Multi-Agent Collaboration

Pygent can coordinate multiple agents in parallel using the TaskManager class. This approach resembles systems like Crew AI where tasks are distributed to specialised agents.

Defining personas

Create a list of Persona objects representing the roles of the agents that will form the "crew":

from pygent import Agent, TaskManager
from pygent.persona import Persona

personas = [
    Persona("writer", "produces files"),
    Persona("reviewer", "checks the result"),
]

manager = TaskManager(personas=personas)

Delegating tasks

Use start_task to launch background subtasks. Specify the desired persona and optionally send supporting files:

main = Agent()

# Send a writer agent to generate a file
writing = manager.start_task(
    "write_file path='note.txt' content='hi'\nstop",
    main.runtime,
    persona="writer",
)

# Another agent reads the file and prints it to the console
reading = manager.start_task(
    "bash cmd='cat note.txt'\nstop",
    main.runtime,
    files=["note.txt"],
    persona="reviewer",
)

Monitoring and collecting results

The TaskManager lets you check task status and copy files back to the main agent:

import time
for tid in [writing, reading]:
    while manager.status(tid) == "running":
        time.sleep(1)
    print("Task status", tid + ":", manager.status(tid))

print(manager.collect_file(main.runtime, writing, "note.txt"))
main.runtime.cleanup()

With these calls you can build chained or parallel workflows, achieving behaviour similar to multi-agent frameworks like Crew AI.