Does the AI run commands on my servers by itself?

Read-only investigation runs on its own (metrics, process lists, services, logs). Anything that changes a machine stops and waits for a person to approve the exact action. Tier-3 actions always need that click.

What can I see afterwards?

Every tool call, the approval and who gave it, in the audit log and in the AI session's own ledger. Nothing about how it reached the answer is hidden.

Is this a real model or a scripted demo?

A real model with real tool calls against a live Linux container in the lab. The fault (a runaway "report-indexer" process) was staged on purpose; the investigation and the fix are the assistant's own.

Which operating systems does this work on?

Any device with the Breeze agent: Windows, macOS and Linux. The demo uses a Linux server.

Per-token AI usage is metered per client with caps and bill-back. See the AI usage and budgets video.

I Asked My RMM's AI Why a Server Was Slow. It Found the Runaway Process

Name: I Asked My RMM's AI Why a Server Was Slow. It Found the Runaway Process | Breeze RMM
Uploaded: 2026-10-23
Duration: 2 min 19 s

A live Linux server has been dragging all morning and nobody knows why. Instead of opening a shell, the tech hands it to the Breeze RMM assistant in plain words: it's slow, find out what's going on, stop whatever is misbehaving, tell me what changed. Watch the assistant pull the device context and the last day of metrics, run real commands through the agent (the live process list, the services), find the runaway indexer pegging the CPU, then stop and ask for approval before it kills anything. After the click the process is gone, the assistant reports what it did and why, and the audit log has every command, the approval and who gave it.

2:19 Captions: English Open on YouTube

What's in the demo

The device page for an online Linux server with CPU pinned for a while
Opening the assistant from the top bar on that device
The brief, in plain words: slow server, find out why, stop it, report back
Tool calls shown as they happen: device context, metrics, process list, services
The assistant stops on its own before a destructive change and shows the process it wants to kill
Approving the kill, then the plain-English report of what changed
The container really is back to normal after the approval
Audit log: every command the assistant ran, the approval, the approver, all timestamped

Questions this video answers

Does the AI run commands on my servers by itself?: Read-only investigation runs on its own (metrics, process lists, services, logs). Anything that changes a machine stops and waits for a person to approve the exact action. Tier-3 actions always need that click.
What can I see afterwards?: Every tool call, the approval and who gave it, in the audit log and in the AI session's own ledger. Nothing about how it reached the answer is hidden.
Is this a real model or a scripted demo?: A real model with real tool calls against a live Linux container in the lab. The fault (a runaway "report-indexer" process) was staged on purpose; the investigation and the fix are the assistant's own.
Which operating systems does this work on?: Any device with the Breeze agent: Windows, macOS and Linux. The demo uses a Linux server.
What does it cost?: Per-token AI usage is metered per client with caps and bill-back. See the AI usage and budgets video.

A note on the demo: the MSP, the client and the server are fictional lab data. The slow-server fault was staged in a container for the recording; the assistant's investigation, its proposed fix and the approval are real and unscripted, so your own run may take a different route to the same answer. Where the model took longer than the narration, the picture is sped up; nothing is cut.

Do this on your own fleet

Everything in the video is in the product you get. Start the Cloud Beta in minutes, or self-host the open-source Community edition for free.

Start Cloud Beta Deploy for Free Book a Demo