Client-based ZTNA installs an agent on each device that connects to a gateway in front of resources, which suits managed devices and gives deep device visibility. Clientless ZTNA gives access through a browser portal, which suits contractors and personal devices but sees less about the device.
Where do these two models come from?
"ZTNA" is an industry label, but NIST SP 800-207 describes the same designs as deployment variations of zero trust architecture. Its device agent and gateway model corresponds to client-based access, and its resource portal model corresponds to clientless access. NIST also lists enclave based deployment and device application sandboxing.
NIST notes that several variations can be in use at once for different business processes in the same enterprise.
How does the client-based model work?
In the agent and gateway model, every enterprise issued device has an installed agent that coordinates connections, and each resource has a gateway directly in front of it, acting as a proxy. When a user requests access, the agent forwards the request to the policy administrator, which asks the policy engine. If approved, the administrator sets up a channel between the agent and the gateway, and encrypted traffic flows. The session ends when the work is done or when a security event triggers it, such as a timeout or a failed reauthentication.
NIST says this model is best for enterprises with a mature device management program and discrete resources that can talk to a gateway, and for organizations that do not want a BYOD policy.
How does the clientless model work?
In the resource portal model, a gateway acts as a portal to resources and no software has to be installed on client devices. NIST says the main benefit is flexibility for BYOD and inter-organizational collaboration, because administrators do not need every device to have an agent.
The costs are also stated plainly. Limited information can be inferred from devices requesting access. The device can only be scanned once it connects to the portal, so it may not be continuously monitored for malware, unpatched vulnerabilities and configuration. Enterprises may have no visibility between sessions. The portal can be discovered and targeted for denial of service, so it must be well provisioned. NIST suggests measures such as browser isolation to compensate.
How do you choose?
Choose by device ownership and application type, not by ideology. Managed laptops accessing internal apps, and engineers who need non-web protocols, point to client-based. A partner who needs one web application for a month points to clientless.
A constructed example: a company has 200 employees on managed laptops and 40 contractors on their own machines. It uses an agent for employees, with device posture checks, and a portal for contractors limited to two web apps, with browser isolation, short sessions and logging. Different trust levels, different mechanisms, one policy engine.
What are the shared requirements?
Both models rely on strong identity and policy. NIST says authentication and authorization of subject and device are discrete functions performed before a session is established. Both need logging, least privilege and a clear inventory of resources. Neither fixes a stolen password if the sign in is phishable, so use phishing resistant authentication for the user in both cases.
How to check or apply this
- List who needs access: employees, contractors, partners, and on which devices.
- List the resources and whether each is web based or uses another protocol.
- Use agent based access where you manage the device and need posture checks.
- Use portal access for unmanaged devices and limit it to specific web applications.
- Add browser isolation or download limits for portal sessions.
- Require phishing resistant sign in for both paths.
- Log session start, denial, and termination, and review them.
- Plan for portal availability and test the fallback.
Quick check
Is clientless ZTNA less secure?
It gives the enterprise less device information, which NIST acknowledges. That can be acceptable if access is narrow, sessions are short and compensating controls are in place.
Can I use both?
Yes. NIST notes that multiple deployment models may be in use for different business processes in one enterprise.
Sources
Last reviewed: 18 October 2026.
