Component
core/security/trustyai-guardrails-gateway
TrustyAI's guardrails gateway in a model's path: a service speaking the OpenAI chat-completions API on one route, values.route, that sends every request through the orchestrator wired on its orchestrator socket with a fixed detector applied, overwriting any detectors a caller names. When the detector reports the latest message at or above values.threshold, the gateway answers with an ordinary completion carrying values.fallback_message, and the model is not asked to reply, though its server's /tokenize route still counts the refused text; with values.check_output set, a flagged reply is replaced the same way. It reads no system message, earlier turn or tool result, refuses a request whose latest message is a list of content parts or a tool result, and answers a refused streamed request with a stream carrying the detections and no text. When the orchestrator or its detector cannot be reached, the request is refused with an error. Its verdict is the detector's: a probabilistic classifier misses some attacks, refuses some legitimate requests and is known to be evadable. It authenticates no caller and serves no model listing. The Release supplies the image in infra.image and the catalog names none: TrustyAI's build adds the compat-openssl11 package, whose declared licence includes the OpenSSL licence, and is published only under a floating tag. Nothing renders until the Release sets `infra.image`. Realized on Kubernetes by its Helm Recipe and for local development by its Compose Recipe, each running that image as a single stateless service.
Provenance
- Content digest
- sha256:091ede15c669b068e2f0474f55a3d5097529b29bacb32efc11db83308e84a4f7
- Published
- 2026-11-10T10:13:56Z
- 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
sha256:6f5bb002d081f8b3b5527f6d5f28f3c4d3b5b43c046415095961fd74ea7da560
Reference at this version
oci://hub.aggrega.ai/component/core/security/trustyai-guardrails-gateway:1.0.0Developer parameters
Values the developer sets in the Aggregate.
| Name | Type | Default | Required | Description |
|---|---|---|---|---|
| route | string | guarded | no | Name of the gateway's one route: it serves chat completions at /<route>/v1/chat/completions, and provides.endpoint ends in /<route>/v1, the base URL a client library takes. |
| threshold | number (min 0, max 1) | 0.5 | no | Score at or above which the detector's verdict refuses the request. The orchestrator applies it, and core/security/trustyai-huggingface-detector reports to the orchestrator only labels scored at or above its own 0.5, so with that detector a value below 0.5 has no effect; above 0.5, a higher value lets through a prompt the detector scores below it. |
| fallback_message | string (min 1) | I'm sorry, I can't respond to that. | no | Text of the assistant message the gateway answers a refused request with, in an ordinary completion that also carries the detections. A refused streamed request is answered with the detections and no text. |
| check_output | boolean | false | no | Also checks the model's reply with the same detector and threshold, and replaces a flagged reply with values.fallback_message. The model is called before its reply is checked. |
Operator sizing
Infrastructure sizing the operator tunes per environment.
| Name | Type | Default | Required | Description |
|---|---|---|---|---|
| port | integer (min 1024, max 65535) | 8090 | no | Port the gateway listens on and its Service accepts callers on. Most operators should omit this and let the Recipe assign the default (8090), the product's own. |
| image | string | — | yes | Image the gateway runs, supplied by the Release; the catalog names none. The Recipes run the published build's own binary at /app/bin/vllm-orchestrator-gateway, so the image is TrustyAI's guardrails gateway or a copy of it. That build adds the compat-openssl11 package and is tagged only as latest, so a Release that must keep one build mirrors it and names the copy here. |
| compute | object | — | no | — |
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 gateway. Every caller reaching this name can send requests through the gateway, so it is an address for the component routing model traffic through it 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 OpenAI-compatible API the gateway serves, with the host, the port and the route already composed. A consumer hands it to a client library as its base URL, and sends chat completions to it. |
| protocol | string | — | no | URL scheme to reach `host` and `port` over, for a consumer composing its own address rather than reading `endpoint`. |
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.
orchestrator
- Selector
- security.aggrega.ai/text-detection=orchestrator
- Inject
- —
This socket declares no attachment slots, so a connection through it carries no attachment.
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.
No attachment slots declared.
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
- usesuses socket «orchestrator»
security.aggrega.ai/text-detection=orchestrator
Declares no attachment slots.
Used by
uses socket «auth_core»
uses socket «auth_core»
uses socket «auth_core»
uses socket «auth_core»
uses socket «guard»
uses socket «upstream»
uses socket «guard»
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
implements