> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neuraltrust.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Firewall

> Deploy and size the optional NeuralTrust Firewall with TrustGuard.

Firewall is an optional **data-plane** component (enabled by default). TrustGuard in your cluster calls its in-cluster service for prompt and response classifiers. It is not part of the SaaS control plane.

## Default CPU deployment

```yaml theme={null}
firewall:
  enabled: true
  firewall:
    enabled: true

trustguard:
  firewall:
    enabled: true
```

The chart renders two gateway replicas and one replica for each default worker: `toxicity`, `toolguard`, `prompt-jailbreak`, `prompt-moderation`, and `response-jailbreak`. Gateway and workers use `firewall-cpu:v2.14.0`.

Default worker requests are 1 CPU and 3 GiB, with 2 CPU and 4 GiB limits. `prompt-moderation` overrides memory to 4 GiB requested and 6 GiB limited. Official images bundle their models; `HUGGINGFACE_TOKEN` is optional.

## GPU workers

Chart defaults remain CPU. GPU mode requires a NeuralTrust-provided `firewall-gpu` image plus explicit GPU resources and scheduling. Keep the gateway on the CPU image:

```yaml theme={null}
firewall:
  firewall:
    gateway:
      image:
        repository: registry.example.com/neuraltrust/firewall-cpu
    workerDefaults:
      image:
        repository: registry.example.com/neuraltrust/firewall-gpu
      resources:
        requests:
          cpu: "1"
          memory: 4Gi
          nvidia.com/gpu: "1"
        limits:
          cpu: "2"
          memory: 8Gi
          nvidia.com/gpu: "1"
      nodeSelector:
        accelerator: nvidia
      tolerations:
        - key: nvidia.com/gpu
          operator: Exists
          effect: NoSchedule
      hostIPC: true
    config:
      cudaMpsActiveThreadPercentage: "25"
      cudaMpsPinnedDeviceMemLimit: "6000M"
```

This matches `values-dataplane-gpu.yaml.example`. Install the vendor device plugin and validate node labels first. CUDA MPS and `hostIPC` may require extra security approval, especially on OpenShift.

## TrustGuard integration

TrustGuard derives `NEURAL_TRUST_FIREWALL_BASE_URL` as `http://firewall.<namespace>.svc.cluster.local` and maps `firewall-secrets/JWT_SECRET` to its client secret. If Firewall is disabled, disable both switches:

```yaml theme={null}
firewall:
  enabled: false

trustguard:
  firewall:
    enabled: false
```

## Verify

```bash theme={null}
kubectl get pods -n neuraltrust \
  -l app.kubernetes.io/name=firewall
kubectl get service firewall -n neuraltrust
kubectl logs -n neuraltrust \
  -l app.kubernetes.io/name=firewall,app.kubernetes.io/component=gateway
```

If GPU pods remain pending, inspect resource availability, taints, node labels, and the NVIDIA device plugin with `kubectl describe pod`.

See [Secrets](/neuraltrust/deployment/secrets) for the shared JWT and [Images](/neuraltrust/deployment/images) for chart pins.