Verification surface
A live verification console makes sensitive actions, policy decisions and their outcomes inspectable rather than implicit.
AI Agent Security · 2026 — Present
Aurels
A control layer for autonomous AI agents. Aurels evaluates sensitive actions before execution, applying policy and semantic risk checks before an agent can perform irreversible operations.
What it is for
As agents move from answering questions to calling tools, the most consequential choice is often whether an action should happen at all. Aurels sits between an agent’s proposal and execution, giving teams a place to make that decision explicit.
It is designed for actions that cannot be casually undone: payment-adjacent operations, procurement workflows, repository changes and other sensitive tool calls.
Aurels execution flow — an explanatory system diagram, not a product screenshot.
Proposed actionAn agent makes a request to use a tool or perform an operation.
Aurels evaluationPolicy rules and semantic risk checks decide whether to issue a time-bounded permit or require review.
What exists now
Execution controls
An API and rules engine combine a semantic risk pre-screen with signed, expiring execution permits.
Traceability
An audit trail records the decision path around an action. GitHub and MCP action-control integrations connect that control to agent workflows.
Engineering discipline
The current product includes 65+ automated tests. The goal is less “autonomous magic” than predictable behavior under pressure.
/ The direction
Control should be available where it changes an outcome, not bolted on as an afterthought.
Aurels is being shaped with teams that need agent workflows to be useful without becoming opaque. The product question is practical: how do you give automation room to move while preserving a meaningful stop button?