Server awareness
without the overhead
Self-hosted monitoring for Docker, Kubernetes, and Proxmox VE — that tells you why something broke, not just what. Smart insights, calm alerts, runs on your hardware.
You run a homelab. Containers go down at 2 AM. Grafana is overkill. Uptime Kuma only does HTTP. You want one tool that monitors everything and tells you what matters.
Everything you need to stay aware
Monitoring, insights, and diagnostics in one lightweight package.
Multi-Runtime Monitoring
Docker, Kubernetes / k3s, and Proxmox VE reporting to the same hub. One agent per Docker host, a DaemonSet on your cluster, or a REST-mode agent for the hypervisor.
Kubernetes-Aware
Cluster events, node conditions as alerts, container resource-limit saturation alerts, PV/PVC inventory, and Ingress auto-discovery into the HTTP endpoint monitor. One DaemonSet, full visibility.
Proxmox VE Hypervisor
Talk to PVE over its REST API from a guest VM — no install on the hypervisor. LXC and QEMU guests appear as containers; ZFS health, storage saturation, cluster quorum, and per-guest backup overdue all become first-class alerts.
HTTP Endpoint Monitoring
Track uptime, response time, and slow checks for your HTTP services. Per-endpoint sparklines, custom intervals and headers, alerts when something goes down. Promote any cluster Ingress with one click.
Smart Insights Engine
Capacity-based health scoring. Flags real saturation, not baseline noise. Time-of-day baselines, predictions, anomaly detection.
Research-Grounded Diagnosis
When something breaks, insightd correlates metrics, logs, restart history, and host state to explain why. Based on Drain, MicroRCA, and S-H-ESD.
Explainable Alerts
15 alert types covering containers, hosts, k8s nodes, and resource-limit saturation. Every alert tells you the value, the threshold, and the context. Webhook delivery to Slack, Discord, Telegram, ntfy.
Disk Forecasting
"X days until full" predictions based on usage trends. Know before you run out of space.
Metric Personalities
Baseline-aware, human-friendly moods on every metric. Your server isn't "at 80% CPU" — it's "Normal for Tuesday evening."
Public Status Page
Shareable uptime page for your services, no authentication required. Enable with a single toggle.
Container Actions
Start, stop, restart, and remove containers from the UI. Remote agent updates via MQTT. All opt-in.
Weekly Digests
HTML email summary of the week: uptime, resource trends, restarts, health score, anomalies, and what needs attention.
Lightweight
~30 MB hub, ~20 MB agent. SQLite storage, no external database. Runs on a Raspberry Pi.
See it in action
A modern, responsive dashboard designed for clarity. Click any screenshot to zoom in.
Up and running in one command
Three containers on one server. To add a host, you run one more agent — that's the only difference.
Requires Docker + Compose v2 on Linux (kubectl for the Kubernetes tab).
curl -sSL https://insightd.org/install.sh | bash
Installs Mosquitto + hub + a local agent with one command. Generates a random MQTT password, runs docker compose up, and waits for the hub to come online. Re-running is safe — existing credentials are preserved.
Prefer to run it yourself?
mkdir -p ~/insightd && cd ~/insightd
# Download the Compose file
curl -sSLO https://raw.githubusercontent.com/goldenproductions/insightd/main/docker-compose.hub.yml
# Write a .env with a generated MQTT password
cat > .env <The script above is ~90 lines of bash. Read it before piping to a shell if you'd rather.
From host to dashboard
Click any piece to see what it does — or add hosts to watch the flow grow.
How insightd compares
The right tool for the homelab sweet spot.
Full monitoring without the config overhead. No PromQL, no dashboards to build, no YAML tuning.
Goes beyond HTTP. Monitors containers, hosts, disks, and explains why things are broken.
Monitoring-first, not management-first. Health scores, baselines, and predictive insights — not just container status.
Frequently asked questions
Is insightd free?
Yes. insightd is MIT licensed and free forever. No paid tiers, no cloud lock-in.
How many servers can I monitor?
Unlimited. Deploy an agent on each server and they all report to a single hub. The lightweight MQTT protocol keeps overhead minimal even at scale.
Does it phone home?
No. Zero telemetry, no cloud dependency, no external API calls. Your monitoring data stays on your hardware.
What are the system requirements?
Minimal. The hub uses ~30 MB of RAM and each agent uses ~20 MB. Storage is SQLite — no external database needed. Runs on anything that runs Docker, including a Raspberry Pi.
Does it support Kubernetes?
Yes. Deploy the agent as a DaemonSet on k8s or k3s — one pod per node. Each pod reports its containers and node-level metrics. Host metrics come from the kubelet, not /proc, so they are accurate even inside a container.
Does it support Proxmox VE?
Yes. Run an agent from any guest VM and point it at PVE's REST API with a token — no install on the hypervisor. LXC and QEMU guests appear as containers, alongside ZFS pool health, storage saturation, cluster quorum, and per-guest backup tracking. Multi-node clusters use one agent process per node.
What notification channels are supported?
Slack, Discord, Telegram, ntfy, and generic webhooks. Plus HTML email alerts and weekly digest summaries.
How does the diagnosis engine work?
When a container is unhealthy, insightd runs seven signal detectors against metrics, robust baselines (MAD z-scores), restart history, host state, Drain-mined log templates, and service topology (Personalized PageRank). The unified diagnoser fuses the results, ranks evidence, and tells you what went wrong and what to try. Based on published research.
Is there a demo?
Demo video and screenshots are being prepared. For now, you can install insightd locally — the setup wizard gets you running in under 5 minutes. It is designed to run on your own hardware, so there is no hosted instance to log into.
Is it a hosted SaaS?
No. insightd runs entirely on your hardware. There is no hosted version, no account to sign up for, and no data leaves your network. If you want a managed service, this is not it.
Is it a Grafana / Prometheus replacement?
No. insightd is not a time-series database and has no PromQL or custom dashboards. It stores SQLite snapshots with 30-day raw retention and 365-day hourly rollups — enough for incident triage and weekly trends, not for high-cardinality long-term metrics. If you already run Prometheus and love it, keep it.
Is it a log aggregator like Loki or ELK?
No. insightd ingests recent container logs for diagnosis and pattern matching (e.g. "ffmpeg fatal" → linked to a restart-spam alert), but it is not a centralized log search platform. There is no cross-host log query language and no long-term log retention.
Does it support multi-tenancy or per-user roles?
No. v0.1 is single-admin: one password, no RBAC, no orgs or teams. It is designed for one trusted operator per hub. If you need fine-grained access control, this is not the right tool yet.
What runtimes are not supported?
Docker, Kubernetes/k3s, and Proxmox VE are first-class today. Bare-metal/systemd services, FreeBSD jails, LXD, Podman rootless, and Nomad are not supported in v0.1. Open an issue if your stack matters to you.
Is it safe to expose to the internet?
Expose the hub UI behind a reverse proxy with TLS and a strong admin password — that part is designed for it. Do not expose the MQTT broker (port 1883) to the public internet; keep it on a LAN, VPN, or Tailscale tailnet. Agent-to-hub authentication is username/password over MQTT, not mTLS.