Skip to content

AEA quick start

If you want to create Autonomous Economic Agents (AEAs) that can act independently of constant user input and autonomously execute actions to achieve their objective, you can use the AEA framework.

This example will take you through a simple AEA to familiarise you with the basics of the framework.

System Requirements

The AEA framework can be used on Windows, Ubuntu/Debian and MacOS.

You need Python 3.6 or higher as well as Go 1.14.2 or higher installed. ​

Option 1: Manual system preparation

Install a compatible Python and Go version on your system (see this external resource for a comprehensive guide).

Manual approach The following hints can help:
  • To install Go, follow the official guide, depending on your platform here
  • Python is already included by default on many Linux distributions (e.g. Ubuntu), as well as MacOS. To check you have the right version, open a terminal and run:
    python3 --version
    
  • To install Python on Windows machines, you can download a specific release here.
  • Ubuntu/Debian systems only: install Python headers, depending on the Python version you have installed on your machine. E.g. for Python 3.7:
    sudo apt-get install python3.7-dev
    
  • Windows users: install tools for Visual Studio.

Option 2: Using an automated install script

We provide a script to automatically install all framework dependencies and the framework itself. This means that if you follow this option, you can skip the installation step that comes later on this page.

Automated install script approach On MacOS or Ubuntu run the following commands to download and install:
curl https://raw.githubusercontent.com/fetchai/agents-aea/main/scripts/install.sh --output install.sh
chmod +x install.sh
./install.sh
On Windows: download https://raw.githubusercontent.com/fetchai/agents-aea/main/scripts/install.ps1, then run install.ps1 with the PowerShell terminal.

Option 3: Using Docker

​ We also provide a Docker image with all the needed dependencies.

Docker approach To use the image you will first have to pull it and than run it with your current local directory mounted as a docker volume. This allows you to keep your agents local while working on them from within the docker container. To pull:
docker pull fetchai/aea-user:latest
To run the image on Linux and MacOs:
docker run -it -v $(pwd):/agents --workdir=/agents fetchai/aea-user:latest 
And on Windows:
docker run -it -v %cd%:/agents --workdir=/agents fetchai/aea-user:latest 
Once successfully logged into the docker container, you can follow the rest of the guide the same way as if not using docker. ​

Preliminaries

Ensure, you are in a clean working directory:

  • either you create it manually mkdir my_aea_projects/ && cd my_aea_projects/, then add an empty directory called packages with the following command mkdir packages/,

  • or you clone the template repo as described in Approach 1 in the development setup guide.

At this point, when typing ls you should see a single folder called packages in your working environment. This will act as your local registry for AEA components.

Unless you are using the docker image, we highly recommend using a virtual environment to ensure consistency across dependencies.

Check that you have pipenv.

which pipenv

If you don't have it, install it. Instructions are here.

Once installed, create a new environment and open it (here we use Python 3.7 but the AEA framework supports any Python >= 3.6).

touch Pipfile && pipenv --python 3.7 && pipenv shell

Installation

The following installs the entire AEA package which also includes a command-line interface (CLI). (You can skip this step if you used the install script above: Option 2 .)

pip install aea[all]

If you are using zsh rather than bash type

pip install 'aea[all]'

If the installation steps fail, it might be a dependency issue. Make sure you have followed all the relevant system specific steps above under System Requirements.

Setup author name

AEAs are composed from components. AEAs and AEA components can be developed by anyone and pushed to the AEA registry for others to use. To use the registry, we need to register an author name.

You can set up your author name using the init command:

aea init

This is your unique author (or developer) name in the AEA ecosystem.

You should see a similar output (with your input instead of the sample username and email):

Do you have a Registry account? [y/N]: n
Create a new account on the Registry now:
Username: fetchai
Email: [email protected]
Password:
Please make sure that passwords are equal.
Confirm password:
    _     _____     _
   / \   | ____|   / \
  / _ \  |  _|    / _ \
 / ___ \ | |___  / ___ \
/_/   \_\|_____|/_/   \_\

v1.0.0

AEA configurations successfully initialized: {'author': 'fetchai'}

Note

If you would rather not create an account on the registry at this point, then run aea init --local instead.

Echo skill demo

This is a simple demo that introduces you to the main components of an AEA.

The fastest way to have your first AEA is to fetch one that already exists!

aea fetch fetchai/my_first_aea:0.26.0
cd my_first_aea

To learn more about the folder structure of an AEA project read on here.

Alternatively: step by step install Create a new AEA
First, create a new AEA project and enter it.
aea create my_first_aea
cd my_first_aea

Add the stub connection
Second, add the stub connection to the project.
aea add connection fetchai/stub:0.20.0

Add the echo skill
Third, add the echo skill to the project.
aea add skill fetchai/echo:0.18.0
This copies the fetchai/echo:0.18.0 skill code containing the "behaviours", and "handlers" into the project, ready to run. The identifier of the skill fetchai/echo:0.18.0 consists of the name of the author of the skill, followed by the skill name and its version.

Echo skill

Just like humans, AEAs can have skills to achieve their tasks. As an agent developer, you can create skills to add to your own AEAs. You can also choose to publish your skills so others add them to their AEAs. More details on skills can be found on this page .

The above agent has an echo skill, fetched from the registry, which simply echoes any messages it receives back to its sender.

Communication via envelopes and messages

AEAs use envelopes containing messages for communication. To learn more, check out the next section.

Stub connection

Besides skills, AEAs may have one or more connections enabling them to interface with entities in the outside world. For example, an HTTP client connection allows an AEA to communicate with HTTP servers. To read more about connections see this page.

In this demo, we use the stub connection (fetchai/stub0.15.0) to send envelopes to and receive envelopes from the AEA.

A stub connection provides an I/O reader and writer. It uses two files for communication: one for incoming envelopes and the other for outgoing envelopes.

The AEA waits for a new envelope posted to the file my_first_aea/input_file, and adds a response to the file my_first_aea/output_file.

The format of each envelope is the following:

TO,SENDER,PROTOCOL_ID,ENCODED_MESSAGE,

For example:

recipient_aea,sender_aea,fetchai/default:1.0.0,\x08\x01\x12\x011*\x07\n\x05hello,

Install AEA dependencies

aea install

Add and create a private key

All AEAs need a private key to run. Add one now:

aea generate-key fetchai
aea add-key fetchai

Run the AEA

Run the AEA.

aea run

You will see the echo skill running in the terminal window (an output similar to the one below).

    _     _____     _
   / \   | ____|   / \
  / _ \  |  _|    / _ \
 / ___ \ | |___  / ___ \
/_/   \_\|_____|/_/   \_\

v1.0.0

Starting AEA 'my_first_aea' in 'async' mode ...
info: Echo Handler: setup method called.
info: Echo Behaviour: setup method called.
info: [my_first_aea]: Start processing messages...
info: Echo Behaviour: act method called.
info: Echo Behaviour: act method called.
info: Echo Behaviour: act method called.
...

The framework first calls the setup methods in the skill's Handler and Behaviour classes in that order; after which it repeatedly calls the act method of Behaviour class. This is the main agent loop in action.

Add a message to the input file

You can send the AEA a message wrapped in an envelope using the CLI's interact command.

From a different terminal and same directory (ensure you are in the same virtual environment: pipenv shell):

cd my_first_aea
aea interact

You can now send messages to this AEA via an interactive tool by typing anything into the prompt and hitting enter twice (once to send the message and once more to check for a response).

Let us send hello to this AEA (type hello and press enter twice). In the original terminal, you will see the Echo Handler dealing with this envelope and its contained message. You should see an output similar to the one below but with a different dialogue_reference.

info: Echo Behaviour: act method called.
info: Echo Handler: message=Message(dialogue_reference=('1', '') message_id=1 target=0 performative=bytes content=b'hello'), sender=my_first_aea_interact
info: Echo Behaviour: act method called.
info: Echo Behaviour: act method called.
Manual approach Optionally, from a different terminal and same directory (i.e. the my_first_aea project), you can send the AEA a message wrapped in an envelope via the input file.
echo 'my_first_aea,sender_aea,fetchai/default:1.0.0,\x12\x10\x08\x01\x12\x011*\t*\x07\n\x05hello,' >> input_file
You will see the Echo Handler dealing with the envelope and responding with the same message to the output_file, and also decoding the Base64 encrypted message in this case.
info: Echo Behaviour: act method called.
Echo Handler: message=Message(sender=sender_aea,to=my_first_aea,content=b'hello',dialogue_reference=('1', ''),message_id=1,performative=bytes,target=0), sender=sender_aea
info: Echo Behaviour: act method called.
info: Echo Behaviour: act method called.
Note, due to the dialogue reference having to be incremented, you can only send the above envelope once! This approach does not work in conjunction with the aea interact command.

Stop the AEA

You can stop an AEA by pressing CTRL C.

Once you do, you should see the AEA being interrupted and then calling the teardown() methods:

info: Echo Behaviour: act method called.
info: Echo Behaviour: act method called.
^C my_first_aea interrupted!
my_first_aea stopping ...
info: Echo Handler: teardown method called.
info: Echo Behaviour: teardown method called.

Write a test for the AEA

We can write an end-to-end test for the AEA utilising helper classes provided by the framework.

Writing tests The following test class replicates the preceding demo and tests it's correct behaviour. The AEATestCase classes are a tool for AEA developers to write useful end-to-end tests of their AEAs. First, get the packages directory from the AEA repository (execute from the working directory which contains the my_first_aea folder):
svn export https://github.com/fetchai/agents-aea.git/trunk/packages
Then write the test:
import signal
import time

from aea.common import Address
from aea.mail.base import Envelope
from aea.protocols.base import Message
from aea.protocols.dialogue.base import Dialogue

from packages.fetchai.protocols.default.dialogues import DefaultDialogue, DefaultDialogues
from packages.fetchai.protocols.default.message import DefaultMessage
from packages.fetchai.protocols.default.serialization import DefaultSerializer
from aea.test_tools.test_cases import AEATestCase


class TestEchoSkill(AEATestCase):
    """Test that echo skill works."""

    def test_echo(self):
        """Run the echo skill sequence."""
        process = self.run_agent()
        is_running = self.is_running(process)
        assert is_running, "AEA not running within timeout!"

        # add sending and receiving envelope from input/output files
        sender_aea = "sender_aea"
        def role_from_first_message(
            message: Message, receiver_address: Address
        ) -> Dialogue.Role:
            return DefaultDialogue.Role.AGENT
        dialogues = DefaultDialogues(sender_aea, role_from_first_message)
        message_content = b"hello"
        message = DefaultMessage(
            performative=DefaultMessage.Performative.BYTES,
            dialogue_reference=dialogues.new_self_initiated_dialogue_reference(),
            content=message_content,
        )
        sent_envelope = Envelope(
            to=self.agent_name,
            sender=sender_aea,
            protocol_id=message.protocol_id,
            message=DefaultSerializer().encode(message),
        )

        self.send_envelope_to_agent(sent_envelope, self.agent_name)

        time.sleep(2.0)
        received_envelope = self.read_envelope_from_agent(self.agent_name)

        assert sent_envelope.to == received_envelope.sender
        assert sent_envelope.sender == received_envelope.to
        assert sent_envelope.protocol_id == received_envelope.protocol_id
        received_message = DefaultMessage.serializer.decode(received_envelope.message)
        assert message.content == received_message.content

        check_strings = (
            "Echo Handler: setup method called.",
            "Echo Behaviour: setup method called.",
            "Echo Behaviour: act method called.",
            "content={}".format(message_content),
        )
        missing_strings = self.missing_from_output(process, check_strings)
        assert (
            missing_strings == []
        ), "Strings {} didn't appear in agent output.".format(missing_strings)

        assert (
            self.is_successfully_terminated()
        ), "Echo agent wasn't successfully terminated."
Place the above code into a file test.py in your AEA project directory (the same level as the aea-config.yaml file). To run, execute the following:
pytest test.py

Delete the AEA

Delete the AEA from the parent directory (cd .. to go to the parent directory).

aea delete my_first_aea

Next steps

To gain an understanding of the core components of the framework, please continue to the next page:

For more demos, use cases or step by step guides, please check the following: