Azure / AutoGen Adapter

Wraps AutoGen v0.4 message delivery with LOCO scheduling. Supports both send_message() (point-to-point) and publish_message() (pub/sub).

Setup

from loco import AsyncLOCOScheduler, SharedResource
from loco.adapters.autogen import AutoGenAdapter

scheduler = AsyncLOCOScheduler([], SharedResource("azure_openai", capacity=3))
adapter = AutoGenAdapter(scheduler, default_model="gpt-4o")

Usage

Point-to-Point

result = await adapter.send_message(
    sender="coordinator",
    recipient="analyst",
    content="Analyze this security event",
    model="gpt-4o-mini",
)

Pub/Sub

results = await adapter.publish_message(
    sender="coordinator",
    topic="security-alerts",
    content="New threat detected",
    subscribers=["analyst-0", "analyst-1", "analyst-2"],
)

Each subscriber delivery is scheduled independently through LOCO.

Model Weights

Model	Weight
gpt-4o	3.0
gpt-4o-mini	1.0
gpt-4	3.0
gpt-35-turbo	1.0
Unknown	2.0 (default)

Demo

See loco-autogen-demo -- 8 AutoGen agents in an enterprise security pipeline with per-agent budget enforcement. Shows reject, alert, and downgrade modes.