Use this pattern when the agent should respond to changing system state, not only static prompts.

• service monitoring assistants
• codebase watchers
• runtime health agents
• agents that combine observation with guarded actions

from loom import Agent, Model, Runtime, RuntimeSignal, SessionConfig

agent = Agent(
    model=Model.anthropic("claude-sonnet-4"),
    instructions="Monitor the system and recommend safe next steps.",
    runtime=Runtime.long_running(criteria=["recommendations are safe and actionable"]),
)

session = agent.session(SessionConfig(id="ops"))

await session.receive(
    RuntimeSignal.create(
        "CPU usage exceeded 90%",
        source="heartbeat",
        type="alert",
        urgency="high",
        payload={"host": "api-1"},
    )
)

result = await session.run("Assess pending runtime signals")

Monitoring enters the runtime as signal input.

That keeps the kernel simple:

• heartbeat adapters emit RuntimeSignal
• gateway and cron adapters emit the same contract
• AttentionPolicy decides whether a signal should cause execution
• tools and remediation still go through Capability and GovernancePolicy

• combine with SignalAdapter for real heartbeat, webhook, or gateway event shapes
• combine with Capability.shell(require_approval=True) when remediation is allowed
• use Runtime.supervised(...) when human review is required

• examples/12_heartbeat_and_safety.py
• examples/16_signal_adapters.py