Control model
An autonomous agent may propose a request, but a request should not automatically imply unlimited authority. SENTRY inserts a policy layer between intent and execution so permissions become explicit rather than assumed.
System layers
Produces the target, asset, amount and requested action that enters the control path.
Applies explicit rules such as allowlists, ceilings and action constraints.
Receives only those requests that satisfy the active policy decision path.
Keeps the decision context and resulting outcome available for later review.
Policy primitives
The operating model is designed around simple controls that remain understandable for both builders and operators.
- Asset allowlists
- Target contract allowlists
- Maximum transaction value
- Daily operating budget
- Allowed action or function scope
- Pre-execution simulation requirements
Execution path
When a request satisfies active policy, it becomes eligible for a connected execution environment on Arc. SENTRY keeps the policy decision surface separate from the execution mechanism so authority remains inspectable.
Reviewable records
Each decision can be associated with execution context such as request identifiers, targets, values, timestamps and resulting status. The goal is straightforward: make autonomous behaviour easier to inspect after the fact.