Component
core/security/nemoguard-jailbreak-detect
NVIDIA's NemoGuard JailbreakDetect, served by its NIM: a classifier that reads a text and reports whether it is an attempt to make a model ignore its instructions, with a score. It changes and forwards nothing. core/security/nemo-guardrails points its jailbreak-detection rail at it through its jailbreak_detection socket. Running it requires an NVIDIA AI Enterprise licence or NVIDIA Developer Program membership, and an NGC API key: NVIDIA restricts access to the image and the model, and the container fetches its model from NVIDIA with the key when it starts. The Release names the image in infra.image and the catalog names none, so nothing renders until it is set. Realized on Kubernetes by helm/nemoguard-jailbreak-detect, one replica that requests a GPU. It authenticates no caller, so it is published on an in-cluster address only.
Provenance
- Content digest
- sha256:a12fae80984fae74116e1dbbab57e130f867ad80e5a193a1f8ef0cbda482d050
- Published
- 2026-11-10T10:13:49Z
- Signing identity
- https://token.actions.githubusercontent.com/https://github.com/aggrega-ai/aggrega-hub/.github/workflows/catalog-publish-sign.yml@refs/heads/main
Signature
- Officialapplication/vnd.dev.sigstore.bundle.v0.3+json
https://token.actions.githubusercontent.com/https://github.com/aggrega-ai/aggrega-hub/.github/workflows/catalog-publish-sign.yml@refs/heads/main
sha256:b6fea30f86e941b677a03c49e0802ba380f8f06328e982ee5344b21da8fc7842
Reference at this version
oci://hub.aggrega.ai/component/core/security/nemoguard-jailbreak-detect:1.0.0Developer parameters
Values the developer sets in the Aggregate.
No developer parameters declared.
Operator sizing
Infrastructure sizing the operator tunes per environment.
| Name | Type | Default | Required | Description |
|---|---|---|---|---|
| port | integer (min 1024, max 65535) | 8000 | no | Port the classifier listens on and its Service accepts callers on, handed to the container as NIM_HTTP_API_PORT, the variable NVIDIA documents for it. Most operators should omit this and let the Recipe assign the default (8000). |
| image | string | — | yes | Full reference of the jailbreak-detection NIM image, including its tag — the image the operator is entitled to pull. There is no default, because access to NIM images is governed by NVIDIA's terms rather than by the catalog, and the catalog therefore names none. |
| compute | object | — | no | — |
| model_cache | object | — | no | Volume the container fetches its model into and reads it from, kept across restarts so the model is fetched once. NVIDIA lists the model at about 530 MB. |
| gpu | object | — | no | GPU scheduling for the classifier's container. NVIDIA supports the NIM on one GPU and its documented run hands the container the host's GPUs, so the container always requests at least one; a runtime_class selects the GPU-enabled container runtime. Requires a GPU-capable cluster. |
| image_pull_secret | string | — | no | Name of a pre-created Kubernetes image-pull Secret granting access to NVIDIA's registry, nvcr.io, where the NIM image is published. Omit when the cluster can pull the image without credentials, as from a mirror it already reaches. |
Output contract
Values this component publishes to consumers over the uses topology.
| Name | Type | Default | Required | Description |
|---|---|---|---|---|
| host | string | — | yes | In-cluster DNS name of the Service fronting the classifier. Every caller reaching this name can submit text, so it is an address for the component that calls the service rather than one to publish. |
| port | integer | — | yes | TCP port that Service accepts callers on. It follows the port an operator set in the Release, so a consumer reads it rather than assuming a default. |
| endpoint | string | — | yes | Base URL of the classifier's API, with the host, the port and the API path /v1 already composed. A caller posts the text to check to the classify route under it. |
| protocol | string | — | no | URL scheme to reach `host` and `port` over, for a consumer composing its own address rather than reading `endpoint`. |
| metrics_endpoint | string | — | no | The full address Prometheus scrapes this component's metrics at — scheme, host, port and path, `http://<host>:<port>/v1/metrics` on the port callers reach. |
Uses sockets
The dependency sockets this component connects through. The target of every connection wired through a socket must satisfy its label selector, and each connection may carry an attachment in the slots the socket declares. A required slot is owed by each connection the Aggregate or the Release wires through the socket.
No uses sockets declared.
Attachment slots
Named binding points at which a Feature attaches to this component. A slot's selector constrains which Feature may fill it and its inject list names how the Feature's material may be delivered. A required slot must be bound on every instance, by an attachment on the instance or at a scope above it, and a sensitive slot is declared to handle secret material. A Feature may also attach under a name no slot declares.
| Slot | Selector | Required | Inject | Sensitive | Description |
|---|---|---|---|---|---|
| ngc_api_key | feature.aggrega.ai/kind=secret | no | env | yes | Optional NGC API key the container fetches its model with when it starts. When attached, the container receives it as the NGC_API_KEY environment variable, read from a Secret by reference. Unattached, the container starts without it and can serve only a model already in its cache volume. |
Self-inject allowlist
No self-inject pipes declared.
Relationships
Typed edges to and from other published entities. A reference with no published target renders as a broken edge rather than a link.
Uses
This entity uses nothing in the catalog.
Used by
uses socket «auth_core»
uses socket «auth_core»
uses socket «auth_core»
uses socket «auth_core»
uses socket «upstream»
uses socket «forward_target»
uses socket «scrape_target»
uses socket «store»
uses socket «jailbreak_detection»
applies to
applies to
applies to
applies to
applies to
applies to
applies to
implements