A desktop tray app that unifies local GPU metrics and remote AI service quotas into a compact overview.
Find a file
Alexey Skobkin 54879c4f24
All checks were successful
ci/woodpecker/pr/ci Pipeline was successful
ci/woodpecker/push/ci Pipeline was successful
ci: run push pipeline only for the default branch
A feature-branch push with an open PR ran two pipelines on the same
commit concurrently — duplicated runner work for the same result, since
the pull_request pipeline already tests the merged state. Restrict the
push event to master; tags and PRs are unaffected.

Co-Authored-By: GLM-5.3 Flash <noreply@z.ai>
2026-11-04 02:45:47 +04:00
.woodpecker ci: run push pipeline only for the default branch 2026-11-04 02:45:47 +04:00
cmd/aimon feat(app): show-on-startup, start-on-login, single-instance guard 2026-10-31 03:46:14 +04:00
docs fix(ui): derive ExhaustAt from the projected value; review follow-ups 2026-11-04 00:35:53 +04:00
internal test(app): deflake notifier queue-drop test with a parking barrier 2026-11-04 02:05:26 +04:00
packaging fix(ui): resolve the Wayland taskbar icon via the GLFW app-id hint 2026-08-29 17:58:50 +03:00
.gitignore docs: initial commit with design documents 2026-08-23 06:00:15 +03:00
.golangci.yml ci: add Woodpecker pipeline, GoReleaser release and golangci-lint config 2026-08-30 02:19:57 +03:00
.goreleaser.yml feat(settings): About tab with version, build info, and config/log paths 2026-10-30 21:23:59 +04:00
AGENTS.md docs: refresh AGENTS.md authoritative list and constraints for S7 2026-10-25 01:43:35 +04:00
go.mod fix(app): send desktop notifications via D-Bus on Linux/BSD, Fyne elsewhere 2026-11-01 02:57:43 +04:00
go.sum chore(deps): bump Fyne to v2.8.2 2026-10-30 20:03:49 +04:00
LICENSE docs: replace the README stub with a product page 2026-08-30 23:15:58 +03:00
README.md feat(ui): show pace burn-out estimate in quota countdown labels 2026-11-03 21:10:55 +04:00

AI Monitor ci

AI Monitor is a desktop tray monitor for local AI workloads and LLM subscription plans. It lives in the system tray and keeps an eye on your AMD GPU, your local model router, and the quotas and spending of the cloud plans you use — without ever opening a browser tab.

Monitor window: Overview tab with local GPU, llama-swap and Hindsight, and remote Z.AI and Codex quotas Settings window: Add source dialog with the source type dropdown open

Features

  • Compact borderless monitor window launched from the tray, with Overview, Local and Remote tabs. Overview cards are clickable: tapping one jumps to that source's full detail card, whose header carries a per-source refresh button for retrying just that source.
  • Usage pace for subscription quotas: how fast a quota window is burning ("Pace 3.4%/h") and where the usage is heading by the window's reset ("≈68% at reset", with a red overflow segment when the projection passes 100%). When a quota would run out before the reset at the current burn, the countdown shows how long it lasts: "≈1h35m / 4h45m" — estimated quota time against time until reset. OpenRouter shows a spend speed in $/h. Estimates derive from a bounded in-memory history; the estimation window (10–60 minutes) is configurable per source.
  • Multiple instances per source: monitor several accounts or endpoints side by side, each with its own poll interval. The Add dialog explains what each kind monitors, existing instances can be duplicated with one click (credentials are not copied), and reusing a name shows a soft hint.
  • Credentials are stored in the system keychain — never in config files or logs. The source configuration itself can be exported to a JSON file and imported back on another machine (credentials are not included and are not affected by an import).
  • Session conveniences: an optional "Show monitor window at startup" preference, a "Start on login" toggle (writes a freedesktop autostart entry; removed again when switched off), and a single-instance guard that quietly exits a second launch.
  • Update checks against the project's releases (automatic, or "Check for updates…" in the tray menu) with a release-notes dialog and a direct link to the download.
  • An About tab in Settings with the version, commit and build date, and clickable shortcuts to the settings directory and the log file; the tray menu has a matching "About" entry, aimon --version prints the same identity in a terminal, and the log itself is written to <config>/fyne/com.skobkin.aimonitor/logs/aimon.log (1 MiB rotating).
  • Desktop notifications for quota windows — "low" (entering the red band, ≤10% remaining), "exhausted" (0% remaining), and "reset" (a used window rolled over) — edge-triggered so each transition notifies once per quota window. Each event has its own toggle in Settings (plus a master switch); notification text carries percentages only, never absolute amounts. On Linux the app talks to the desktop's notification service directly, so popups auto-hide after the platform's default timeout.
  • Optional masking of sensitive values such as spending, for screenshots and shared screens.
  • Sources are independent: one failing or disabled source never breaks the others, and the last good data stays visible while a source retries.

Monitored sources

Source Type Shows
AMD GPU via amdgputop-web Local GPU and VRAM utilization, temperature; power, GTT and clocks in detail
llama-swap Local Running models and service health
Hindsight Local Deployment health, memory banks and facts, activity rates
MiniMax Token Plan Remote 5-hour and weekly remaining quota, usage pace and end-of-window projection
Z.AI GLM Coding Plan Remote 5-hour token quota and monthly MCP usage quota, usage pace and projection
OpenRouter Remote Key usage and spend limit, account credits and balance, $/h spend pace
ChatGPT Codex Remote Plan, 5-hour rate-limit windows, remaining credits, usage pace and projection

Download

Grab a prebuilt binary from the releases page: aimon for Linux amd64 (tar.gz) and Windows amd64 (zip).

On KDE Plasma (Wayland), see packaging/README.md for the taskbar icon setup.