Pulse catches a dead agent before your customers do.

Find the silent failure
first.

A web service crashes loudly. An agent just stops firing, or returns the wrong answer with the same confidence as the right one. Pulse watches the agent itself: heartbeats, full run traces, and an alert path your on-call actually respects.

support-triage · schedulelive
support-triage missed its 14:00 run paged on-call · last ok 13:00 · 247 runs clean before

heartbeats

Know the moment an agent goes quiet.

Tell AgentPing the schedule each agent should keep. Miss a window and you get paged in seconds, with the last good run and what was meant to happen already attached. Heartbeats are for agents on a schedule; request-driven agents get the same watch through traces and alerts.

  • Expected schedules per agent, down to the cron expression
  • Paged on a missed run, not hours later when a customer notices
  • Every page carries the last successful run for instant context
support-triage · hourlylive
missed the 14:00 run paged on-call · last ok 13:00 · 247 clean runs before

alerting

One ping when it breaks, one when it recovers.

Set rules for what matters: an agent goes quiet, runs start failing, health drops, quality drops, or spending runs hot. AgentPing opens one issue and tells you once. While it stays open it does not page you again; when it clears, you get a single all-clear. Reminders only if you haven't acknowledged. We page you, we don't spam you.

  • Rules scoped to one agent or a whole team
  • Email, Slack, Discord, Telegram, webhook, plus Teams and PagerDuty on Team and up
  • An Issues page that answers "what is wrong right now" in one place
issues · open2 open
invoicer3 issuesruns failing · health 71%
news-fetcherquietoverdue 18m · last ok 02:00
delivered to email, Slack, Telegram acknowledge from the message · resolves itself when it clears

traces

See every step the agent took.

Every run records its tool calls, timings, the model used, and the output it produced. When one agent hands off to another, the trace follows, so a multi-step failure is one story, not five disconnected logs.

  • Run-level traces with tool calls, timings, and outputs
  • Step through any run to see where it went wrong
  • Multi-agent trace propagation across handoffs, no lost context
live · agent activitystreaming
daily-summary2m 03sclaude-sonnet-4-5 · $0.084
news-fetchertimeoutgpt-5 · n/a
support-triage8sclaude-haiku-4-5 · $0.003
research-agentexit 1gpt-5 · $0.002
content-writer3m 41sclaude-opus-4-7 · $0.156

alerting

Alerts your on-call actually respects.

When a run fails or a scheduled agent goes silent, the alert lands in the channel your team already lives in, with the error and the last good run attached. No new dashboard to babysit.

  • Route alerts to Slack, PagerDuty, Microsoft Teams, email, or webhook
  • Every alert carries the failing run and its last successful one
  • The SDK never blocks your agent, even when we are unreachable
alerts1 firing
research-agent failed: exit 1 14:02 · last ok 13:54 · PagerDuty + #ai-ops
news-fetcher: timeout 13:58 · provider latency · email

status pages

The page you send your client.

A public, link-shareable status page for the agents you run: operational, degraded, or down, with 90 days of history per agent. The states come from real runs, the same signal that pages your on-call, so you never declare an incident by hand. Your client sees their agents are healthy without a seat in your dashboard.

  • Unguessable link; add a passcode for a page you keep between you and one client
  • States, never data: no payloads, costs, run counts, or customer ids can leak
  • Regenerate the link, change the passcode, or disable the page to revoke access instantly
  • 1 page on Starter, 5 on Team, unlimited on Business

Status page docs

status · acme-corp agentslive
support-triageoperational90 days · 99.8%
daily-summaryoperational90 days · 100%
invoicerdegradedsince 09:14 · recovering

Built so nothing fails quietly.

Everything you need to trust that the agents you shipped are still running, and to find out first when they are not.

traces

Trace propagation

Follow the thread when one agent kicks off another, so a five-step failure is one trace to read, not five logs to correlate by hand.

sharing

A status page for clients

A public page showing each agent operational, degraded, or down with 90 days of history. States, never data, so nothing sensitive can leak.

detail

Latency and tool calls

Every tool call, timing, and output captured, so a slow run is a known quantity.

status

Status at a glance

Every run lands as success, failed, timeout, or cancelled, so a bad deploy shows up immediately.

coverage

Webhook for anything

One curl monitors cron, n8n, GitHub Actions, or Claude Routines. If it can curl, it counts.

safety

Never blocks your agent

Hard 2-second timeout, bounded queue. If we go down, your agent runs as if we were not there.

From silence to signal.

Know it fired

Schedules and heartbeats turn "I think it is fine" into a green tick or a page.

See what it did

Step through the run, read the tool calls, find the exact step that went wrong.

support-triage8s$0.003
research-agentexit 1$0.002

Page the right people

Failures route to the channel your on-call already trusts.

research-agent failedexit 1 · PagerDuty

Questions, answered.

Will the SDK slow my agents down?
No. Telemetry runs off the hot path with a hard 2-second timeout and a bounded local queue. If AgentPing is unreachable, your agent runs exactly as if it were not installed. The SDK never blocks and never crashes your code.
How does heartbeat monitoring work?
You tell AgentPing the schedule an agent is meant to keep. If a scheduled run does not arrive in its window, Pulse pages you, with the last successful run and what was supposed to happen attached.
What can I see in a run trace?
Every run records its tool calls, latency, the model used, and the output it produced. Step through any run to see exactly what it did, and follow the trace when one agent hands off to another.
Where do alerts go?
Slack, PagerDuty, Microsoft Teams, email, or any webhook. Each alert carries the failing run and its last successful one, so on-call gets context, not just a red light.
Can I monitor an agent that has no SDK?
Yes. A single authenticated webhook, one curl, works from cron, n8n, GitHub Actions, Claude Routines, or anything that speaks HTTP. If it can curl, it is monitored.
Can I show a client their agents are healthy without giving them a login?
Yes. A status page is a public, link-shareable view of the agents you choose: operational, degraded, or down, with 90 days of history each. Add a passcode to keep it private, and revoke access instantly by regenerating the link. It shows states, never payloads or costs. One page on Starter, five on Team, unlimited on Business.

Point Pulse at your agents. See it in minutes.

Pulse catches a dead agent before your customers do. Two lines of code, or one curl. Live in minutes, free to start, 14-day trial on paid plans.

Free to start. No card. The SDK never blocks your agents.