Input controls
Evaluate incoming content before it becomes model context or reaches a tool.
Platform
Guardrails
Put policy checks in the execution path.
Define reusable controls for inputs and outputs, attach them to the capabilities that need them, and inspect every policy decision with the rest of the request trace.
Policy evaluation
Ready
PoliciesAssignmentsRunsEvents
Live configuration
input.pii_detection passed
output.schema_check passed
output.content_policy passed
output.schema_check passed
output.content_policy passed
DecisionAllow
Policies3 evaluated
EvidenceRecorded
ReusableDefine policies once and assign them deliberately.
In-pathChecks run with the capability execution.
AuditableDecisions are retained with request context.
Platform capabilities
Everything needed to operate this layer.
Build the capability once, keep its operating context visible, and improve it without spreading AI infrastructure across your application.
Output controls
Check generated responses before they are returned to the calling application.
Policy assignments
Apply relevant rules to specific capabilities rather than relying on one global setting.
Grounding checks
Use retrieval evidence and response conditions as part of release and runtime review.
Decision evidence
Record which checks ran, what they observed, and why a response was allowed or blocked.
Evaluation workflow
Connect runtime controls to broader quality evaluation before and after deployment.
Operating workflow
From configuration to production evidence.
The interface follows the work in the order teams actually perform it.
01
Define the policy
Describe the condition and expected action.
02
Assign it
Attach the control to the relevant intent.
03
Test edge cases
Run representative unsafe and ambiguous inputs.
04
Review decisions
Inspect policy outcomes in traces and evaluations.
What improves
A cleaner operating model for production AI.
Move complexity into a visible platform boundary while keeping the product integration direct.
01
Consistent enforcement
Controls live in one governed layer rather than scattered application handlers.
02
Clearer incident review
Policy evidence is available alongside the request that triggered it.
03
Safer change management
Teams can evaluate policy behavior before relying on it in production.
Questions
Before you build.
Practical answers about how this part of LiyaEngine fits into the platform.
Can guardrails run before and after generation?+
Yes. Controls can protect incoming content and evaluate generated output before it returns to the caller.
Are guardrails global?+
Policies can be reusable, while assignments determine which capabilities use each control.
Can I inspect a guardrail decision?+
Yes. Policy outcomes are part of the execution evidence available for review.
Build the capability. Keep the operational context.
Start with a tenant-owned domain and expose the first intent through a stable, governed API.