Pulse Patrol

The rounds you keep meaning to do

Alerts catch what you thought to write a rule for. Patrol goes looking for the rest on a schedule, from every ten minutes to every seven days, and tells you what it found in plain English. It runs on every tier, including the free one, using your own LLM key or a local model.

Start Free Try the Live Demo

Pulse Patrol finding · Just now

Warning

Backup job "pbs-daily" failing

Last 3 runs failed with "connection refused". PBS at 192.168.1.50 may be unreachable.

Pulse Patrol finding · 4 hours ago

Critical

ZFS pool "tank" 94% full

At current growth rate, pool will be full in 3 days. ZFS performance degrades significantly above 80%.

Why this isn't just another chatbot

Your infrastructure, not just a chat window

When you ask a general chatbot about your server, you have to describe everything manually. Patrol already sees your entire estate at once, with historical context no generic LLM can reach.

Full infrastructure state

Every Proxmox node, Docker host, Kubernetes workload, TrueNAS pool, vSphere cluster, backup job and ZFS pool you have connected, held in one context.

Historical intelligence

Trends over 24 hours and 7 days, baselines learned from your environment, capacity predictions and anomaly detection.

Operational memory

Your notes, the alerts you dismissed, the fixes that worked before, and the behaviour you told it to treat as expected.

This is what makes Patrol catch issues that static alerts miss.

It's not just "CPU > 90%". It's "CPU spiked to 85% but that's normal for this VM during backups, however the backup hasn't completed in 3 hours which is unusual."

Pro exclusive

The 3am problem, solved

Traditional alerts wake you up with noise. Pulse wakes you up with answers. Pro has Patrol investigate each alert the moment it fires, propose the exact command, and run it only under the approval policy you choose.

● Traditional monitoring · 3:14 AM

⚠️ ALERT: CPU > 90%
⚠️ ALERT: Memory > 85%
⚠️ ALERT: IO Wait > 20%
...and 12 more

"Great, everything is on fire. Where do I even start?"

● Patrol investigated alert · 3:14 AM

Root Cause: Runaway backup process

PBS backup job for VM 102 (postgres) is stuck in a write loop, consuming 94% CPU and causing IO saturation. All other alerts are symptoms.

→ One root cause, not 15 alerts

↓

Then Patrol offers to fix it

Pulse Assistant

Investigating patrol finding

That backup job for VM 102 (postgres) is stuck on pve1. The PBS task UPID:pve1:00003A2F has been running for 6+ hours with no progress.

Recommended action: Kill the stuck task to free resources.

PROPOSED FIX

                                pvesh delete /nodes/pve1/tasks/UPID:pve1:00003A2F
                            

✓

Fix applied successfully

Task terminated. CPU usage returning to normal.

Commands require explicit approval. You stay in control.

Skip the 20-minute triage. One root cause, one command, verified.

Autonomy

You choose how far it goes

Agent commands are disabled by default. Nothing runs until you turn it on and pick a mode, and every action is verified and written to the audit trail.

Ask first

Patrol investigates and proposes the exact command. Nothing runs until you press apply. This is where most people stay, and that is fine.

Safe auto-fix

Patrol carries out the low-risk, reversible fixes on its own and still asks before anything that could bite. You read about it afterwards instead of at 3am.

Autopilot

For environments where you have decided the command safety policy is doing the work. Still bounded by that policy, still verified, still logged.

Bring your own agent

Prefer Claude Code or another MCP client? Pulse ships an MCP server exposing the same governed tools the assistant uses, under Settings → Pulse Intelligence.

Patrol findings are free. Letting Patrol act is Pro.

Every install gets scheduled findings with your own LLM key. Pro is what turns a finding into an investigation and a verified fix.