When velocity spikes or keys go wrong, Edge Shield steps in. It watches, throttles, quarantines, or freezes — automatically. Your money stays safe even when an agent does not.
Every agent — and the network itself — sits behind an adaptive circuit breaker. Each level escalates what gets blocked, from watching to a full global freeze.
Levels 3 and 4 are the only ones that stop money. Refunds, escrow releases and settles always pass.
No human reacts this fast. The edge watches every debit — and when a spike lands, the shield moves before the second one is over.
The detector counts transactions per repo in a rolling 1-second window. The instant it crosses the threshold, the request is rejected before the spend lands.
The credential that produced the burst is soft-stopped. The rest of the agent's keys stay live.
All debits under that owner/repo are frozen. The event is written to the ledger with reason and timestamp — nothing is silent.
The repo is placed under Edge Shield Level 3 with an automatic cool-down. A webhook fires so the owner knows the moment it happens.
The shield escalates when the ledger is under pressure — and quietly steps back down when the window passes.
Wallet locks are counted across the network in a rolling 5-minute window. If they stack up, the problem is not one agent.
One bad agent gets quarantined on its own. Three at once means the network tightens for everyone — new spends slow, refunds keep flowing.
Every trip carries a timer. When it expires, the shield re-opens on its own and logs the release.
Owners can release early from the dashboard. Otherwise the shield simply heals.
At Level 3 and 4 the shield blocks every new debit. But money that already belongs to an agent can always come back.
Discretionary spends are refused the moment the gate is checked — before any balance moves.
The freeze only cuts new risk. Returning value is never held hostage by the shield.
Reasons are returned with every block — shield_level, reason, cools_at — so agents know exactly what tripped.
Trips, releases and cool-downs are stored as shield_events — scope, level, reason and timestamp. Nothing is silent.
It acts on its own in milliseconds — but every level can be inspected, released and overridden by the owner of the agent.
Current shield level, reason and trip count for any agent.
GET /api/shieldClear a quarantine before the cool-down expires.
/api/me/shield/releaseAdmin can raise or drop any shield, network-wide.
POST /api/admin/shieldShield state is stored per owner/repo. A frozen agent keeps its balances, keys and policy intact — and no other agent is touched. One bad agent never becomes a systemic event.
What the shield does alone
The shield runs at the edge on every request. It does not wait for someone to notice an incident.
What the owner always keeps
Automation contains the damage. The owner decides what happens next — with every action recorded.
Edge Shield
Spikes are intercepted. Quarantines heal. One bad agent never becomes a systemic event.
Join AgentPayNow
Bind a repo. Send a POST. Agents settle — humans stay out of the way.
Join AgentPay│
