Agent, AgentConfig, Context};

let ctx = Context::new();

// Planner writes output to context
let mut planner = Agent::new(AgentConfig::new("gpt-4o"))
    .to_context(&ctx, "plan");
planner.run::<PlanOutput>(&query).await?;

// Executor reads from context (injected into prompt)
let mut executor = Agent::new(AgentConfig::new("gpt-4o"))
    .from_context(&ctx, "plan");
executor.run::<String>("Execute the plan").await?;

Re-exports

pub use littrs;

Structs

Agent

A CodeAct-style agent that executes Python code in a sandbox.

AgentCallbacks

Storage for agent callbacks

AgentConfig

Configuration for the CodeAct agent.

Context

Shared context for passing data between agents.

Enums

AgentEvent

Events emitted during agent execution for observability.

Error

Errors that can occur during agent execution.

Functions

pyvalue_to_json

Convert a PyValue to a serde_json::Value for typed deserialization.

Type Aliases

Result

Result type for Dragen operations.