Multi-agent communication is the core functionality of any Multi-Agent System. Unlike in all other frameworks, Agency Swarm not only allows you to define communication flows in any way you want (uniform communication flows), but also to configure the underlying logic for this feature. This means that you can create entirely new types of communication or adjust it to your own needs. Below you will find a guide on how to do all this, along with some common examples.

​
Pre-Made SendMessage Classes

Agency Swarm contains multiple commonly requested classes for communication flows. Currently, the following classes are available:
Class NameDescriptionWhen to UseCode Link
SendMessage (default)This is the default class for sending messages to other agents. It uses synchronous communication with basic COT (Chain of Thought) prompting and allows agents to relay files and modify system instructions for each other.Suitable for most use cases. Balances speed and functionality.link
SendMessageQuickA variant of the SendMessage class without Chain of Thought prompting, files, and additional instructions. It allows for faster communication without the overhead of COT.Use for simpler use cases or when you want to save tokens and increase speed.link
SendMessageAsyncThreadingSimilar to SendMessage but with async_mode='threading'. Each agent will execute asynchronously in a separate thread. In the meantime, the caller agent can continue the conversation with the user and check the results later.Use for asynchronous applications or when sub-agents take significant amounts of time to complete their tasks.link
SendMessageSwarmInstead of sending a message to another agent, it replaces the caller agent with the recipient agent, similar to OpenAI’s Swarm. The recipient agent will then have access to the entire conversation.When you need more granular control. It is not able to handle complex multi-step, multi-agent tasks.link
To use any of the pre-made SendMessage classes, simply put it in the send_message_tool_class parameter when initializing the Agency class: 
from agency_swarm.tools.send_message import SendMessageQuick

agency = Agency(
    ...
    send_message_tool_class=SendMessageQuick
)
That’s it! Now, your agents will use your own custom SendMessageQuick class for communication.

​
Creating Your Own Unique Communication Flows

To create your own communication flow, you will first need to extend the SendMessageBase class. This class extends the BaseTool class, like any other tools in Agency Swarm, and contains the most basic parameters required for communication, such as the recipient_agent.

​
Default SendMessage Class

By default, Agency Swarm uses the following tool for communication: 
from pydantic import Field, field_validator, model_validator
from .SendMessageBase import SendMessageBase

class SendMessage(SendMessageBase):
    """Use this tool to facilitate direct, synchronous communication between specialized agents within your agency. When you send a message using this tool, you receive a response exclusively from the designated recipient agent. To continue the dialogue, invoke this tool again with the desired recipient agent and your follow-up message. Remember, communication here is synchronous; the recipient agent won't perform any tasks post-response. You are responsible for relaying the recipient agent's responses back to the user, as the user does not have direct access to these replies. Keep engaging with the tool for continuous interaction until the task is fully resolved. Do not send more than 1 message to the same recipient agent at the same time."""
    my_primary_instructions: str = Field(
        ...,
        description=(
            "Please repeat your primary instructions step-by-step, including both completed "
            "and the following next steps that you need to perform. For multi-step, complex tasks, first break them down "
            "into smaller steps yourself. Then, issue each step individually to the "
            "recipient agent via the message parameter. Each identified step should be "
            "sent in a separate message. Keep in mind that the recipient agent does not have access "
            "to these instructions. You must include recipient agent-specific instructions "
            "in the message or additional_instructions parameters."
        )
    )
    message: str = Field(
        ...,
        description="Specify the task required for the recipient agent to complete. Focus on clarifying what the task entails, rather than providing exact instructions. Make sure to include all the relevant information needed to complete the task."
    )
    message_files: list[str] | None = Field(
        default=None,
        description="A list of file IDs to be sent as attachments to this message. Only use this if you have the file ID that starts with 'file-'.",
        examples=["file-1234", "file-5678"]
    )
    additional_instructions: Optional[str] = Field(
        default=None,
        description="Additional context or instructions from the conversation needed by the recipient agent to complete the task."
    )

    @model_validator(mode='after')
    def validate_files(self):
        # prevent hallucinations with agents sending file IDs into incorrect fields
        if "file-" in self.message or (self.additional_instructions and "file-" in self.additional_instructions):
            if not self.message_files:
                raise ValueError("You must include file IDs in message_files parameter.")
        return self


    def run(self):
        return self._get_completion(message=self.message,
                                    message_files=self.message_files,
                                    additional_instructions=self.additional_instructions)
Let’s break down the code. In general, all SendMessage tools have the following components:
  1. The Docstring: This is used to generate a description of the tool for the agent. This part should clearly describe how your multi-agent communication works, along with some additional guidelines on how to use it.
  2. Parameters: Parameters like message, message_files, additional_instructions are used to provide the recipient agent with the necessary information.
  3. The run method: This is where the communication logic is implemented. Most of the time, you just need to map your parameters to self._get_completion() the same way you would call it in the agency.get_completion() method.
When creating your own SendMessage tools, you can use the above components as a template.

​
Common Use Cases

For detailed Common Use Cases, please refer to the Common Use Cases subpage.

​
Conclusion

Agency Swarm has been designed to give you, the developer, full control over your systems. It is the only framework that does not hard-code any prompts, parameters, or even worse, agents for you. With this new feature, the last part of the system that you couldn’t fully customize to your own needs is now gone! So, we want to encourage you to keep experimenting and designing your own unique communication flows. While the examples above should serve as a good starting point, they do not even merely scratch the surface of what’s possible here! We are looking forward to seeing what you will create. Please share it in our Discord server so we can all learn from each other.