Usage
What the agents spent, by model — on this machine, and across the whole fleet.
The system tile is what the machine is doing now. Usage is what
already happened: where the tokens went.
Nothing here is a dollar amount. We do not resell tokens and we never see the
bill at Anthropic, OpenAI, Moonshot or xAI. What the screen shows is the count
each agent wrote down in its own log — input, output, cache — by model.
On the machine
Open the usage tile. Two objects and one control:
- a stacked plot across the top, one colour per model, brightest is biggest;
- a legend underneath that is also the table;
- a group-by that turns the same numbers into a reading by model (the
default), or by agent.
A day you tap becomes the column you are looking at. A gap is a dash, never a
zero: an agent whose log carries no counter is named below the rule, with a
sentence, rather than drawn as a quiet week.
There is no poll — one button rereads it.
Across the fleet
beafk cloud has the same screen, for every machine at once. It is Usage in
the cabinet, next to the fleet — not admin-only.
That one adds a third group-by: machine. The numbers still come from the
machines themselves. Your browser opens each one it can reach and asks. A
machine that is asleep, or on a tailnet this phone is not on, lands as a named
row with the reason — never as a total that quietly left it out.
What it is not
It is not your invoice from the model makers, and it is not beafk's bill. beafk
charges per machine. What a turn cost in tokens is on
this screen; what it cost in money is on theirs.
For processor, memory and disk — the things that make a box feel slow — see
is the machine all right?.