Skip to main content

🤖 👥 Agent Teams

Agent Teams is a Python library for orchestrating teams of AI agents built on pydantic-ai with coordination over the A2A protocol. It brings production-grade patterns from distributed systems to multi-agent coordination.

Key Features​

  • Team Lifecycle Management — Create, start, stop, pause, resume, and delete agent teams via a REST API or CLI.
  • Execution Modes — Sequential, parallel, and supervisor-driven task execution.
  • Health Monitoring — Heartbeat-based liveness detection, stuck-agent identification, mass death handling, and exponential-backoff restarts.
  • Reaction Engine — Automatic event handling with retry logic, escalation, and configurable responses (notify, reassign, restart).
  • Lifecycle Hooks — Priority-ordered, short-circuit-capable hooks for every major team event.
  • A2A Protocol Extensions — Team coordination metadata as a first-class A2A extension.
  • Typer CLI — Full-featured command-line interface for managing teams without writing code.
  • FastAPI REST API — Embeddable or standalone API server with SSE event streaming.

Architecture at a Glance​

Installation​

pip install agent-teams

For development (with test dependencies):

pip install -e ".[test]"

Quick Example​

from agent_teams.types import TeamConfig, AgentMemberConfig, ExecutionMode
from agent_teams.manager import TeamManager

manager = TeamManager()

# Create a team
config = TeamConfig(
    name="research-team",
    execution_mode=ExecutionMode.PARALLEL,
    members=[
        AgentMemberConfig(name="researcher", model="openai:gpt-4o"),
        AgentMemberConfig(name="writer", model="anthropic:claude-sonnet-4-20250514"),
    ],
)
team = await manager.create_team(config)

# Start and assign work
await manager.start_team(team.team_id)
result = await manager.assign_task(team.team_id, "Analyse recent AI papers")

Or use the CLI:

# Start the server
agent-teams serve --port 8765

# Create and run a team
agent-teams create team.yaml
agent-teams start <team-id>
agent-teams assign <team-id> "Analyse recent AI papers"

Documentation​