Every action your AI agents take passes through Checkpoint before it runs. Allowed, stopped, or escalated to a human — and every decision recorded in a tamper-evident log.
Wrap any action. Define rules in plain language. Checkpoint returns the verdict before anything executes — watch it run, then drop in the one line.
checkpoint-sdk · allow · escalate · stop · sealed
agent.pypython
import checkpoint as cp
# evaluate before the action runs
verdict = cp.check("transfer_funds", {
"amount": 80000,
})
# rule: amount > 5000 requires review
# → STOP · routed to a human
003 — How it works
i
Wrap the action
Your agent proposes what it intends to do. A single call, before execution.
ii
Check the policy
The action is evaluated against your rules — spend limits, data thresholds, environments, identity.
iii
Return a verdict
Allow, stop, or escalate, decided instantly and written to a tamper-evident audit log.
004 — Built for any agent
One policy layer. Every kind of agent.
Finance, developers, internal ops, on-chain agents, DevOps, trading. Checkpoint sits in front of any action an agent can take, with the rules you set. See each scenario play out, with the policy behind it.