Check resource alerts for CPU, memory, and disk usage against configured thresholds. It reads and compares; nothing is sent anywhere and nothing is recorded, which is what a running watcher does instead

Show recorded alert and remediation history. Entries are only written while a watcher is running, so an empty list means nothing was recording rather than nothing went wrong

Create a Docker compose backup archive for all services or one service. Volumes are read while the containers run, so a database mid-write can land inconsistent; an archive is not evidence it restores, which is what backup_drill answers

Verify a backup by booting an app in an isolated Docker environment and checking that it responds. A second copy runs beside the live one on a network and port of its own, and everything it made is removed either way. A pass means the archive is not corrupt and the app starts on it, not that every row is there

List existing backup archives in the configured backup directory. It reads names, sizes and dates — that an archive is here says nothing about whether it restores, which is what backup_drill answers

Restore Docker volumes from a backup archive, overwriting the data the app is running on. Destructive: confirm intent before calling. Bind mounts declared by the archive are always refused here, because an agent has no way to name a host path it may write to

Check the config file this server is running on: which file was used, which rule selected it, what was read from each section, and anything wrong or silently ignored

Summarize a Docker container's image, state, restart policy, ports, mounts, networks, and health. Environment variable values are never included

List Docker containers with their status, image, and ports. Stopped containers are included, so a name appearing here is not a name that is running — read state

Get logs from a Docker container: the last lines only, 50 by default, and it returns rather than following the stream

Restart a Docker container by name. It goes down and comes back, and the result says the restart command succeeded, not that the app inside is serving again. Read the logs first if you do not know why it needs restarting

Get resource usage statistics (CPU, memory, network, block I/O) for all running Docker containers

Stop a Docker container by name. Nothing here starts it again: there is no start tool, so the operator brings it back themselves

List the processes running inside a Docker container, read from the host. Nothing runs in the container: no exec, no TTY. Each process's command is the kernel's name for it (ps comm, at most 15 characters), never read from its command line, because a token passed on a command line would otherwise be in the result

Run a read-only diagnosis for resource pressure, stopped containers, public ports, backup hygiene, notifications, and report baseline readiness

Install a self-hosted app via docker compose. Pre-checks docker, ports, and duplicates automatically, and a refusal comes back as a result with the reasons rather than as an error. warning is present when an earlier install of the app left state in Docker volumes this install could not carry over

List available self-hosted apps that can be installed. The catalogue is compiled into the binary, so this reaches no network and answers the same on any machine

Stop an installed app and delete all data including containers, config, and volumes. Nothing here restores it and no backup is taken first: take one before calling if the data matters. A volume that still exists afterwards (external, or used by another container) is listed in kept_volumes, and one Docker could not be asked about in unchecked_volumes

Check the status of an installed app, as its containers report it. An app homebutler did not install is not known here

Stop an installed app and remove its containers. The app directory and its volumes stay on disk; install_purge is the one that deletes them. warning is present when part of the app's state is in Docker volumes outside the app directory, left there by an earlier install

Export server inventory/topology as a Mermaid diagram locally, or JSON locally/remotely. Inside a container a local export is refused, because a container cannot see the machine it is on

Collect server inventory/topology including system status, Docker containers, app ports, and system ports

Scan the local network to discover devices (IP, MAC, hostname). It probes every address on the subnet and takes up to 30 seconds, so it is an answer to a question somebody asked rather than a way to begin. Inside a container it refuses, because the subnet it would scan is the container's own

Send one test notification through every configured channel and report which ones arrived. A real message goes out to each, so anyone reading those channels sees it

List open network ports with associated process information. The process behind a port is not always readable without privilege, and missing_process says so rather than leaving the field quietly empty

List the top processes by CPU or memory, with a total count and any zombies broken out separately

Reboot one explicitly targeted Proxmox guest after confirmation and return the accepted task UPID

Gracefully shut down one explicitly targeted Proxmox guest after confirmation and return the accepted task UPID

Start one explicitly targeted Proxmox guest after confirmation and return the accepted task UPID

List Proxmox QEMU and LXC guests, optionally filtered by node, status, or type

Get detailed Proxmox node status

Render the pinned install command for one Proxmox VE Community Script. Never fetches or runs it; the caller reviews and runs it themselves on the Proxmox host

List the curated Proxmox VE Community Scripts catalog (community-scripts/ProxmoxVE)

Get Proxmox VE version, cluster status, and resources

Inspect one asynchronous Proxmox task by node and opaque UPID

Get the 50 most recent Proxmox tasks for a node

Generate a butler-style health report with snapshot comparison, warnings, notable changes, and suggested actions. It saves a snapshot unless no_save is set, which moves the window every later comparison is measured from — so a loop that calls this leaves nothing to compare against

Get system status including CPU, memory, disk usage, and uptime, for the machine this binary runs on. Inside a container it refuses rather than describe the container as the host; name a server to read a machine over SSH

Send a Wake-on-LAN magic packet to wake a machine. The packet is fire-and-forget: a successful result means it was sent, not that anything woke up

Add a Docker container, systemd unit, or PM2 app to the watch list. This writes the list and nothing begins watching it — a supervisor has to be installed separately, which only the operator can do. Adding a target twice is reported as added=false rather than an error

Run a one-shot restart check on watched targets and report restarts detected since the last check. Only docker targets can be inspected this way; systemd and pm2 targets are reported as skipped rather than assumed healthy

List recorded restart incidents, newest first. Captured logs are excluded unless include_logs is set, because every incident carries a hundred lines of output twice over

List the targets being watched, with their kind and what the last check recorded

Remove a target from the watch list, leaving its recorded incidents in place. Only the list changes: whatever was supervising it keeps running until the operator stops it