github SikamikanikoBG/homelab-monitor v0.39.0
v0.39.0 — see where your watts went

35 minutes ago

Contributions in this release: @haigou-web (three PRs), @WaldInsel (first PR) and @jmdjr, with reviews from @pehota. Thank you.

Added

  • Windows hosts show their Docker containers (#279, #345, thanks @haigou-web). probe.ps1 now reads Docker Desktop the way probe.py reads Docker on Linux — the container list, CPU %, RAM, writable-layer disk and restart policy — so a Windows host's Containers tab is no longer empty. A host without Docker, or with a daemon that hangs, simply has no Containers tab, as on Linux.
  • AI Models → Hall of Fame (#23, #347, thanks @haigou-web). Every model that has held the GPU over the selected range, ranked by how long it stayed loaded.
  • Costs → Energy used: kWh by hour and by day, over time. The busy-hours heatmap folds history into an average week; this chart keeps the timeline — the energy the machine actually consumed (GPU + CPU + DRAM) each hour for the last 7 days, or each day for the last 90, with the cost for that hour or day next to it at your tariff. Gaps show when nothing was recorded, a young hub's chart starts at its first reading, and it works per host and without a price set. (GET /api/costs/energy?by=hour|day&days=N&host=)
  • Make the sidebar yours: reorder tabs and hide the ones you don't use (#33). ✎ Customize at the foot of the sidebar turns the nav into an editor: drag a tab to move it within its group (Monitoring, AI, …), or focus it and press Alt+↑/↓; the eye hides or shows it. No AI servers? Hide the whole AI group and its heading goes too. Settings-group tabs can be moved but not hidden, so the config is always one click away. A hidden tab isn't locked away: its URL and the search box still open it. The layout lives in this browser only (hl_tab_order / hl_tab_hidden), and Settings → General → Sidebar layout resets it. Tabs added by a future release still show up for you, at the end of their group.

Changed

  • A new locale is one file (#283, #346, thanks @haigou-web). The language switcher discovers locales/*.json from /api/locales instead of a hardcoded list, and CI runs scripts/i18n-sync.py --check, so a key added to en.json without its translations fails the PR instead of the user.
  • Simplified Chinese is complete and reads naturally (#281, #349, thanks @WaldInsel). Placeholder English strings translated, terminology made consistent; zh-CN is at 97%, and what remains are terms like CPU, GPU and VRAM that read the same in Chinese.

Fixed

  • The agent assumed an 8k context on llama.cpp (#364, thanks @jmdjr). llama-server reports its runtime context as meta.n_ctx, which the model list ignored, so the agent fell back to 8192 tokens even on a 64k server. It now reads it.
  • The Today / 7 days / 30 days cost cards counted only the GPU on the hub. They summed the hourly rollup's wsec, which is the GPU's energy alone, so a box with CPU/DRAM power readings (RAPL) was under-billed there while the range total below counted all three. They now add CPU and DRAM too, so the cards, the range total and the new energy chart agree. Remote hosts were already right.
  • Benchmarks said your ollama wasn't answering while its connectivity Test passed. The Test probes the ollama you registered under Settings → AI servers; the Benchmark Lab only ever talked to COPILOT_OLLAMA_URL, which defaults to 127.0.0.1:11434 — so an ollama on another box was green in one place and "not responding" in the other. The lab now benchmarks the registered ollama, unless you set COPILOT_OLLAMA_URL yourself (that still wins), and the error names the address it actually tried.

Pinned image: sikamikaniko123/homelab-monitor:0.39.0

Don't miss a new homelab-monitor release

NewReleases is sending notifications on new releases.