Arch Linux User Repository

Heimdall is a lightweight, cross-platform hardware monitoring system with a real-time terminal dashboard. Unprivileged daemons stream metrics from every host over a low-bandwidth gRPC link to a central hub; a btop/mactop-class Go TUI renders the fleet live. It sees (metrics) and hears (opt-in logs) across every realm.

Point it at your homelab, GPU boxes, a rack of servers, or just your laptop, and watch CPU, memory, disk, network, temperature, GPU, and power from every machine in one terminal.

Why Heimdall?

• One terminal for the whole fleet — not a separate btop/mactop window per box.
• Lightweight — four small static binaries. No Prometheus + Grafana + exporters + a time-series database to stand up (though Heimdall speaks OpenMetrics if you already run them).
• Unprivileged by default — hosts need no inbound ports and no root. Power, GPU, and thermal come from no-sudo paths where they exist, and an optional helper where they don't.
• Built for distance — compact gRPC over a low-bandwidth link; happy across a VPN or Tailscale.
• Zero-config when you want it — daemons can discover their hub over mDNS; tag hosts to organize a growing fleet; alert on thresholds; scrape it all into your existing Grafana.

Capabilities

Feature codenames are Norse — see the glossary for what they mean and how to say them.

• Cross-platform daemon — Windows/macOS/Linux × amd64/arm64; enrolls over TLS, auto-reconnects, low bandwidth.
• SOLID metric adapters — CPU (with per-core), memory, disk + disk I/O, network throughput, temperature, GPU, power, internet + per-NIC gateway latency, uptime, host context. New signals add without touching existing adapters; a failing adapter is isolated, never dropping the host.
• Unprivileged by default — power, GPU, and thermal from no-sudo paths (Apple Silicon IOReport, Linux RAPL + hwmon) and an optional helper for the rest; adapters self-report unavailable / needs-helper rather than failing.
• Real-time TUI — live grid, per-host detail, gradient gauges, stale/offline with last-known values; high-fidelity and graceful-degradation render modes.
• Bifröst federation — a hub relays upstream (local → cloud); multiple dashboards subscribe to one hub.
• Ratatoskr discovery — daemons find their hub over mDNS (--hub auto); no hardcoded address on a LAN.
• Realms tags — tag hosts and hubs (env=prod, region=apac); hub tags inherit to their hosts, and the fleet is groupable by tag or origin hub.
• Mímir export — a Prometheus/OpenMetrics /metrics endpoint over the whole fleet, plus short-range history, so it drops into an existing Grafana.
• Gjallarhorn alerting — declarative threshold rules with hysteresis; fire to a webhook on breach and clear.
• Read-only control plane — allow-listed, no-sudo remote queries, audited.
• Opt-in log streaming — rate-limited, on its own channel.

Quick start

Monitor your own machine — build, then run all three pieces locally:

make build-tui
./bin/heimdall-hub &                                               # central server (:9090)
./bin/heimdall-daemon --hub localhost:9090 --name "$(hostname)" &  # collector
./bin/heimdall-dashboard                                           # the TUI (subscribes to :9090)

Just want to see the interface? ./bin/heimdall-dashboard --demo renders a simulated fleet — no hub or daemon needed.

→ Full walkthrough: Quickstart.