AI agents now write code, run tools, and query internal systems in design, research, and engineering environments. An agent usually runs with the access of the person who started it, so it can open that person’s files, reach their systems, and send whatever it reads to any model provider it can reach. Some of these agents run without the security team’s knowledge, using personal API keys.
This datasheet explains how Ensage AI governs those agents. It covers the three components Ensage places in your environment, follows one coding agent through six steps from detection to evidence record, and lists the capabilities at each stage of an agent’s life. It also states plainly what is in scope and what is not.
What this datasheet covers
01
Discover
How zLink identifies AI agents, coding tools, and local models at the process level, including agents your team did not deploy, without needing cooperation from the agent.
02
Authorize
How project assignment, agent policy, and configuration governance decide what each agent can reach. Also how credential substitution keeps real API keys off the endpoint.
03
Contain
How the per-agent sandbox, file-level Data Access Control, the circuit breaker, and the kill switch limit one agent without isolating the machine it runs on.
04
Observe
How the AI Session Controller inspects prompts, responses, and tool calls inline, redacts sensitive data before it leaves your environment, and reports token spend.
05
Maintain
How zCenter keeps the evidence record (agent identity, artifact hash, policy versions, decisions, and session records) under your custody.
"An agent usually runs with the access of the person who started it."
Written for
CISOs and security architects evaluating controls for AI coding agents, command-line agents, and local models
Security teams that need an inventory of the agents already running, including shadow agents using personal API keys
Security and engineering leaders in semiconductor and EDA design, pharmaceutical research and clinical operations, and financial services
Technical evaluators who need components, deployment model, data custody, and coverage scope in one place