troop 0.0.9

A lightweight framework for multi-agent orchestration, forked from OpenAI's Swarm library.

Troop

A simple and lightweight framework for multi-agent orchestration, forked from OpenAI's Swarm library.

There are many great and powerful LLM agent libraries out there. troop is trying to become a simple, flexible and production-ready alternative that let's you focus on your projects instead of spending time on learning complex tools.

OpenAI and Anthropic have both shared their thoughts on what

Install

Requires Python 3.10+

pip install git+https://github.com/pietz/troop.git

Usage

Currently, troop is a drop-in replacement for swarm from OpenAI. Change the imports and your existing code should work. While I want to try to keep it this way, future features might introduce breaking changes.

from troop import Troop, Agent

client = Troop()

def transfer_to_agent_b():
    return agent_b

agent_a = Agent(
    name="Agent A",
    instructions="You are a helpful agent.",
    functions=[transfer_to_agent_b],
)

agent_b = Agent(
    name="Agent B",
    instructions="Only speak in Haikus.",
)

response = client.run(
    agent=agent_a,
    messages=[{"role": "user", "content": "I want to talk to agent B."}],
)

print(response.messages[-1]["content"])

Overview

Troop focuses on lightweight agent coordination and execution through two core abstractions:

1. Agent: Encapsulates instructions and tools
2. Handoffs: Allows agents to transfer control to other agents

These primitives enable building scalable multi-agent systems while maintaining simplicity.

Documentation

Running Troop

Initialize a client:

from troop import Troop
client = Troop()

Client.run() Arguments

Argument	Type	Description	Default
agent	Agent	Initial agent to be called	(required)
messages	List	List of message objects	(required)
context_variables	dict	Additional context variables	{}
max_turns	int	Maximum conversation turns	float("inf")
model_override	str	Override agent's model	None
execute_tools	bool	Execute tool calls	True
stream	bool	Enable streaming responses	False
debug	bool	Enable debug logging	False

Agent Configuration

Field	Type	Description	Default
name	str	Agent name	"Agent"
model	str	Model to use	"gpt-4"
instructions	str/func	Agent instructions	"You are a helpful agent."
functions	List	Available functions	[]
tool_choice	str	Tool choice override	None

Functions

Functions can:

• Return strings (or values castable to strings)
• Return other agents for handoffs
• Access context variables
• Update context through Result objects

Example with context variables:

def greet(context_variables, language):
    user_name = context_variables["user_name"]
    greeting = "Hola" if language.lower() == "spanish" else "Hello"
    return f"{greeting}, {user_name}!"

agent = Agent(functions=[greet])

Streaming

stream = client.run(agent, messages, stream=True)
for chunk in stream:
    print(chunk)

Testing

Use the built-in REPL for testing:

from troop.repl import run_demo_loop
run_demo_loop(agent, stream=True)