Support
Most hosting support in 2026 is a chat window that tries very hard not to fetch anybody. Ours fetches somebody. There is a telephone number, it is answered at four in the morning, it is on every plan including the ones that cost nothing, and it is printed here rather than kept behind a login.
Technical support, every hour of every day. Sales, billing and abuse have their own numbers on the contact page.
Why this page is not a list of adjectives
Everyone writes “24/7 expert support”. Almost nobody writes down how long they have to reach a failed part, or what you are owed when they do not. Those two things are the whole of a support promise, and both are further down this page — taken out of the service level agreement rather than out of a copywriter.
The rest of it is simpler than the industry likes to make it sound. Somebody picks up. They can see the machine. They tell you what is wrong with it, including when what is wrong with it is not ours.
What it looks like
This is the shape of a good first exchange. The AI on shift looks your machine up before it answers; if a person is on shift it hands them the answer privately and says nothing, and you get the human. Not a bot standing guard until it gives up and offers you a form.
My stream stopped about twenty minutes ago and the panel will not load. Account 248261.
Looking now. The machine is powered on and healthy, and its address answers from our edge — so this is not the server being off.
Two of your stream ports are closed and the third is running with no source connected. Is your encoder still pointed at it?
…the encoder is showing an error. Let me restart it.
Composed from a real exchange in the archive. Details changed.
Before you call
Six things are yours to do, from the panel, with no ticket, no charge and no limit. Between them they cover most of what people write in about.
Through the machine’s own management controller, so it works even when the operating system has stopped answering.
Thirty-three operating system images in one click, thirty-two of them free, as many times as you like.
Screen and keyboard on the machine itself, which is what makes a firewall rule you wrote yourself survivable.
On every address, yours to edit — which matters more than it sounds if the machine sends mail.
Delegated access hands another registered customer entry to your dashboard, and either side can withdraw it.
The looking glass runs ping, traceroute, MTR and BGP queries from our own edge in New York and Bucharest.
In writing
Support is included with every server and every plan, the free ones as well, so there is nothing to weigh up and nothing to choose. The part of the agreement worth actually reading is the other one: a replacement window for a failed part that runs from the moment you tell us — 24 hours on the free tier, down to 2 on Gold and Platinum — and a written schedule of what you are owed for every hour past it.
The clock starts when you tell us, by telephone or live chat. The full schedule, the separate rule for network unavailability, and how to claim are below.
Who is on shift
Everybody has a chatbot now, and most of them are a wall. Ours is arranged the other way round, and the arrangement is worth a paragraph because it is the part you cannot see from outside.
It is not a search box with a personality. Before it replies it can pull up your account, read the server’s own hardware health, check whether the address answers from our edge and look at the switch port. Most first replies are a fact about your machine rather than a suggestion to try turning it off and on.
It writes its answer as a private note to whoever is on, and waits. If they reply, you never see the draft — you get the human, and they got a head start. It only sends its own answer if nobody does. So it is not a gate standing in front of a person; it is a floor underneath one.
Three in the morning on a Sunday is the hour this matters. The alternative is not a better answer, it is silence until someone wakes up. Anything it cannot finish it hands on as a ticket with a team’s name attached, and says so instead of guessing.
Read you a password, ask you for one, or take an irreversible action on your server because a chat window asked it to. Reinstalls, cancellations and anything that destroys data are yours to trigger from the panel or a person’s to carry out.
You can always just ask for a person. Saying so plainly is enough — there is no phrase to guess and no menu to defeat — and the telephone has never had an AI on it at all.
What a faster answer costs
The telephone, the live chat, the ticket desk and the email address are on every plan and are never metered — the free one included. What a paid tier buys is time: how fast we answer, how fast a failed part is replaced, and whether the machine is monitored when it goes quiet at three in the morning.
| Plan | We answer within | Failed part replaced | Advanced support | Monitored | Per month |
|---|---|---|---|---|---|
| Basic24x7x12 | 12h | 24h | — | — | Included |
| Bronze24x7x4 | 4h | 4h | 30m | 5m | $19 |
| Silver24x7x2 | 2h | 3h | 60m | 2m | $39 |
| Gold24x7x1 | 1h | 2h | 90m | 1m | $69 |
| Platinum24x7x1/2 | 30m | 2h | 120m | 30s | $99 |
The monitoring column is the one that changes an ordinary month. We read the switch port your server is plugged into rather than pinging the machine, so a firewall cannot hide an outage from us and nothing has to be installed on your side; when the port goes quiet you get an email. On the top two tiers we open the ticket ourselves rather than waiting to be asked.
Every figure in this table is read from the same catalogue the checkout bills from, so it cannot drift from what you are actually charged. If the table is missing, that link is down — the free plan still carries a 12-hour response and a 24-hour hardware replacement, in writing.
The full schedule
The control above shows one band at a time. Here is the whole ladder, out of the service level agreement rather than out of a brochure, with the separate rule that covers the network and the seven days you have to claim under either. The replacement window is your tier’s — 24 hours on Basic, 4 on Bronze, 3 on Silver, 2 on Gold and Platinum — and every band below is measured from the moment we are past it.
Percentages of the monthly recurring charge for the affected service, by how far past your window the replacement ran. The clock starts when you tell us, by telephone or live chat — which is the reason the number is printed at the top of this page rather than hidden behind a login — and the window is the one your tier carries, so a Gold customer is owed the first band two hours in where a Basic customer is owed it after twenty-four.
More than fifteen consecutive minutes of network unavailability earns one day of credit for every cumulative hour of it in that calendar month, and part hours count proportionally once the month has passed an hour in total. Two hours and a quarter is two and a quarter days. The cap is one month: we cannot owe you more than you paid.
Within seven business days of the end of the month, tell us three things: who you are, the date with the start and end time of the outage or the failure, and a sentence on what it looked like. You get an answer by email either way. Approved credit lands on the next invoice.
Two things void it, and both are in the agreement: not giving us the access needed to troubleshoot properly, and an account that is not in good financial standing. Credit is credit — it is set against what you owe and is never paid out as cash.
Before you write
Most of the delay in a support thread is not us thinking. It is the first four exchanges, where we ask which account, which address, and when. Open with these and the person reading has everything they need to start looking before they have finished your message.
The account ID or the IP address. Either one is enough; the account ID is on every invoice and at the top of the dashboard, and the IP address is in the setup email we sent when the server was built.
The hour it started, with your timezone. “Since about 14:20 UTC” is worth more than “since this morning”, because everything on our side is timestamped and we can go straight to that minute. If something changed just before — a deploy, a firewall rule, a reboot, a payment — say so even if you are sure it is unrelated.
The actual text, not a description of it. A screenshot works. “It says connection refused” and the real line from the log are different amounts of information, and the second one often ends the conversation on the first reply.
Run a traceroute from where you are, or use our looking glass to see the path from our edge back toward you. A great many “the server is down” reports are a route, a home firewall or an office network, and finding that out takes a minute rather than an afternoon.
Where the desk stops
Every host has a line. Most of them find out where it is halfway through your problem. Ours is here, so you can plan around it instead of discovering it.
We will always look, and we will tell you what we can see. What we will not do is spend a night inside software we did not write while calling it hardware support. If you want somebody inside your own stack rather than beside it, that is a managed service rather than a support desk, and server management is the page for it.
Written down already
This is the part of the desk that never sleeps and never has a queue. Encoder by encoder, with the settings that actually matter and the ones that do not — and, for the people running their own machines, the server-side ones too.
Getting a radio stream up, and what to check first when it stops.
BUTT, Mixxx, SAM, VirtualDJ, the SHOUTcast DSP and the rest, configured.
Barix, Teradek and the other boxes that sit in a rack and just broadcast.
RTMP in, players out, and where a video stream usually breaks.
OBS, Wirecast, XSplit and FFmpeg, with the settings that matter.
Epiphan, AVerMedia, Hauppauge and the rest of the rack.
Uploading, transcoding and serving a library rather than a live feed.
File transfer, mailboxes and the everyday hosting questions.
Putting a public address on a virtual machine, three ways, with IPv6 and reverse DNS.
In your own words
The desk answers in the language a message arrives in. This site is published in eleven — English, Romanian, Spanish, Portuguese, Italian, French, German, Polish, Greek, Turkish and Chinese — and the chat has handled all of them and several it is not published in. Nobody should have to debug a stream in a second language at three in the morning.
The one thing to keep in English if you can: the pasted error. A translated log line is a log line we cannot search for.
Other doors
Not support channels in the sense of a queue, but real accounts with real people behind them — and the two public records where a complaint has to be answered in the open.
If something has gone properly wrong and the desk has not fixed it, complaints@primcast.com reaches a different set of eyes from support@.
Asked often
