Runtime policy enforcement for AI agent actions, applied before each action executes

Skyrelis addresses a gap that opens once AI agents stop answering questions and start taking actions. Identity and access management decides which systems an agent may reach, and prompt-level guardrails filter what a model says, but neither inspects the specific action an agent is about to perform: the payment it is about to submit, the CRM record it is about to overwrite, the email it is about to send. Skyrelis places a policy layer at that point. A lightweight SDK instruments the agent and routes its tool calls, model interactions and data access through a central gateway that evaluates each one before execution.

Policies are written in plain language and can be scoped narrowly, to a particular agent, a particular customer tenant, or a particular class of action, which matters for multi-tenant deployments where one agent serves many customers under different obligations. Each evaluated action is allowed, blocked, or held for human approval, and every decision is risk-scored and logged, producing the evidence trail auditors ask for. The company states that deployment requires no changes to agent code. Skyrelis was founded in 2025 by Jaz Lin, a network security engineer whose background spans seven years at Cisco followed by product and security roles at VMware, Rubrik and RingCentral, and operates from San Mateo, California.