Waiting for your first agent

Leave this running and use your agents as normal. They appear here the moment one of them does anything.

Your agents

Every decision, receipted 0 allowed0 blocked0 asked

receipts.csv

This is the real export, built from the decisions above. Running it yourself, this button saves the file.

Nothing decided yet. Every allow, block and question appears here as it happens, each hash-chained to the one before it.

Owner console

Checks

Enforces the money and speed limits on the left. Turn it off and agents run uncapped, though everything is still recorded.

Stops an agent repeating the same action over and over, including the alternating kind that a simple streak check never catches.

Judges what an agent may DO, not just what it may spend. Piping the internet into a shell is refused outright. Deleting a whole tree, rewriting git history, reading or writing credentials, and publishing or deploying all stop and ask you first. Ordinary work passes untouched, and these fire even on a full budget, because a dangerous command is usually a cheap one.

Reads the task the person typed and says when the model looks wrong for it, in either direction. Where the governor sits in front of a coding agent as a hook it can only advise, because a hook cannot change a model. On the API route it can switch it for you.

when it nears the limitAt three quarters of the spend limit, either pause and ask you, or stop the agent outright without asking.

on a model mismatchTell me leaves the choice to you. Switch it rewrites the request to the cheaper model, and only ever downgrades. That works on the API route, where the governor owns the request.

everything below is priced as auto-detected

Money

spend, per agent The most one agent may spend in a session. It asks you at three quarters of this and stops at all of it. Twenty dollars is a normal working day. $20

—

of agent work

$15

it asks you at

$20

it stops at

4.00M

tokens, if you care

everything together, a day A ceiling across every agent at once. A per-agent limit cannot bound a team: seven agents on a twenty dollar limit can spend a hundred and forty. off

Spent so far: —

Speed and shape

speed, per agent Dollars a minute, for one agent. Ordinary work runs about ten to twenty five cents a minute, so this only catches a genuine runaway. A daily cap notices once the day is gone; this notices in the first minute. $2/min

speed, everything together The same measure across the whole fleet. This is the one that catches a session fanning out into dozens of subagents, which is how people lose four figures in an afternoon. $10/min

Burning right now: nothing

new agents a minute A team that starts over an hour is a choice. Eight appearing inside a minute is an orchestrator spawning orchestrators, and that is exponential. 8

errors before it stops retrying A rate-limited call fails cheaply. The retry after it does not. This stops a script that is paying to be told no. 6

Want it gone?

This removes Enforcer from your agent's settings and stops every check straight away. Nothing you have already recorded is deleted, and your agents carry on working normally. You can put it back any time with one command.