How it works

Separate the decision from the action.

Begin with a small workflow, explicit limits and evidence you can inspect. Add execution only through a connector that actually enforces those limits.

1. Define the intention

Name the actor, action, recipient, amount and required evidence. Identify which inputs come from trusted systems and which are supplied by the agent.

2. Evaluate the policy

The local spend-policy evaluator checks supplied inputs. Loop stores intentions and requests manual governance evaluations. Neither tool moves money.

3. Enforce at the provider boundary

This is integration work. A trusted gateway must own credentials, validate authority, enforce approvals, reserve limits and prevent replay before submitting the action.

4. Reconcile the outcome

Record the provider’s actual response. Pending or unknown outcomes stay unresolved until reconciliation, even when the earlier policy decision was allow.

Available tools; explicit next steps

Local policy evaluation and manual Loop evaluation are available as packages. Payment and refund execution connectors are design-partner work. A stored schedule does not run a job, and a decision record does not certify compliance.

Start with the developer tools