v4.7.0

RemotePower v4.7.0 — "IntegrationsMatters"

A reach-outward release. RemotePower now monitors the popular software your homelab and fleet already run, and the agent can run as a container to watch a Docker host without installing anything on it. No breaking changes; everything existing keeps working.

After upgrading, hard-reload the dashboard once (service-worker cache remotepower-shell-v4.7.0).

Homelab software integrations

A new server-side integration subsystem polls popular self-hosted software for health on a cadence and folds the result into the Alerts inbox and the dashboard. Read-only; nothing is installed on the target. Set one up under Settings → Integrations: pick a type, point it at the service on your LAN, add an API token, and Test.

26 connectors across the stack:

• DNS: Pi-hole (v6), AdGuard Home
• Storage / NAS: TrueNAS (CORE/SCALE), Unraid
• Virtualization / orchestration: Kubernetes / k3s, VMware vCenter/ESXi, Proxmox Backup Server
• Network: UniFi Network
• Reverse proxy / certs: Traefik, Nginx Proxy Manager, Caddy
• Observability: Netdata, Grafana, Uptime Kuma
• Media: Jellyfin, Plex
• Apps: Home Assistant, Nextcloud
• Download clients: qBittorrent, Transmission, Deluge, SABnzbd, NZBGet
• Media automation: Servarr (one connector for Sonarr / Radarr / Prowlarr / Lidarr), Bazarr
• Requests: Overseerr / Jellyseerr

An unhealthy or unreachable target raises an integration_down alert (severity from the result; auto-resolved on recovery) that routes through your channels. Health also surfaces as an Integration health dashboard widget and live status badges on each card.

Security: every outbound call goes through the SSRF guard (loopback / link-local / cloud-metadata refused, RFC1918 LAN allowed; peer IP re-validated at connect time; redirects refused). Credentials are stored server-side and redacted from every API response; the raw URL is admin-only. See integrations.

Enterprise declutter: a Show Homelab software switch (Settings → Integrations, default on) is an instance-wide kill switch — unchecking it stops polling, hides the configuration section, and removes the dashboard widget.

Containerized agent

The Linux agent can now run as a container that monitors its Docker host and reports to the server — no package install on the host. In the UI: Enroll device → Generate Docker compose, then docker compose up -d. The agent reads the host's facts (shared PID/network namespaces, host rootfs mounted read-only), names itself after the host, and persists its credentials in a volume. Published as ghcr.io/tyxak/remotepower-agent (multi-arch amd64+arm64).

Standard capabilities, no --privileged (SMART/DMI is an opt-in profile; container inventory via the Docker socket is opt-in with a host-root warning). Host package inventory + CVE scanning read the host package DB directly; host compliance scans and systemd/journal collection are honestly reported as unavailable from a container. See docker-agent.

Fleet GPU page

A new Monitoring → GPUs page shows every GPU across the fleet in one rich view — NVIDIA and AMD — with utilisation and VRAM meters, temperature, power and fan, hottest/busiest first, plus a fleet summary (count, per-vendor, total power). Agents report via nvidia-smi / rocm-smi; AMD also has a tooling-free amdgpu sysfs fallback (so a host with no ROCm tooling still reports utilisation, VRAM and temperature). NVIDIA telemetry gains fan speed.

Each GPU card also carries temperature + utilisation trend sparklines (the last ~4 hours of samples), so you can see a GPU heating up or a job ramping without opening the host.

Thermal alerting. A GPU at or above the temperature threshold (default 85 °C, configurable) raises the standard high-temperature alert and auto-resolves when it cools. It reuses the existing hardware-temperature alert — GPU sensors simply participate alongside the CPU/board sensors — so there's no new alert type and your existing routing applies unchanged.

Unmonitored devices now show their data everywhere

Telemetry and inventory views — thermal, power, storage, exposure, predictive-health / SMART, patches, listening ports, processes and the new GPU page — now display unmonitored hosts too, flagged so the UI marks them as unmonitored. Only alerting stays suppressed for unmonitored devices (the same gate the alert pipeline already used), so you can still see an unmonitored host's data without it paging you.

CSP report hygiene

The in-app CSP violation reporter now ignores reports whose source is a browser extension (moz-extension://, chrome-extension://, safari-web-extension://, and the like). Those violations come from a user's installed extensions, not from RemotePower, so dropping them keeps the security log free of noise the app didn't cause.

---

Upgrade is in-place; see upgrading. Older release notes live in CHANGELOG.