PyPi | Read the Docs | Github | mail

mango (modular python agent framework) is a python library for multi-agent systems (MAS). It is written on top of asyncio and is released under the MIT license.

mango allows the user to create simple agents with little effort and in the same time offers options to structure agents with complex behaviour.

The main features of mango are:

• Container mechanism to speedup local message exchange
• Message definition based on the FIPA ACL standard
• Structuring complex agents with loose coupling and agent roles
• Built-in codecs: JSON and protobuf
• Supports communication between agents directly via TCP or via an external MQTT broker

A detailed documentation for this project can be found at mango-agents.readthedocs.io

Installation

mango requires Python >= 3.8 and runs on Linux, OSX and Windows.

For installation of mango you should use virtualenv which can create isolated Python environments for different projects.

It is also recommended to install virtualenvwrapper which makes it easier to manage different virtual environments.

Once you have created a virtual environment you can just run pip to install it:

$ pip install mango-agents

Getting started

Creating an agent

In our first example, we create a very simple agent that simply prints the content of all messages it receives:

    from mango import Agent

    class RepeatingAgent(Agent):

        def __init__(self):
            super().__init__()
            print(f"Creating a RepeatingAgent. At this point self.addr={self.addr}")

        def handle_message(self, content, meta):
            # This method defines what the agent will do with incoming messages.
            print(f"Received a message with the following content: {content}!")

        def on_register(self):
            print(f"The agent has been registered to a container: {self.addr}!")

        def on_ready(self):
            print("All containers have been activated!")

Agents must be a subclass of mango.Agent. Agent's are notified when they are registered mango.Agent.on_register and when the container(s) has been activated mango.Agent.on_ready. Consequenty, most agent features (like scheduling, sending internal messages, the agent address) are available after registration, and only after mango.Agent.on_ready has been called, all features are available (sending external messages).

Creating a container

Agents live in containers, so we need to know how to create a mango container. The container is responsible for message exchange between agents.

    from mango import create_tcp_container

    # Containers have to be created using a factory method
    # Other container types are available through create_mqtt_container and create_ec_container
    container = create_tcp_container(addr=('localhost', 5555))

This is how a tcp container is created. While container creation, it is possible to set the codec, the address information (depending on the type) and the clock (see read the docs Scheduling).

Running your first agent within a container

The container and the contained agents need asyncio (see asyncio docs <https://docs.python.org/3.10/library/asyncio.html>_) to work, therefore we need write a coroutine function and execute it using asyncio.run.

The following script will create a RepeatingAgent, register it, and let it run within a container for 50ms and then shutdown the container:

    import asyncio
    from mango import create_tcp_container, Agent, activate

    class RepeatingAgent(Agent):
        def __init__(self):
            super().__init__()
            print(f"Creating a RepeatingAgent. At this point self.addr={self.addr}")

        def handle_message(self, content, meta):
            print(f"Received a message with the following content: {content}!")

        def on_register(self):
            print(f"The agent has been registered to a container: {self.addr}!")

        def on_ready(self):
            print("All containers have been activated!")


    async def run_container_and_agent(addr, duration):
        first_container = create_tcp_container(addr=addr)
        first_agent = first_container.register(RepeatingAgent())

        async with activate(first_container) as container:
            await asyncio.sleep(duration)

    asyncio.run(run_container_and_agent(addr=('localhost', 5555), duration=0.05))

In this example no messages are sent, nor does the Agent do anything, but the call order of the hook-in functions is clearly visible. The function mango.activate will start the container and shut it down after the code in its scope has been execute (here, after the sleep).

Creating a proactive Agent

Let's implement another agent that is able to send a hello world message to another agent:

    import asyncio
    from mango import Agent

    class HelloWorldAgent(Agent):
        async def greet(self, other_addr):
            await self.send_message("Hello world!", other_addr)

        def handle_message(self, content, meta):
            print(f"Received a message with the following content: {content}")

    async def run_container_and_agent(addr, duration):
        first_container = create_tcp_container(addr=addr)
        first_hello_agent = first_container.register(HelloWorldAgent())
        second_hello_agent = first_container.register(HelloWorldAgent())

        async with activate(first_container) as container:
            await first_hello_agent.greet(second_hello_agent.addr)

    asyncio.run(run_container_and_agent(addr=('localhost', 5555), duration=0.05))

If you do not want to await sending the message, and just let asyncio/mango schedule it, you can use mango.Agent.schedule_instant_message instead of mango.Agent.send_message.

Support

• Documentation: mango-agents.readthedocs.io
• E-mail: [email protected]

License

Distributed under the MIT license.