Skip to content

Protocols

Protocols define agent to agent interactions, which include:

  • messages, which define the representation;

  • serialization logic, which define how a message is encoded for transport; and, optionally

  • dialogues, which define rules over message sequences.

The framework provides one default protocol, called default and introduced below. This protocol provides a bare bones implementation for an AEA protocol which includes a DefaultMessage class and associated DefaultSerializer and DefaultDialogue classes.

Additional protocols - i.e. a new type of interaction - can be added as packages or generated with the protocol generator.

We highly recommend you do not attempt to write your own protocol code; always use existing packages or the protocol generator!

Components of a protocol

A protocol package contains the following files:

  • __init__.py
  • message.py, which defines message representation
  • serialization.py, which defines the encoding and decoding logic
  • two protobuf related files

It optionally also contains * dialogues.py, which defines rules of the message exchange * custom_types.py, which defines custom types

All protocols are for point to point interactions between two agents or agent-like services.

Metadata

Each Message in an interaction protocol has a set of default fields:

  • dialogue_reference: Tuple[str, str], a reference of the dialogue the message is part of. The first part of the tuple is the reference assigned to by the agent who first initiates the dialogue (i.e. sends the first message). The second part of the tuple is the reference assigned to by the other agent. * dialogue_reference: Tuple[str, str], a reference of the dialogue the message is part of. The first part of the tuple is the reference assigned to by the dialogue initiator, the second part of the tuple is the reference assigned to by the dialogue responder. The default value is ("", "").
  • message_id: int, the identifier of the message in a dialogue. The default value is 1.
  • target: int, the id of the message this message is replying to. The default value is 0.
  • performative: Enum, the purpose/intention of the message.
  • is_incoming: bool, a boolean specifying whether the message is outgoing (from the agent), or incoming (from the other agent). The default value is False.
  • counterparty: Address, the address of the counterparty of this agent; the other agent, this agent is communicating with.

The default values for the above fields assume the message is the first message by the agent in a dialogue. Therefore, the message_id is set to 1 indicating the first message in the dialogue, target is 0 since the first message is the only message that does not reply to any other, and is_incoming is False indicating the message is by the agent itself.

By default, the values of dialogue_reference, message_id, target, is_incoming, counterparty are set. However, most interactions involve more than one message being sent as part of the interaction and potentially multiple simultaneous interactions utilising the same protocol. In those cases, the dialogue_reference allows different interactions to be identified as such. The message_id and target are used to keep track of messages and their replies. For instance, on receiving of a message with message_id=1 and target=0, the responding agent could respond with a another with message_id=2 and target=1 replying to the first message. In particular, target holds the id of the message being replied to. This can be the preceding message, or an older one.

Contents

Each message may optionally have any number of contents of varying types.

Dialogue rules

Protocols can optionally have a dialogue module. A dialogue, respectively dialogues object, maintains the state of a single dialogue, respectively all dialogues, associated with the protocol.

The framework provides a number of helpful classes which implement most of the logic to maintain dialogues, namely the Dialogue and Dialogues base classes.

Custom protocol

The developer can generate custom protocols with the protocol generator. This lets the developer specify the speech-acts as well as optionally the dialogue structure (e.g. roles of agents participating in a dialogue, the states a dialogue may end in, and the reply structure of the speech-acts in a dialogue).

We highly recommend you do not attempt to write your own protocol code; always use existing packages or the protocol generator!

fetchai/default:0.9.0 protocol

The fetchai/default:0.9.0 protocol is a protocol which each AEA is meant to implement. It serves AEA to AEA interaction and includes two message performatives:

from enum import Enum

class Performative(Enum):
    """Performatives for the default protocol."""

    BYTES = "bytes"
    ERROR = "error"

    def __str__(self):
        """Get the string representation."""
        return self.value
  • The DefaultMessage of performative DefaultMessage.Performative.BYTES is used to send payloads of byte strings to other AEAs. An example is:

    from packages.fetchai.protocols.default.message import DefaultMessage
    
    msg = DefaultMessage(
        performative=DefaultMessage.Performative.BYTES,
        content=b"This is a bytes payload",
    )
    

  • The DefaultMessage of performative DefaultMessage.Performative.ERROR is used to notify other AEAs of errors in an interaction, including errors with other protocols, by including an error_code in the payload:

    class ErrorCode(Enum):
        """This class represents an instance of ErrorCode."""
    
        UNSUPPORTED_PROTOCOL = 0
        DECODING_ERROR = 1
        INVALID_MESSAGE = 2
        UNSUPPORTED_SKILL = 3
        INVALID_DIALOGUE = 4
    
    An example is:
    msg = DefaultMessage(
        performative=DefaultMessage.Performative.ERROR,
        error_code=DefaultMessage.ErrorCode.UNSUPPORTED_PROTOCOL,
        error_msg="This protocol is not supported by this AEA.",
        error_data={"unsupported_msg": b"serialized unsupported protocol message"},
    )
    

Each AEA's fetchai/error:0.9.0 skill utilises the fetchai/default:0.9.0 protocol for error handling.

fetchai/oef_search:0.10.0 protocol

The fetchai/oef_search:0.10.0 protocol is used by AEAs to interact with an SOEF search node to register and unregister their own services and search for services registered by other agents.

The fetchai/oef_search:0.10.0 protocol definition includes an OefSearchMessage with the following message types:

class Performative(Enum):

    """Performatives for the oef_search protocol."""
    REGISTER_SERVICE = "register_service"
    UNREGISTER_SERVICE = "unregister_service"
    SEARCH_SERVICES = "search_services"
    OEF_ERROR = "oef_error"
    SEARCH_RESULT = "search_result"
    SUCCESS = "success"

    def __str__(self):
        """Get string representation."""
        return self.value

We show some example messages below:

  • To register a service, we require a reference to the dialogue in string form (used to keep different dialogues apart), for instance

    my_dialogue_reference = "a_unique_register_service_dialogue_reference"
    
    and a description of the service we would like to register, for instance
    from aea.helpers.search.models import Description
    
    my_service_data = {"country": "UK", "city": "Cambridge"}
    my_service_description = Description(
        my_service_data,
        data_model=my_data_model,
    )
    
    where we use, for instance
    from aea.helpers.search.generic import GenericDataModel
    
    data_model_name = "location"
    data_model = {
        "attribute_one": {
            "name": "country",
            "type": "str",
            "is_required": True,
        },
        "attribute_two": {
            "name": "city",
            "type": "str",
            "is_required": True,
        },
    }
    my_data_model = GenericDataModel(data_model_name, data_model)
    
    We can then create the message to register this service:
    msg = OefSearchMessage(
        performative=OefSearchMessage.Performative.REGISTER_SERVICE,
        dialogue_reference=(my_dialogue_reference, ""),
        service_description=my_service_description,
    )
    

  • To unregister a service, we require a reference to the dialogue in string form, for instance

    my_dialogue_reference = "a_unique_unregister_service_dialogue_reference"
    
    the description of the service we would like to unregister, say my_service_description from above and construct the message:
    msg = OefSearchMessage(
        performative=OefSearchMessage.Performative.UNREGISTER_SERVICE,
        dialogue_reference=(my_dialogue_reference, ""),
        service_description=my_service_description,
    )
    

  • To search a service, we require a reference to the dialogue in string form, for instance

    my_dialogue_reference = "a_unique_search_dialogue_reference"
    
    and a query we would like the search node to evaluate, for instance
    from aea.helpers.search.models import Constraint, ConstraintType, Query
    
    query_data = {
        "search_term": "country",
        "search_value": "UK",
        "constraint_type": "==",
    }
    query = Query(
        [
            Constraint(
                query_data["search_term"],
                ConstraintType(
                    query_data["constraint_type"],
                    query_data["search_value"],
                ),
            )
        ],
        model=None,
    )
    
    We can then create the message to search these services:
    oef_msg = OefSearchMessage(
        performative=OefSearchMessage.Performative.SEARCH_SERVICES,
        dialogue_reference=(my_dialogue_reference, ""),
        query=query,
    )
    

  • The SOEF search node will respond with a message, say msg of type OefSearchMessage, of performative OefSearchMessage.Performative.SEARCH_RESULT. To access the tuple of agents which match the query, simply use msg.agents. In particular, this will return the agent addresses matching the query. The agent address can then be used to send a message to the agent utilising the P2P agent communication network and any protocol other than fetchai/oef_search:0.10.0.

  • If the SOEF search node encounters any errors with the messages you send, it will return an OefSearchMessage of performative OefSearchMessage.Performative.OEF_ERROR and indicate the error operation encountered:

    class OefErrorOperation(Enum):
    
        """This class represents an instance of OefErrorOperation."""
        REGISTER_SERVICE = 0
        UNREGISTER_SERVICE = 1
        SEARCH_SERVICES = 2
        SEND_MESSAGE = 3
    
        OTHER = 10000
    

fetchai/fipa:0.10.0 protocol

This protocol provides classes and functions necessary for communication between AEAs via a variant of the FIPA Agent Communication Language.

The fetchai/fipa:0.10.0 protocol definition includes a FipaMessage with the following performatives:

class Performative(Enum):
    """Performatives for the fipa protocol."""

    ACCEPT = "accept"
    ACCEPT_W_INFORM = "accept_w_inform"
    CFP = "cfp"
    DECLINE = "decline"
    INFORM = "inform"
    MATCH_ACCEPT = "match_accept"
    MATCH_ACCEPT_W_INFORM = "match_accept_w_inform"
    PROPOSE = "propose"

    def __str__(self):
        """Get the string representation."""
        return self.value

FipaMessages are constructed with a performative, dialogue_reference, message_id, and target as well as the kwargs specific to each message performative.

def __init__(
    self,
    performative: Performative,
    dialogue_reference: Tuple[str, str] = ("", ""),
    message_id: int = 1,
    target: int = 0,
    **kwargs,
)

The fetchai/fipa:0.10.0 protocol also defines a FipaDialogue class which specifies the valid reply structure and provides other helper methods to maintain dialogues.

For examples of the usage of the fetchai/fipa:0.10.0 protocol check out the generic skills step by step guide.

Fipa dialogue

Below, we give an example of a dialogue between two agents. In practice; both dialogues would be maintained in the respective agent.

We first create concrete implementations of FipaDialogue and FipaDialogues for the buyer and seller:

from aea.common import Address
from aea.helpers.search.models import Constraint, ConstraintType, Description, Query
from aea.mail.base import Envelope
from aea.protocols.base import Message
from aea.protocols.dialogue.base import Dialogue as BaseDialogue
from aea.protocols.dialogue.base import DialogueLabel

from packages.fetchai.protocols.fipa.dialogues import FipaDialogue, FipaDialogues
from packages.fetchai.protocols.fipa.message import FipaMessage


class BuyerDialogue(FipaDialogue):
    """The dialogue class maintains state of a dialogue and manages it."""

    def __init__(
        self,
        dialogue_label: DialogueLabel,
        self_address: Address,
        role: BaseDialogue.Role,
        message_class: Type[FipaMessage] = FipaMessage,
    ) -> None:
        """
        Initialize a dialogue.

        :param dialogue_label: the identifier of the dialogue
        :param self_address: the address of the entity for whom this dialogue is maintained
        :param role: the role of the agent this dialogue is maintained for

        :return: None
        """
        FipaDialogue.__init__(
            self,
            dialogue_label=dialogue_label,
            self_address=self_address,
            role=role,
            message_class=message_class,
        )
        self.proposal = None  # type: Optional[Description]


class BuyerDialogues(FipaDialogues):
    """The dialogues class keeps track of all dialogues."""

    def __init__(self, self_address: Address) -> None:
        """
        Initialize dialogues.

        :return: None
        """
        def role_from_first_message(
            message: Message, receiver_address: Address
        ) -> BaseDialogue.Role:
            """Infer the role of the agent from an incoming/outgoing first message

            :param message: an incoming/outgoing first message
            :param receiver_address: the address of the receiving agent
            :return: The role of the agent
            """
            return BaseFipaDialogue.Role.BUYER

        FipaDialogues.__init__(
            self,
            self_address=self_address,
            role_from_first_message=role_from_first_message,
            dialogue_class=FipaDialogue,
        )


class SellerDialogue(FipaDialogue):
    """The dialogue class maintains state of a dialogue and manages it."""

    def __init__(
        self,
        dialogue_label: DialogueLabel,
        self_address: Address,
        role: BaseDialogue.Role,
        message_class: Type[FipaMessage] = FipaMessage,
    ) -> None:
        """
        Initialize a dialogue.

        :param dialogue_label: the identifier of the dialogue
        :param self_address: the address of the entity for whom this dialogue is maintained
        :param role: the role of the agent this dialogue is maintained for

        :return: None
        """
        FipaDialogue.__init__(
            self,
            dialogue_label=dialogue_label,
            self_address=self_address,
            role=role,
            message_class=message_class,
        )
        self.proposal = None  # type: Optional[Description]


class SellerDialogues(FipaDialogues):
    """The dialogues class keeps track of all dialogues."""

    def __init__(self, self_address: Address) -> None:
        """
        Initialize dialogues.

        :return: None
        """
        def role_from_first_message(
            message: Message, receiver_address: Address
        ) -> BaseDialogue.Role:
            """Infer the role of the agent from an incoming/outgoing first message

            :param message: an incoming/outgoing first message
            :param receiver_address: the address of the receiving agent
            :return: The role of the agent
            """
            return FipaDialogue.Role.SELLER

        FipaDialogues.__init__(
            self,
            self_address=self_address,
            role_from_first_message=role_from_first_message,
            dialogue_class=FipaDialogue,
        )

Next, we can immitate a dialogue between the buyer and the seller. We first instantiate the dialogues models:

buyer_address = "buyer_address_stub"
seller_address = "seller_address_stub"
buyer_dialogues = BuyerDialogues(buyer_address)
seller_dialogues = SellerDialogues(seller_address)

First, the buyer creates a message destined for the seller and updates the dialogues:

cfp_msg = FipaMessage(
    message_id=1,
    dialogue_reference=buyer_dialogues.new_self_initiated_dialogue_reference(),
    target=0,
    performative=FipaMessage.Performative.CFP,
    query=Query([Constraint("something", ConstraintType(">", 1))]),
)
cfp_msg.counterparty = seller_addr

# Extends the outgoing list of messages.
buyer_dialogue = buyer_dialogues.update(cfp_msg)
If the message has been correctly constructed, the buyer_dialogue will be returned, otherwise it will be None.

In a skill, the message could now be sent:

# In a skill we would do:
# self.context.outbox.put_message(message=cfp_msg)

However, here we simply continue with the seller:

# change the incoming message field & counterparty
cfp_msg.is_incoming = True
cfp_msg.counterparty = buyer_address
In the skill, the above two lines will be done by the framework; you can simply receive the message in the handler.

We update the seller's dialogues model next to generate a new dialogue:

# Creates a new dialogue for the seller side based on the income message.
seller_dialogue = seller_dialogues.update(cfp_msg)

Next, the seller can generate a proposal:

# Generate a proposal message to send to the buyer.
proposal = Description({"foo1": 1, "bar1": 2})
message_id = cfp_msg.message_id + 1
target = cfp_msg.message_id
proposal_msg = FipaMessage(
    message_id=message_id,
    dialogue_reference=seller_dialogue.dialogue_label.dialogue_reference,
    target=target,
    performative=FipaMessage.Performative.PROPOSE,
    proposal=proposal,
)
proposal_msg.counterparty = cfp_msg.counterparty

# Then we update the dialogue
seller_dialogue.update(proposal_msg)

In a skill, the message could now be sent:

# In a skill we would do:
# self.context.outbox.put_message(message=proposal_msg)

The dialogue can continue like this.

To retrieve a dialogue for a given message, we can do the following:

retrieved_dialogue = seller_dialogues.get_dialogue(cfp_msg)