Every AI agent that touches your infrastructure is a new access path. OneIQ keeps AI off production entirely, by giving it access to a governed context cloud - so your security team can say yes without a six-month review.
The problem
Point an AI agent straight at vCenter, a cloud API, or a monitoring platform and it works, right up until it doesn't. Platform credentials are almost never scoped narrowly enough, so the agent gets ungoverned access to everything, not just the one fact it's asking about.
Telling the agent to "stay in scope" in the prompt isn't an enforcement mechanism - it's a request. It can be ignored, worked around, or simply fail silently. And once an agent has reached your infrastructure, there's usually no clean audit trail of what it actually saw or asked for.
Off-production scope limiting. OneIQ doesn't filter the path from AI to production - it removes it. AI only has access to a governed context cloud to ask its questions, never live production systems.
OneIQ doesn't grant AI agents access to your live environment. Instead, it continuously builds a governed context cloud of your infrastructure. The context cloud contains the same telemetry, configuration, and state, held in a separate, read-only layer that exists purely for AI and human queries to run against.
A firewall or VPN still puts the AI on a path to production, just a filtered one. OneIQ doesn't filter that path, it removes it from the reach of AI completely. OneIQ lifts your infrastructure context into a governed context cloud, and every AI activity happens entirely off production.
Every question an agent asks, every query copilot runs, is answered from OneIQ's context cloud, never from the live systems. There's no live connection between the AI conversation and the systems actually running your workloads; so a rogue agent, a bad query, or a compromised model has nothing to reach.
Read-only, governed context cloud
Production infrastructure telemetry flows continuously into OneIQ's context cloud and keeps it up to date. AI agents and copilots query that cloud as much as they want. There's simply no direct path from AI back to production for anything to travel down: rogue, compromised, or otherwise.
Scoping what AI sees
An Agent Zone is a named, customer-approved boundary. A single workload, one cluster, a region of servers, or a migration wave, that scopes an AI agent to exactly the slice of your infrastructure context it needs. Nothing more. The boundary is enforced by OneIQ, in product logic, not by asking an LLM to behave.
Bound to one cluster, fleet, or infrastructure slice, never the whole environment.
An agent can query a zone. It can't change anything inside it.
Every query passes through one governed entry point, logged end-to-end.
Pull a zone's access immediately: no waiting on a maintenance window.
Under the hood
OneIQ holds its own read-only credentials to your infrastructure. The AI agent never receives production access of any kind. It talks to OneIQ over a governed MCP interface, and the platform resolves the question against the Agent Zone it's been granted, nothing further.
For engagements that call for it, Agent Zones can also mask identifying details at the source before anything leaves the zone. For example: stripping hostnames and IPs, so a third party's AI, a partner's benchmarking tool, or an external model can reason over the shape of the data without the identifying details attached.
For your security team
Agents query a read-only context cloud, never live infrastructure.
One set of read-only credentials, not one per agent.
Any MCP-compatible agent, copilot, or LLM. The enforcement lives in OneIQ, not the AI model.
What was asked, what was returned, and which zone was touched, every time.
Security brief
If you need to bring someone else into the approval conversation, this one-page brief covers the deployment model, data handling, and access controls in the language your security team expects.
deployment_model: "infrastructure-to-context engine (MCP)"
endpoint_agents: false
firewall_changes: false
credential_type: "read_only"
vendor_access: false
audit_log: true
zone_revocation: "immediate"
data_retention: "configurable" Live zone, any environment
Pick any environment - virtual, HCI, physical, cloud, or a mix. We'll deploy a governed Agent Zone against it, live.