AI Controls

AI Controls govern how Governance Autopilot operates for your organization. All AI features are disabled by default. Enabling them requires a deliberate configuration step, and every AI feature can be disabled independently or all at once.

The master toggle

AI for your organization is either on or off. This is a single organization-level setting, off by default, changeable only by an Owner.

When AI is disabled:

• All Governance Autopilot surfaces (Insights, Designer, Suggestion Inbox) are hidden from the management console
• No nightly discovery agent runs are scheduled
• No inference calls are made
• IEC verdicts are unchanged - your agents' policy evaluation behavior is exactly as it was before AI was enabled

When AI is enabled:

• The Governance Autopilot section becomes visible in the management console
• The nightly discovery agent is scheduled for your organization
• Designer sessions become available to Admin and Owner roles
• The Suggestion Inbox activates

Existing policy rules are not affected by toggling this setting in either direction. Re-enabling after a period of AI being off restores all surfaces without data loss - Insights produced before the toggle are still accessible.

Only an Owner can change this setting.

BYOM - Bring Your Own Model

Formael does not fund inference calls. Each organization provides its own model provider connection. Formael does not have access to your provider credentials - the key is encrypted and used only to route inference calls on your behalf.

To configure your model provider:

1. Navigate to Organization Settings → AI Controls
2. Enter your model provider key in the Model Provider field
3. Select the model tier for each surface (see tier table below)
4. Save

If no model provider key is configured, the Designer will return a configuration error. The Insights feed and Suggestion Inbox will continue to work for previously-generated data.

Model tiers

You can configure which tier to use for each surface independently. The Designer uses the Standard tier by default and can escalate to Advanced on a per-turn basis when you explicitly request deeper analysis.

Per-surface feature gates

Each Governance Autopilot surface can be enabled or disabled independently without affecting the others:

All three must be enabled for the full Governance Autopilot experience. You can roll out surfaces incrementally - for example, enabling Insights for a period before opening the Designer to your team.

Daily cost cap

A daily cost cap limits total Governance Autopilot inference spend per organization per calendar day (UTC). The default cap is $10.00/day. When the cap is reached:

• The Designer displays: "Daily AI budget reached. Try again tomorrow or raise the cap in AI Controls."
• The nightly discovery agent is paused for the remainder of the day
• The Insights feed and Suggestion Inbox continue to work - they read existing data and make no new inference calls

The cap is tracked against your model provider's reported token costs. If cost reporting is unavailable for your provider, the cap is estimated conservatively.

Discovery agent frequency

The nightly discovery agent runs once per day by default. Enterprise organizations can configure the frequency:

Hourly runs are not available. The discovery agent is designed for asynchronous, offline analysis - it is not a real-time monitoring system.

Viewing AI usage

Organization Settings → AI Controls → Usage shows:

• Inference spend per day over the last 30 days, broken down by surface
• Total AI interactions logged, by surface and by day
• Discovery agent run history: start time, duration, cost, number of Insights emitted

Usage data is available to Owner and Admin roles.

API reference

The AI Controls are available via the Management API:

GET  /manage/v1/organizations/:id/ai-controls
PUT  /manage/v1/organizations/:id/ai-controls

The PUT endpoint requires an Owner-scoped JWT. Management tokens cannot modify AI controls.

Example payload:

{
  "aiEnabled": true,
  "policyStudioInsights": true,
  "policyStudioDesigner": true,
  "policyStudioInbox": true,
  "policyStudioMaxCostUsdPerDay": 25.0,
  "byomVirtualKey": "sk-..."
}

Related

• Governance Autopilot - overview of what AI controls enable
• API Authentication - Owner-scoped JWT requirements