Documentation

Which plan you need

A pricing page can tell you what things cost. It cannot tell you what your server will spend. /recommend measures how busy the place actually is and answers the real question — which plan fits, what a cheaper one would cost you, and when you would run out. It is free, and it reads no message content.

Why a command and not a calculator

Every number on this site is real, and none of them answer the question you actually have. You don’t want to know what a 400-message summary costs — you want to know whether your server, with its handful of channels that never stop and its dozen that went quiet in March, will get through a month on Starter.

Nobody can answer that from a pricing page, because the answer is a property of your server rather than of the product. So the bot goes and looks.

Run /recommend in your server.

It reads how busy the place really is, compares that against what every plan includes and what the server has already spent, and tells you which one fits — including when the honest answer is the cheaper one, or the one you already have.

It is not a sales tool, and it is built so it can’t become one

The advice is free, costs no credits, and works when the server has run out — which is exactly when a metered version would go quiet. Nothing about the recommendation is a stored rule about which plan to push; it is written fresh from your server’s own numbers each time.

What it measures

Only channels you can already read — the same boundary as every other command, so a recommendation can never be assembled out of rooms you can’t open.

Message rateper channel: How busy each channel is, from the timestamps of a bounded sample. A channel quiet enough to fit inside that sample is measured exactly; a busier one is measured as a rate. The analysis knows which of the two it got and says so rather than blurring them.
The shape of a week: Which days carry the traffic, so a server that only moves at weekends isn’t sized as if it were busy on Tuesday. Only exactly-measured channels contribute — a truncated sample under-counts its own older days, and an under-counted past looks identical to growth.
Directionbusiest few: The busiest channels are also sampled from about a month ago, turning “how busy is this server” into “which way is it moving”. That trend is what decides whether an allowance that fits today still fits in six weeks.
Participants: How many distinct people are talking in each channel, as a count. Ten people trading short messages and one person posting logs are the same volume and very different servers.
What you have actually spent: The server’s real credit history through the bot. If the balance hit zero, that number is treated as a floor on demand rather than a measure of it — a server that ran out was stopped, not satisfied, and the analysis is told the difference.
Cost, in credits: Message volume is converted using the same charging function that bills you, so every figure in the answer is a credit you would really be charged.

A young channel is rated against its own age rather than the whole window, so a channel created on Thursday doesn’t read as though it were silent all week.

What it never reads

No message content is read, stored, or sent anywhere. This is not a policy applied afterwards; there is nothing to apply it to, because the analysis never needs the words.

Timestamps become rates. Discord message IDs carry the time they were created, so how fast a channel moves is arithmetic on IDs.
Character counts become credits, and are discarded. What a message said never enters the calculation — only how much of it there was.
Authors become a number. The analysis knows nine people are talking, not which nine.

The only text that leaves the bot is channel and server names. Everything else that crosses that boundary is a count, a rate, or a date.

Which is also why it’s free

Credits meter the message content a model reads, and this sends none. There is nothing to charge for — see what a credit is.

Reading the answer

The plan name is the least useful part of the answer. What you are being handed is the set of trade-offs behind it, in your server’s own numbers.

What a cheaper plan would actually cost you. Not “Starter is insufficient” — how far into the month it gets you, and what you would have to give up to make it last.
The levers, priced. Dropping a busy channel to a lighter quality, moving a daily digest to weekly, raising a live channel’s refresh threshold, or reading a shorter window each time. Each one is worth a specific number of credits in your server.
When you run out. Based on the direction your server is moving, not on today’s snapshot held flat.

Sometimes the answer is that you are on the right plan already, or paying for one larger than you need. Both are real outcomes and both get said plainly.

A recommendation is a reading, not a verdict

It is measured from a couple of weeks of history. A server that is about to run an event, open to the public, or go quiet for a term break knows something the numbers don’t — and you should weigh that over anything here. The burn-down on the credits page lets you play the same arithmetic out by hand.

Cost and limits

Free. It costs no credits and works on the free trial and on an empty balance — the moment the question actually gets asked.

Twice a day per server, resetting at midnight UTC. Reading a whole server is real work, and a server’s shape does not change between lunch and dinner, so a third run would cost something and tell you the same thing. Past the limit you get SB-REC-429 and the previous answer still stands.

A server with almost nothing in it gets SB-REC-204 instead of a guess. Until there is enough conversation to measure, the trial’s 3,000 credits are the right size, and that is what it will tell you.

Who can run it: Anyone in the server. It measures only what the person running it can already read, so a member can check the case before asking whoever holds the card.
Where else it appears: In the setup DM when the bot is first added, in /admin overview, in /help, and on the message you get when the server runs out of credits.
What it never does: Change anything. It reads, it advises, and it buys nothing — upgrading stays a deliberate act through Discord.

If you would rather work it out yourself, everything it uses is public: the credit formula, the full plan comparison, and the month-at-this-rate simulation. The command exists because it has your numbers and you would otherwise have to guess at them, not because the arithmetic is secret.