ground

$0 / forever

Free forever, and it is the whole instrument. A paid plan does not detect drift better — there is one detector and this is it.

what it looks like

grounded     Φ 0.00   ⟨benchmark|pars|ship⟩ advancing d7
advancing    Φ 0.06   ⟨benchmark|pars|ship⟩ advancing d5
goal-drift   Φ 0.49   ⟨pars|test|write⟩ advancing d4
goal-drift   Φ 0.59   ⟨refactor|tokenis⟩ advancing d4
goal-drift   Φ 0.56   ⟨help|logg|tidy⟩ advancing d5

An agent asked to ship a parser and a benchmark. By the third step it is writing tests, then refactoring the tokeniser, then tidying logging — each a reasonable thing to be doing, none of them the job.

It reports advancing every single step. It believes it is fine, and on its own terms it is. The ground was set on the first check and cannot be revised, which is why the fourth line reads 0.59 instead of agreeing with the agent.

what it allows

field reads1,000 / month
field writes10 / month
field history24 hours
drift retention1 dayenough to debug a run you are watching
the check itselfunlimitedit runs locally — nothing meters it

Going over returns 429 with the reset time — a hard stop, not an overage. No invoice you did not expect. Ask for more and it is a config change.

what it does not do

It does not remember. A day of drift history is enough to debug a run you are watching and not enough to see a pattern across a week. It will not page you when an agent spirals at 3am, because nothing here is watching but you.

pip install laserbrain