Runtime policy
Inspect every boundary before unsafe data crosses it.
Apply input and output checks for prompt injection, sensitive data, unsafe content, jailbreak markers, and dangerous code at the gateway or inside a workflow.
One-time starter credit per account on Everstack Cloud. Expires 30 days after signup.
- 01Input and output checks
- 02PII detection
- 03Prompt injection controls
- 04Traceable violations
01Operating model
One path from application intent to controlled execution.
Apply input and output checks for prompt injection, sensitive data, unsafe content, jailbreak markers, and dangerous code at the gateway or inside a workflow.
01
Check inputs before model spend
Inspect freeform user content for sensitive data, prompt injection, unsafe categories, and configured token limits before it reaches a provider.
02
Check outputs before delivery
Apply jailbreak, toxicity, and dangerous code checks after generation but before the response reaches the application or user.
03
Make the policy path explicit
Run guardrails across gateway traffic or place pass and block branches directly inside a workflow graph.
02Capabilities
The controls needed to run it in production.
Everstack keeps guardrails connected to the same identity, policy, deployment, and telemetry boundaries as the rest of the platform.
Sensitive data detection
Detect configured patterns such as payment data, identity numbers, email addresses, phone numbers, and IP addresses.
Prompt injection checks
Block common instruction overrides, role impersonation, and system prompt extraction attempts.
Content filtering
Inspect configured safety categories before input reaches a model and before output reaches a user.
Jailbreak detection
Flag known refusal bypass markers and developer-mode patterns in generated output.
Code execution prevention
Detect dangerous function patterns when generated content must not contain executable system actions.
Violation records
Attach pass or block outcomes to execution context, traces, and workflow logs for audit and tuning.
03Where it fits
Built for real workloads, not isolated demos.
Start with the narrowest useful surface, then connect more of the platform as the workload needs stronger execution, policy, or evidence.
Public AI applications
Apply consistent policy to unpredictable user input and model output.
Sensitive workflows
Stop protected data before it reaches an external model or downstream tool.
Auditable policy enforcement
Keep each intervention connected to the request and workflow that triggered it.
04Questions
What teams usually ask first.
A direct view of how this surface behaves, where it runs, and how it connects to the rest of Everstack.
Where do guardrails run?
They can run across gateway requests through configuration or as explicit input and output nodes inside Workflow Studio.
Are guardrails enabled automatically?
No. Gateway guardrails are explicitly configured so teams can choose the checks, actions, and thresholds that match their application.
Can a violation be audited later?
Yes. Guardrail outcomes are recorded in execution context and appear in traces and workflow execution logs.
Reference
The pages you open once you start building.
Continue through the stack