Skip to main content
Unlisted page
This page is unlisted. Search engines will not index it, and only users having a direct link can access it.
Version: 11.3.0

See and Govern the AI Tools Already on Your Workstations

Purpose​

This playbook gives the running order for bringing the AI tools already on your workstations under HYPR AgentPass™. First you see them, then you route them through the gateway, then you make sure they cannot go around it.

Each step is a summary. The detail, with screenshots, field values and caveats, is on the page the step links to.

When you finish:

  • Every workstation AI tool is visible, with its user and device.
  • Its traffic flows through the gateway, where policy applies.
  • Direct provider traffic from that tool is blocked.

Architecture​

Endpoint Guard on each workstation reports the tools it sees, then sends their traffic to the gateway. The gateway records and governs it. For the diagram and the request path, see Architecture: The Request Path.

Prerequisites​

You need the following:

  • Administration console access, with permission to change gateway settings
  • The endpoint client, deployable to your workstations
  • An account with the AI provider your tools use, Anthropic or OpenAI

For the full list, see Prerequisites.

Procedure​

Complete these tasks in order.

Enroll Your Workstations​

Turn on Agent Observability, create an enrollment token, and deploy the endpoint client with it. One token per rollout keeps each deployment separate. Enrolled devices appear in the enrolled-devices list, and start reporting.

See Enabling Agent Observability.

Read What Is Running​

Open Discovered, and click the Discovered Agents tab. Each row is an AI tool seen on a workstation or passing through the gateway, attributed to a user and device where possible. Note the tools, and the providers they call. The rest of this playbook works from that list.

See Discovering Unmanaged Agents.

Review Browser AI Tools (Optional)​

On the Shadow AI Discovery tab of the same view, review the AI SaaS your workforce reaches through the browser, and decide per tool. Approve what your organization sanctions, block what it does not, and ignore what was never shadow AI. This is a separate signal, so the tasks that follow do not depend on it.

See Shadow AI Discovery.

Add the AI Provider Connector in Observe Mode​

Add a connector for each provider you noted, and set its Onboarding Mode to Observe only. The gateway records traffic from unmanaged clients without blocking or onboarding anything, so you can watch before you change behavior.

See Anthropic or OpenAI, and Onboarding Mode.

Turn On Managed Mode and Route the Tools​

Open Desktop Tools, turn on managed mode, and select the tools to manage. Point each tool's inference at your connector. Enrolled devices pick up the configuration on their next poll.

See Enabling Managed Desktop Tools and Routing and Observation.

Block the Bypass​

Turn on Egress enforcement for each routed tool, and decide what a workstation does when the gateway is unreachable. From this point, the tool's direct provider traffic is blocked on the device.

See Egress Enforcement.

Require Approval for New Agents​

Change the connector's Onboarding Mode from Observe only to Require Approval. A tool's next connection is held until its supervisor approves it, and the person at the workstation sees the onboarding prompt.

See Onboarding Mode. For what the person sees, see Onboarding Your Agent.

Validate​

Check the following:

  • On Desktop Tools, the Devices dashboard shows your workstations, with the routed and enforced configuration in the Applied column.
  • In the Request Log, the tool's requests are attributed to the workstation user.
  • A direct call from the tool to the provider fails on the device.

See Managed Devices.

Logs and Audit Trail​

Every governed request is in the Request Log. What the gateway keeps of each request, and for how long, is set under Advanced → Data Management.

See Reading Your Request Log and Data Management.