Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.aperium.apps.hillspire.com/llms.txt

Use this file to discover all available pages before exploring further.

The Policies tab is the working list of every guardrail in your tenant. From here you create new policies, edit existing ones, change a policy’s mode (Monitor / Enforce / Disabled), and search across all of them.
Policies tab showing a header with a Create Policy button, a counter '10 total · 8 enforcing · 2 monitoring', filter controls (Search policies, All Stages, All Types, All Modes), and rows for Jailbreak Detection (Input, Built-in, priority 10, action block, mode Enforce), Permission Validation (Tool, Built-in, priority 10, action block, mode Enforce), PII Detection (Output, Built-in, priority 10, action redact, mode Enforce), Data Leakage Prevention (Output, Built-in, priority 20, action redact, mode Enforce), and Prompt Injection (Input, Built-in, priority 20, action block, mode Enforce).

​
Reading a row

Each row condenses a policy into six attributes:
  • Policy. Display name plus a one-line description and the scope (Global or a specific agent).
  • Stage. When the policy runs: Input, Tool, or Output. See the Overview for what each stage covers.
  • Type. Built-in for the bundled policies that ship with Aperium, or Custom for ones your team authored.
  • Priority. A number that controls evaluation order. Lower runs first. Built-ins use 10, 20, 30. Use the same scale for custom policies so the relative ordering stays predictable.
  • Action. What the policy will do when it matches: block, redact, warn, confirm, or allow.
  • Mode. Enforce, Monitor, or Disabled. The dropdown lets you toggle the mode without leaving the list.
The header counter (10 total · 8 enforcing · 2 monitoring) reflects the same totals as the dashboard.

​
Filtering

Four controls narrow the list:
  • Search policies. Free-text search across name and description.
  • All Stages. Filter to Input, Tool, or Output.
  • All Types. Filter to Built-in or Custom.
  • All Modes. Filter to Enforce, Monitor, or Disabled.
Combine filters freely. A common one is All Modes → Monitor to find every policy still being tuned.

​
Creating a policy

Click + Create Policy at the top right. There are two paths:
  • Start from a template. Recommended for almost every case. The Templates tab lists every starter and lets you pick one. You’ll be able to edit the controls (action, sensitivity, extra patterns) before saving.
  • Build a custom policy. Create one from scratch using the policy editor. You’ll specify the stage, the action, the priority, the scope (global or per-agent), and the matching rules (regex patterns or a condition rule).
Save in Monitor mode by default. Watch the dashboard for several days, then promote.

​
Editing an existing policy

Click a row to open the policy editor. Inside you can change:
  • Display name and description. What admins see.
  • Stage and priority. Rare to change; doing so reorders evaluation.
  • Scope. Global (every agent in the tenant) or restricted to specific agents.
  • Action. What happens on a match.
  • Mode. Enforce, Monitor, or Disabled.
  • Policy-specific controls. For example, a regex pattern list, a sensitivity level dropdown, or a rate-limit threshold. The controls available depend on the policy type. See Templates for the controls each built-in template exposes.
Save and the new configuration takes effect on the next request. Built-in policies can have their config changed but cannot be deleted; you can disable them if you really don’t want them.

​
Promoting from Monitor to Enforce

The most common reason to revisit a policy is to promote it once you’re confident it’s tuned. Two ways:
  • From the Policies tab, switch the Mode dropdown on the row from Monitor to Enforce and save.
  • From the editor, change the mode and save.
Either way the change is logged in the audit log (with [MONITOR] removed from new events going forward) and takes effect immediately.

​
Priority and order of evaluation

Within a stage, policies run in priority order: lower number first. Policies at the same priority run in parallel. If any non-modifying policy returns Block or Confirm, evaluation halts and the rest of the stage is skipped. Modifying actions (Redact, Modify) chain together within a priority level: each modifying policy sees the output of the previous one. This is how PII Detection and Data Leakage Prevention can both redact different patterns in the same response.

​
Built-in policies you’ll see by default

A new tenant ships with these built-in policies pre-installed (some Enforced, some in Monitor for tuning):
StagePolicyDefault actionDefault mode
InputJailbreak DetectionblockEnforce
InputPrompt InjectionblockEnforce
InputContent FilteringblockEnforce
InputRate LimitingblockMonitor
InputInput ValidationblockMonitor
ToolPermission ValidationblockEnforce
ToolDangerous Operation DetectionconfirmEnforce
OutputPII DetectionredactEnforce
OutputData Leakage PreventionredactEnforce
OutputResponse ValidationblockMonitor
You can change the action or mode of any of these from the policy editor. They map one-to-one with entries on the Templates tab if you want to read the full description of what each one does.