Monitoring live agents

The Agents view answers a question traces alone cannot: is the agent process alive right now? Every instrumented process with the heartbeat enabled reports in, and Rius classifies each instance by liveness:

The Agents view with instances in every status

Instances group under their agent name; the chips at the top filter by status. Open trace jumps from a running instance straight to the trace it is working on.

Enable heartbeats

Heartbeats are off by default in this release. Turn them on at init:


import glassflow
 
glassflow.init(
    api_key="glassflow_...",
    service_name="support-agent",
    heartbeat=True,                # or GLASSFLOW_HEARTBEAT=true
)

heartbeat_interval sets the seconds between pings (default 15, clamped to 5-300). The staleness thresholds below are derived from it, so a slower interval also slows detection.
agent_name sets the identity instances group under; it defaults to service_name. Use it when several services are one logical agent.
Every process is one instance: a fresh instance ID per process start, so three replicas of the same agent show as three instances under one name.

The full parameter list is in the SDK reference.

Status semantics

Status is computed server-side from the last heartbeat; the client is never trusted for it.

Status	Meaning	When
`running`	Alive and mid-run	Recent ping and at least one open trace
`ready`	Alive and idle	Recent ping, no open traces
`stale`	Missed pings	No ping for ~2 intervals (30s at the default)
`gone`	Presumed dead	No ping for ~4 intervals (60s at the default); no clean shutdown was recorded
`stopped`	Clean shutdown	The process exited normally and sent its final ping

The difference between stopped and gone is the whole point: stopped means the process said goodbye; gone means it vanished mid-flight (crash, OOM kill, network partition) and would have kept pinging otherwise. Records age out a day after the last ping, so a gone instance stays visible long enough to investigate.

Heartbeats are a liveness and readiness signal for agent processes, not stuck-agent detection. A process that is alive but wedged mid-run keeps pinging and shows running; detecting frozen runs is a separate, planned capability.

Next steps

Viewing traces: follow a running instance’s open trace.
Installation and configuration: every init option, including the heartbeat knobs.