Deployment
The model binding, contract, effective policy, pack configuration, and policy version in force.
AI Runtime Control Plane · Boundary policy preview
See how Quantlix evaluates a request before the model runs
This no-storage preview uses the same contextual detectors and enforcement-pack rules as the production request path. It shows the action and the exact input that would reach the provider. It does not call a model or create runtime evidence.
No signup · Synthetic input only · Nothing stored · No provider call
Interactive policy lab
Preview one governed request
Choose an outcome, inspect the input, and watch the request cross the runtime boundary.
Production policy modules
Boundary runtime preview
stage: inputmode: no-store
Configure request
Input and enforcement
enterprise-baseline
General-purpose input baseline for credential-shaped secrets, payment instruments, critical identifiers, and personal data.
POST /run
Synthetic only · not stored522 chars left
Automatic abuse protection runs in the background.
Runtime inspector
Decision and request path
Preflight ready
Watch the request cross the boundary
Run the preview to resolve the real policy action and reveal exactly what would reach the provider.
- then
Request ingress
ReadySynthetic input is ready for evaluation.
- then
Contextual policy
SelectedEnterprise baseline — general purpose
- then
Canonical decision
PendingAllowed · Warned · Redacted · Blocked
Provider boundary
GatedContinues only when the selected policy permits it.
This anonymous preview stops before persistence. Production traffic adds the enforcement event, request ID, trace, and eligible audit evidence.
Production request path
How Boundary works in production
The preview above isolates the contextual-policy stage. A real governed request runs through the deployment's full configured control path and creates runtime records.
Step 1
Request received on a supported path
Your application calls Quantlix /run or a supported OpenAI-compatible gateway route.
Step 2
Deployment contract and effective policy loaded
Quantlix resolves the contract, policy version, guardrails, budgets, and configured contextual pack.
Step 3
Input controls evaluated
The configured controls select an explicit action: allowed, warned, redacted, or blocked.
Step 4
Provider called only when permitted
Blocked input stops before inference. Redacted input continues using the transformed payload.
Step 5
Configured output controls run
On supported routes, output policy runs when the deployment is configured for egress enforcement.
Step 6
Runtime evidence recorded
An enforcement event appears in Decision Audit; continued requests also link to their run and trace.
Supported route details matter: /v1/embeddings is input-policy only, and output enforcement applies only where the route and deployment configuration support it. See the production enforcement guide for the current matrix.
One request · four views
Where the same request appears in Quantlix
Boundary is one control in the Quantlix AI Runtime Control Plane—not a separate prompt-filtering product with a different operating model.
Decision Audit
The enforcement event: action, rule or signal, request ID, policy version, and timestamp.
Run detail
For requests that continue: provider, model output, timings, cost, retrieval, and eval context.
Audit bundles
Eligible production records assembled into a manifest-led export with verification instructions.
Run this on your own AI path
A pilot routes scoped traffic through a real deployment so you can inspect actual request IDs, enforcement events, runs, traces, and eligible audit-bundle exports.
Scope: this page previews input-side contextual policy only. It does not run deployment schema validation, budget gates, provider inference, output policy, persistence, trace chaining, or export generation. Production behavior depends on the effective deployment configuration and supported route.
FAQ
Does this create an enforcement event? No. A real governed request creates the event shown in Decision Audit. This anonymous preview is deliberately not evidence.
Are these real pack rules? Yes. The allowlisted pack IDs, detector revision, and decision ladder come from the production policy modules. There is no separate browser regex fallback.
Which pack is the general option? Enterprise baseline is the general-purpose contextual pack. The other choices specialize the input decision for developer, high-risk, education, finance, GDPR-oriented, or healthcare-related traffic.
What happens to redacted input? The transformed payload—not the original sensitive value—is what continues to the provider on a configured production path.