Changelog

What changed in each version. This page is the changelog; the notes on each GitHub release say the same in both languages, and the application links here from the version in its header.

0.2.2 — 15 September 2026

Fixed

• The speed test works again. The provider began refusing downloads of 100 MB and more, and the largest step of the measurement asked for 150 — so on a fast line every run ended in "the measurement could not be made". The step now stops below that limit, and a step that is refused no longer throws away what the smaller ones already measured. A run costs at most 192 MB instead of 252 MB.
• The speed-test panel follows a change of language. Its heading and button always did; the data notice, the history and the line reporting the last run stayed in the language they were first written in until the page was reloaded.

Added

• Two more figures under "More figures": the fastest measured and the peak observed rate, each way. The measured one comes from the speed tests and is the only figure on the page that is about the line; the peak one is the busiest minute this device moved, counted and labelled like the transferred volume above it — this device, local network included.
• The complaint report carries the measured speeds: the slowest and the fastest download and upload of the reported period, and how many measurements they rest on. The slowest is there because the fastest alone argues the provider's case. The method paragraph names who the measurements were taken against and states that the application never measures on its own.

Changed

• The event log marks where one day ends and the next begins. Over a period longer than a day its rows carried clock times only, so a row at 23:04 under one at 00:35 belonged to the day before without saying so.
• The two panels under the log are switched instead of folded open. Their headings sit in one row and stay there; whichever is open takes the whole width below them.

0.2.1 — 7 September 2026

Added

• A capacity marker. Enter the contracted download and upload rates in the settings (optional), and the figures flag minutes in which more data moved than a line of that rate can carry — the shape of a copy to a device on the local network, which the transferred volume otherwise makes look like internet traffic. Up and down are judged separately, with a margin of one and a half; without contracted rates the marker does not appear.

Changed

• Transferred volume no longer counts tunnels, and says what it counts. A tunnel moves the same bytes twice, so a VPN doubled the figure. The tiles are labelled "this device, incl. local network", and the dashboard lists which interfaces were counted and which were not.
• The menu bar icon on macOS is a template: the system tints it to match a light or dark menu bar and inverts it while the menu is open. The three states differ by the badge on the globe — a check, a cross, an exclamation mark. Windows keeps the colour icons.

Fixed

• The log table keeps its column widths. The widest cell on the visible page used to decide all four columns, so one "Monitoring interrupted" among a page of "Online" rows pushed the whole table, and paging pushed it back.

0.2.0 — 28 August 2026

Added

• A duration filter over the calendar. A dropdown above the heatmap picks which lengths of outage count at all, as a multiple choice over the same groups the distribution below already prints. It exists because a five-second probe interval produces a great many five-second outages, and a calendar coloured by their number says more about the interval than about the line. A second setting decides how far the filter reaches: the calendar alone, or the figures, the log and the exports with it.
• The filter removes outages from the counts, never days from the recording — a day whose only outages were blips stays marked as watched. A report built under an active filter says so and names the lengths it counted, because an outage count that quietly leaves some out is a number the reader cannot check.
• Patterns behind the outage count: worst time of day and worst day of the week, the longest run without an outage, and how the outages of a period were spread by length. The two “worst” figures are a share of the time actually watched in that slot, not a count.
• Transferred volume, per minute, read from the network interfaces.

0.1.1 — 18 August 2026

Changed

• The period is picked from the heatmap: click its year, a month name, a calendar week — the week numbers are new — or a single day. The button row is down to three: Today, 30 days, 90 days.
• A running block is taken whole: picking August on the 17th means all of August, not the part that has already happened.

Removed

• The “more periods” dropdown, and with it the separate Last week, Last month and Last year entries. Each of them is now simply the block before the current one on the heatmap.

Fixed

• Upgrading over a running copy works. Windows cannot replace an executable that is in use, and this installer asks for no administrator rights, so winget upgrade failed every time — a connection monitor is running whenever you would want to upgrade it. The installer now closes the running copy first and starts it again afterwards.
• Ending the tray any other way than through its own Quit item is recorded. On Windows that was every other way there is, and such a shutdown was read as a crash on the next start.
• The complaint report ends with the current day when the chosen period has not finished yet, and says so in its method paragraph. Before, a report over the running month listed a row for every day still to come and measured the monitored time against the whole month.

Added

• A “Buy me a coffee” button at the foot of the dashboard and on this site. It is a link and nothing else: the picture is served from here rather than from their servers, so nothing is requested from anywhere else until you click it.

0.1.0 — 17 August 2026

Added

• Continuous TCP monitoring against a configurable list of target servers, with a failure threshold: an outage is only counted once none of the targets answers for longer than the configured number of seconds.
• Tray icon showing the current connection state, with the dashboard one click away.
• Local dashboard, served in the browser, with optional token-protected access from the local network, and its own version shown in the header.
• A theme setting — auto, light or dark — for the dashboard, applied server-side so the page never paints in the wrong one and flips.
• Connection quality tracking — latency and the share of probes that got no answer — drawn as a curve.
• Complaint-report PDF for the internet provider: what was measured, how it was measured, every outage with timestamps, a day overview, a time-of-day profile, and a covering letter.
• Nine selectable periods — today, this week, this month, this year, last week, last month, last year, the last 30 days, the last 90 days — driving the figures, the year heatmap, the event log, the CSV export and the report alike.
• Event log with paging.
• CSV export of the log for the selected period.
• Start at login, on both Windows and macOS.
• Sleep-aware accounting: time the machine spent asleep, or with the program not running, is reported as unmonitored rather than as an outage. A sleeping machine is one standby however often the operating system wakes it in the background: macOS wakes a closed laptop every quarter of an hour for well under a minute, and a wake that short is folded into the standby around it rather than logged as monitoring that resumed and stopped again.
• Each heatmap cell says in its tooltip how long that day was actually watched, beside the number of outages. A day counts as monitored on any measurement at all, so a night spent asleep otherwise looks exactly like a day watched around the clock.