Agent Asynchronous Loops

Introduction

Agents need to communicate, perform tasks, and respond to events simultaneously and independently within any decentralized system. This guide shows how to create asynchronous agents that operate in parallel, enabling them to handle their own workflows while still interacting with other agents or external processes.

By using asynchronous loops and attaching agents to external event loops, you can build agents that manage tasks simultaneously, send periodic updates, and process incoming messages in real-time. This approach is particularly useful when working with distributed systems, where agents must collaborate or handle multiple simultaneous operations without interruptions.

Supporting documentation

• Creating an agent
• Creating an interval task
• Communicating with other agents
• Agent Handlers

Walk-through

The following scripts show how to define agents, manage their life-cycle and attach them to external asynchronous loops.

Script 1

The first script depicts how to attach an agent to an external event loop and allow it to run simultaneously with other asynchronous tasks.

First of all, let's create a Python script:

windows
echo. > external_loop_attach.py

Now, paste the below code into it:

external_loop_attach.py
 
import asyncio
import contextlib
 
from uagents import Agent, Bureau, Context
 
loop = asyncio.get_event_loop()
 
agent = Agent(
    name="looper",
    seed="<YOUR_SEED>",
    port=8001,
    endpoint=["http://127.0.0.1:8001/submit"],
)
 
bureau = Bureau(
    agents=[agent],
)
 
 
@agent.on_event("startup")
async def startup(ctx: Context):
    ctx.logger.info(">>> Looper is starting up.")
 
 
@agent.on_event("shutdown")
async def shutdown(ctx: Context):
    ctx.logger.info(">>> Looper is shutting down.")
 
 
async def coro():
    while True:
        print("doing hard work...")
        await asyncio.sleep(1)
 
 
if __name__ == "__main__":
    print("Attaching the agent or bureau to the external loop...")
    loop.create_task(coro())
 
    # > when attaching the agent to the external loop
    loop.create_task(agent.run_async())
 
    # > when attaching a bureau to the external loop
    # loop.create_task(bureau.run_async())
 
 

This script is for an agent using an external event loop. We first import the required libraries for this script to be run correctly. We then proceed and instantiate an agent called looper using a seed. Remember that you need to provide a seed within the <YOUR_SEED> field parameter. We then need to create a bureau to manage the agents. We can now add the looper agent into the bureau.

We can proceed and define a startup() function decorated using the .on_event("startup") decorator. The function is triggered when the agent is started and it logs a message indicating the agent has started. Similarly, the shutdown() function is triggered when the agent shuts down, logging an appropriate message.

We go on and define a function coro() which simulates a separate, long-running task that will print "Doing hard work..." every second. This task runs independently of the agent and showcases the agents' ability to handle multiple simultaneous tasks. We now need to attach both the agent and the external task (coro) to the same event loop using loop.create_task(). This allows both the agent and other tasks to execute simultaneously in an asynchronous way.

In the __main__ block, we define a message to be printed indicating that the process of attaching the agent or bureau to the asynchronous event loop has started. The loop.create_task(coro()) adds a coroutine (coro) to the event loop. coro is the task performing "doing hard work..." asynchronously. This allows the task to run simultaneously with other tasks managed by the event loop without blocking the rest of the program.

The agent is attached to the external event loop by using loop.create_task() to schedule the agent's asynchronous operation. The method agent.run_async() is a non-blocking function that runs the agent within the event loop, allowing the agent to perform its tasks and handle events simultaneously.

Finally, in the last line we create a context manager that suppresses KeyboardInterrupt exceptions, which are typically raised when the user presses Ctrl+C to stop the program. This ensures that the program can shut down without printing a traceback or throwing an error when the user stops it manually.

Script 2

The goal of the second script is to create an agent that runs tasks inside an external event loop. The agent can execute certain actions (e.g., print messages or respond to events) while simultaneously performing a separate background task.

Let's start by creating a Python script:

windows
echo. > external_loop_run.py

Then, let's paste the below code into it:

external_loop_run.py
 
import asyncio
 
from uagents import Agent, Bureau, Context
 
loop = asyncio.get_event_loop()
 
agent = Agent(
    name="looper",
    seed="<YOUR_SEED>",
    loop=loop,
    port = 8000,
    endpoint = ["http://127.0.0.1:8000/submit"],
)
 
bureau = Bureau(
    agents=[agent],
    loop=loop,
)
 
 
@agent.on_event("startup")
async def startup(ctx: Context):
    ctx.logger.info(">>> Looper is starting up.")
 
 
@agent.on_event("shutdown")
async def shutdown(ctx: Context):
    ctx.logger.info(">>> Looper is shutting down.")
 
 
async def coro():
    while True:
        print("doing hard work...")
        await asyncio.sleep(1)
 
 
if __name__ == "__main__":
    print("Starting the external loop from the agent or bureau...")
    loop.create_task(coro())
 
    # > when starting the external loop from the agent
    agent.run()
 
 

We start by importing the required libraries to correctly run this script. We then create an asynchronous event loop using asyncio.get_event_loop(). This loop is used to handle all asynchronous operations, such as the agent's actions and background tasks.

We proceed and create an agent called looper using the Agent class. The agent takes three parameters: name, seed, and loop. Remember to provide a seed for your agent otherwise a random address will be generated every time you run the agent. We then create a bureau object using the Bureau class. The bureau is created with a single agent, looper.

We can then define our agent functions to handle the agent's lifecycle events:

1. startup(): This function runs when the agent is started. It logs a message to indicate that the agent has been started up.
2. shutdown(): This function runs when the agent is shut down. It logs a message to indicate that the agent has been stopped.

In the next step, we define the coro() function; As before, this function defines an infinite loop where the agent performs a task ("doing hard work...") every second. This simulates a long-running background task. The await asyncio.sleep(1) pauses execution for one second between each iteration, allowing other tasks to run during that time.

Finally, in the __main__ block, we define a message to be printed indicating that the external event loop is being started. The loop.create_task(coro()) schedules the coro() coroutine to run in the background, simultaneously with the agent's operations.

Expected Output

We are now ready to run the scripts.

The output should be similar to the following:

• Script 1:

  Attaching the agent or bureau to the external loop...
  

• Script 2:

  Starting the external loop from the agent or bureau...
  INFO:     [looper]: >>> Looper is starting up.
  INFO:     [looper]: Agent inspector available at https://agentverse.ai/inspect/?uri=http%3A//127.0.0.1%3A8000&address=agent1qwep424538eh7fcruqcnx8la3q3tgl4tgksrcdtahqs7dqgs4rewsx4jefu
  INFO:     [looper]: Starting server on http://0.0.0.0:8000 (Press CTRL+C to quit)
  doing hard work...
  doing hard work...
  doing hard work...
  doing hard work...
  doing hard work...