pro
$400 / month
Max use, more machines, and the readings no single agent can take about itself — including the ones that exist only between two of them.
what it looks like
collisions() [{'agents': ['a','b'], 'overlap': 0.8}]
a grounded Φ 0.04 ⟨cache|benchmark⟩ advancing d3
b grounded Φ 0.06 ⟨cache|benchmark⟩ advancing d4Two agents, both perfectly on their own ground, both advancing, both correct at every step — and doing the same job. The duplication exists only as a relation between them, so it is invisible from inside either one however capable it is.
No threshold on Φ will ever surface that. It is not a tenth verdict; the nine describe one agent against one ground and keep meaning exactly that. A supervisor holds strictly more information than any member, and this is what it buys.
what it allows
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 adds
- Every agent on one screen, rather than one terminal each.
- Collisions: pairs whose grounds overlap, the reading no single agent can take.
- Statistics and dashboards — what your agents actually did, not one verdict at a time.
- Returning a drifting agent, not only observing it: the return is styled and injected.
- A year of readable history, so a quarter can be compared with the one before it.
what it does not do
It does not manage your agents for you. supercode advises and writes findings; it never sets a running agent’s ground, because a supervisor that could choose the reference it measures against would be measuring itself — the exact failure this whole instrument exists to make visible.