Foresight is delivered as part of the Glasswall Halo CDR engine and is **disabled by default**.

This page covers enabling Foresight on a Glasswall Halo Single Node VM (OVA) deployment.

## Prerequisites

- **Halo version 2.19.0** - a Halo release that includes Foresight. See [v2.19.0 release notes](/halo/v2190).
- **Foresight entitlement** - a Halo license that includes the Foresight entitlement (`Glasswall Halo: Foresight`). If the entitlement is missing or invalid, Halo continues to process files normally and Foresight results are reported as unavailable.
- **System requirements** - To support Foresight, the VM must have **24 CPU cores and 48GB of memory**.



## Deploying Foresight

Foresight ships **disabled by default** on the Single Node VM; the installer sets `foresight.enabled=false`. To enable it, run the `enable_foresight.sh` script included on the appliance:

```bash
/home/glasswall/enable_foresight.sh
```

The script enables the Foresight sidecar on the `cdrplatform-engine` deployment and sizes it for the appliance. No further configuration is required:

- The Foresight image is pre-pulled into the appliance's local image store at build time, so Foresight works on an **air-gapped** appliance with no outbound registry access.
- The sidecar reuses the engine's existing license.

The engine pods then roll to add the Foresight sidecar.

### Enabling Foresight manually

Alternatively, run the Helm upgrade that the script performs:

```bash
helm upgrade cdrplatform-engine ~/cdrplatform-engine -n cdrplatform \
    --reuse-values \
    --set foresight.enabled=true \
    --set foresight.resources.requests.cpu=1.3 \
    --set foresight.resources.requests.memory=3Gi \
    --set foresight.resources.limits.cpu=1.3 \
    --set foresight.resources.limits.memory=3Gi \
    --atomic
```


### Disabling Foresight

To turn Foresight off again, run the same upgrade with `foresight.enabled=false`:

```bash
helm upgrade cdrplatform-engine ~/cdrplatform-engine -n cdrplatform \
    --reuse-values \
    --set foresight.enabled=false \
    --atomic
```

The engine pods roll to remove the Foresight sidecar; standard Glasswall CDR protection is unaffected.

## Resource considerations

The Foresight sidecar performs machine learning inference on the CPU; no GPU is required. On the Single Node VM, `enable_foresight.sh` sizes each sidecar to match the engine container, with requests and limits of **1.3 CPU and 3Gi memory**:

| | CPU | Memory |
|---|---|---|
| Requests | `1.3` | `3Gi` |
| Limits | `1.3` | `3Gi` |

The appliance runs three engine replicas, so allow headroom for three Foresight sidecars. Because the models are bundled into the image, nothing is fetched at runtime.

## Next steps

- [Configuring Foresight in Halo](/halo/configuring-foresight-in-halo)