Features
Smart helpers
Suggestions and warnings the panel works out by itself from its own data: usage watch, month-end forecast and budget, power schedule suggestion, payment risk, support help, quietest hours for maintenance, a periodic digest, right-sizing and the health score.
On this page
What they have in common
Everything on this page is rules and statistics. It is computed inside the panel from the data in the panel's own database. No outside AI service, and no other outside service, is involved, and no data leaves your installation.
- It says why you are seeing it: every suggestion and finding comes with the numbers behind it.
- It can be hidden: hide a suggestion you do not want; a new finding shows up again.
- It does nothing by itself: any suggestion that changes a server runs only when someone with the permission clicks it.
Where the computing happens and why the thresholds are what they are is told in How the smart helpers work without an outside AI service.
| Helper | Who sees it | Where |
|---|---|---|
| Usage watch | Customer and staff | Server page, insights; Administration → Abuse → "Unusual use" |
| Month-end forecast and budget | Customer | The "This month" card on the overview, insights |
| Power schedule suggestion | Customer | Server page, insights |
| Payment risk | Customer and staff | Insights; Administration → Billing → "At risk" |
| Support help | Customer and staff | The new-ticket window; "Similar resolved tickets" on a ticket's page |
| Quietest hours | Staff | Administration → System → Automatic updates; a connection's Hosts tab |
| Periodic digest | Staff; customers if an admin enables it | My account → Notifications |
| Right-sizing | Customer and admin | Insights |
| Health score | Customer | Server page |
Usage watch
Once an hour the panel compares each server's CPU and network use with its own recent history. Servers are never compared with each other.
- CPU: the last 3 full hours are compared with the same hours of the day (one hour either side) over the 14 days before. A finding opens when use is at least 20 points above the usual level and that is also a sharp departure by the server's own normal variation; it closes when the difference falls below 10 points. A value between the two thresholds keeps an open finding open but does not open a new one. A spike of a single hour is not a finding.
- A weekly rhythm is not a finding: a departure must also hold against the same hours of the same weekday one and two weeks earlier. A server that works on weekdays and idles at the weekend is not "lower than usual" at the weekend, and a job that runs every Sunday night is not "higher than usual". This check only removes findings: a weekly job that fails to run is not noticed.
- A gap in the measurements: if there are no samples for the last hours (the platform did not answer, the server is off), an open finding does not close at once; it is kept for 6 hours, and if the measurements return the same finding goes on and the owner is not told a second time.
- Lower than usual: looked at only on a server whose usual level is at least 15%. A finding opens when the last hours fall to 20% of the usual level or less, and closes when they are back at 50%; a service that was running may have stopped.
- Network: traffic is kept per day. The larger of today so far and yesterday's total is compared with the median of the 14 days before that. A finding opens when traffic is at least 5 GiB and at least 3 times a usual day, and also clearly outside the server's normal variation.
- Minimum history: 12 hours from 4 different earlier days for CPU, 7 days for network. With less, the result is "not enough data".
- A lasting change: if use stays at its new level for days, the panel takes that as the new normal and the finding closes by itself in about a week.
The customer sees a finding in the "Usage watch" section of the server page and in the insights on the overview, with its numbers and since when; it can be hidden for 7 days. One notification goes out when a finding opens. By default that is inside the panel only; e-mail and the other channels are switched on under My account → Notifications.
Staff see the same findings on the "Unusual use" tab under Administration → Abuse. CPU at 90% or more gets the note "resembles mining"; traffic that is at least 90% outbound gets "resembles an outbound flood". The list is statistics: it does not create an abuse finding, and that module's own checks and approval flow go on as before.
Month-end forecast and budget
The "This month" card on the overview shows, per currency, the amount so far, the month-end forecast and the monthly total. Prices are read from where billing keeps them; nothing is priced a second time here.
- Servers and virtual datacenters billed monthly: the monthly price, in proportion to the time they have existed this month.
- Servers billed by the hour: the usage billing has recorded this month (the hours of servers deleted since are included) plus the hourly rate of the servers still accruing, times the hours left in the month. A server held because the balance ran out is not carried forward.
- A budget is measured against two things: the sum of the servers' monthly prices, and this month's forecast. The second also sees hourly usage and servers that were deleted during the month but had already cost money.
- Thresholds: at 80% of the budget the forecast is "close to" it, above 100% it is "heading over". The card gets a second line and an insight appears, with the numbers.
- Notification: when the forecast passes the budget, one notification a month goes out (by default in the panel and by e-mail). If "budget exceeded" was already sent that month, this one is not sent on top of it.
No comparison is made for an account without a budget. For an account without hourly servers the figures are the same as before. A budget stops nothing; it only tells you.
Power schedule suggestion
When a server's CPU history shows that it has been idle in the same hours of the week, every week, for weeks (nights, weekends), the panel suggests the schedule that keeps it off through those hours.
- Data: the hourly CPU averages of the last 4 weeks, laid out over the 168 hours of the week in your time zone.
- An idle hour: seen in at least 3 weeks, and in each of them with an hourly average below 3% and a highest sample below 10%. An hour with no measurement (the server was off) counts as in use.
- Night window: the longest stretch of hours that is not used on any day. The server is switched on 1 hour before the hours it is seen in use and off 1 hour after them; what remains must be at least 6 hours. A server that is not used at weekends goes off on Friday evening and on again on Monday morning.
- Weekly window: if there is no night window, the longest idle stretch of the week (a weekend, say), at least 24 hours.
- No suggestion for a server that would be off for less than 20 hours a week, for one that is idle all the time (that is a right-sizing matter), or for one that already has an active power-on or shutdown schedule.
- Minimum history: 21 days. With less the page says "not enough data" and how many days there are.
A suggestion is only a suggestion. When "Create" is pressed, the suggestion is computed again on the server side from the history at that moment and created the way any schedule is: the same power permission, the limit of 10 schedules per server, and an audit record. "Not now" hides the suggestion for 7 days.
Payment risk
None of these rules charges anyone, suspends a service or writes to anyone. Reminders, suspension and deletion stay in billing's own flow.
The customer is warned in two cases, only where automatic collection is on and the next renewal is at most 30 days ahead:
- The card billing will charge (the default card; the newest one if there is no default) expires before the renewal date and the balance does not cover the amount. Billing tries no other card, so a spare card that lasts long enough does not remove the warning; making that card the default does. The card (last four digits, month / year), the date and the amount are given.
- No card lasts long enough; there is a balance, but it is smaller than the renewal. The balance, the shortfall, the date and the amount are given.
An account that pays its invoices by hand is told nothing. The warnings can be hidden; they come back in the next period.
Staff with the billing.view permission see these accounts on the "At risk" tab under Administration → Billing: those with an unpaid invoice past its due date, and those with at least 2 failed charges in the last 30 days. A payment page the customer left or let expire does not count as a failed charge. The list is ordered by amount at risk, largest first, and has 200 rows at most. "Growing" means part of the overdue amount was added in the last 7 days. It is an order in which to look, not a total.
Support help
Matching is done inside the panel, by shared words. The text goes to no outside service and is not stored.
- The rule: the text is split into words; case and Turkish letters make no difference. Words shorter than 3 letters and common ones such as "hello", "please" or "the" are dropped. A word that appears in many tickets counts for little, a rare one for more. At least 2 shared words and a minimum similarity are required; at most 3 results are shown, each with its shared words as the "reason shown".
- Staff see "Similar resolved tickets" on a ticket's page: the newest 300 tickets that were closed in the last 365 days and answered by staff are compared. Each result shows the last public staff reply that closed that ticket, and it can be copied; internal notes are never shown. Nothing is sent by itself.
- A customer writing a new ticket is shown, before sending, what may already answer it. Only sources the customer can already see are read: their own earlier tickets that staff answered (the newest 50) and, if the status page is on, the incidents on it.
Quietest hours for maintenance
For work that disturbs servers (the panel updating itself, a host going into maintenance) the panel works out from the usage history the hours in which the servers are least busy, and shows them in the two places where staff plan that work. It only suggests: it moves, schedules and changes nothing by itself.
- The unit of load is busy vCPUs: each server's hourly CPU average times its vCPUs, added up over the servers measured in that hour.
- The rule: the hours of the last 4 weeks are laid out over the 168 hours of the week. An hour of the week is compared only if it was seen in at least 3 weeks. The window looked for is 2 hours long.
- Two results: the quietest window that starts at the same hour on all seven days (daily), and the quietest 2 hours of the whole week (weekly; Sunday 04:00–06:00, say).
- The numbers shown: the window's average load, the average of the whole week, the load of the busiest hour, how many weeks and days of history it rests on, and how many servers were measured per hour.
- Minimum history: 21 days. With less it says "not enough data" and how many days there are. If no connection reports CPU figures it says "not measured".
Administration → System → Automatic updates shows these hours for all servers of the installation, in the panel server's own time zone. The "Use as the update window" button saves the daily window through the ordinary request of the update setting: the same permission, one maintenance task at a time, and an audit record.
A connection's Hosts tab shows the same for the whole connection and for every host that runs panel servers, to pick the hour before a host is emptied. The button that empties a host is as it was and runs at once; the panel does not schedule the migration for that hour.
Periodic digest
Instead of a notice per event, one message says where things stand, on the day and at the hour you choose. It is off unless turned on. Its content is gathered from what the panel already works out (insights, payment risk, the month-end forecast); there is no new rule and no outside service.
- Staff with the
dashboard.viewpermission turn their own digest on under My account → Notifications. It holds: open findings by severity, findings opened and closed since the last digest, accounts at payment risk, operations that failed since the last digest, and the capacity findings. Every line is something that person can already see in the panel. - Customers get one only after an administrator turns the customer digest on under Settings → Notification rules (off by default) and the customer turns their own on. It holds: findings on their own servers, this month's amount so far and the month-end forecast, and a renewal within 14 days. No other customer's data is read.
- When: weekly (Monday 08:00 by default) or daily; the day and the hour are in the person's own time zone.
- Delivery: it is an ordinary notification. The person chooses the channel (in the panel, e-mail, chat apps; e-mail by default), the administrator's notification rule applies, and during quiet hours it does not go to chat apps.
- An empty digest is not sent: with nothing new to say, no message goes out for that period.
- Preview: "Send me a preview now" sends the digest as it stands at once and shows its words on the screen; 5 times an hour.
Right-sizing and the health score
These belong to the first set and have not changed.
- Right-sizing: the CPU samples of the last 7 days are used; a suggestion needs at least 3 days and 24 samples. A server is idle when the average is below 2% and 95% of the samples are below 5%; oversized (if it has more than 1 vCPU) when the average is below 10% and 95% of the samples are below 30%; undersized when the top 5% of samples are above 85%. The suggested vCPU count is the one that would bring use at busy moments to about 60%. Servers inside a virtual datacenter are left out.
- Health score: it starts at 100; an error state, open security findings, failed operations in the last 24 hours, a saturated or often-high CPU, and being switched off or idle take points away. 85 and above is healthy, 60 and above is worth a look, below that is a problem. The reasons are listed on the server page.
Limits
- Platforms without CPU measurement: XCP-ng, oVirt, KVM (libvirt) and Incus do not report per-server CPU. For those servers the usage watch reaches no conclusion about CPU, no schedule is suggested, right-sizing does not work, and they are left out of the quietest hours; the server page says so.
- Network: a server with no traffic records is not judged on its network. Traffic is counted on VMware and Proxmox only.
- A new server: until enough history has built up, the result is "not enough data".
- The rules have unit tests. They were also run end to end on a seeded history (200 servers of 20 customers, five patterns of use, 14 days of samples): with ordinary use the usage watch opened no finding, and it found the three real departures that were added. They were not tried with real customer workloads. Delivery of the periodic digest by e-mail and chat apps was not tried with a real mail server or app.
Source: docs/smart-features.md and docs/operations-modules.md (both in Turkish).
Install command
curl -fsSL https://lightningvm.app/install.sh | sudo shUbuntu 22.04 / 24.04 / 26.04. Requirements and options
