Skip to main content

A TrustGuard data plane is the detection engine in your request path. It evaluates prompts, responses and tool calls against your policies and returns findings; your applications or gateway call it synchronously and decide how to enforce. It runs in your cluster, enrols into a control plane, and sends the text its ML detectors score to the inference plane.

TrustGuard is NeuralTrust’s commercial offering and is licensed to enterprise customers. This page installs it on its own — the Detection only reference architecture — enrolled into NeuralTrust SaaS in your tenant’s region. To enrol it into a control plane you run instead, follow the same steps with the change in Connect to your own control plane. To run TrustGate and TrustGuard together, follow Hybrid instead.

​
What you are installing

This install runs trustguard-data-plane (port 8081) and dataagent-trustguard. Its ML detectors run on the NeuralTrust-hosted inference plane, in your control plane’s region; set global.firewall.deploy: true to run the Firewall in your cluster instead.
The TrustGuard image is published to the NeuralTrust container registry. Pulling it requires a registry key, which NeuralTrust issues with an enterprise offer — contact NeuralTrust to get one. See Container images for pulling directly or mirroring into your own registry.

​
Before you start

The Hybrid prerequisites apply here. For TrustGuard alone, that comes down to:
  • Your tenant’s region, EU or US — see Choose your region
  • Where the ML detectors run: the hosted inference plane (default), or your cluster — see Inference plane
  • A cluster with an ingress controller and Helm 3.8+: 2 nodes of 8 vCPU / 16 GiB with the hosted Firewall, or 3 with it in your cluster, as the minimum of an autoscaled node pool — see Capacity
  • A base domain you can point at the cluster ingress, for example platform.example.com
  • Redis and PostgreSQL the cluster can reach — see Check your datastores first
  • The registry key
  • The network rules below
  • Console access in your region, to create a private TrustGuard. Its firewall JWT is shown only once

​
Network for TrustGuard alone

Allow outbound TCP 443 to the hosts of your region. TrustGuard needs no TrustGate config-sync host, and no inbound rule: nothing outside the cluster has to reach it. Static IPs for the EU hosts are in Hybrid → Network. The inference plane can also be reached over Private Service Connect — see Inference plane → Reach it.

​
Install

1

Prepare the namespace and chart sources

Do step 1 of the Hybrid install to create the namespace and the image pull Secret, then get the chart sources so values-trustguard.yaml.example is on disk.
2

Create the TrustGuard in the console

In the console of your tenant’s region, create a private TrustGuard in TrustGuard → Agent Security. Take the CONFIG_SYNC_TOKEN, the DataAgent enrollment JWT and the FIREWALL_JWT from the wizard output — see Console setup. The console shows the FIREWALL_JWT only once.
3

Create the three Secrets

To run the Firewall in your cluster, skip trustguard-firewall and add global.firewall.deploy: true to values-cluster.yaml below.
4

Install

Put the cluster-specific values in their own file:
values-cluster.yaml
For a US tenant, also add saasRegion: "us" under global in that file — see Choose your region.Then install with the TrustGuard slice, which turns on global.products.trustguard and wires the three Secrets:
5

Verify

You should see trustguard-data-plane and dataagent-trustguard, plus firewall and its workers with global.firewall.deploy: true. TrustGuard reports Ready only after it has pulled its configuration, so a pod that is Running but never Ready means config sync is failing — see Hybrid → Verify for the causes.With the hosted inference plane, the TrustGuard data plane logs firewall client loaded from FIREWALL_JWT at startup, with the base_url it calls; check that it names your region’s Firewall host. TrustGuard is published at trustguard.<domain>.
Before handing it over, work through the validation checklist: DNS and TLS from the caller’s network, policy decisions, and telemetry arriving.

​
Connect to your own control plane

To enrol this data plane into a central control plane you run, rather than into NeuralTrust SaaS, make three changes to the steps above:
  1. Point it at your domain. Set global.controlPlane.domain in values-cluster.yaml instead of a region; it moves config sync, DataBridge and telemetry together.
  2. Take the config-sync and enrolment tokens from your console, not from NeuralTrust’s, and mint one enrolment token per data plane — see Telling data planes apart. Your console does not issue firewall keys: ask NeuralTrust for one for this cluster, or run the Firewall in it.
  3. Allow egress to your central cluster instead of to NeuralTrust’s control plane: trustguard-configsync.<domain>, databridge.<domain> and telemetry.<domain> on 443, plus the inference plane host your firewall key names. If the central cluster serves chart-generated certificates, trust its CA as in Trusting a chart-generated control plane.

​
Next steps

Inference plane

Regions, what the Firewall receives, and running it yourself.

TrustGate data plane

Put the gateway in front, in its own data plane.

High availability

Replicas across zones, and the ladder beyond one cluster.

Troubleshooting

Install, pod, config-sync, Firewall, and telemetry failures.