Skip to main content
Guardrails are compliance rules enforced by the platform. They protect the organisation from data leaks, runaway costs, and abuse.

​
PII filtering

Redacts personally identifiable information before it reaches the language model.
  1. Navigate to your workspace
  2. Open the Compliance tab
  3. Enable PII filtering
  4. Select the fields to redact: email, phone, SSN, credit card
If PII filtering is enabled at the organisation level, workspaces cannot disable it.

​
Write confirmation

Requires confirmation before executing tools that modify data.
  1. Enable Write confirmation in the Compliance tab
  2. Optionally configure write tool patterns (e.g. update_*, delete_*)
Tools with the is_write flag are automatically subject to this guardrail.

​
Cost caps

Sets a monthly spending limit per workspace.
  1. Set Monthly cost cap in the Compliance tab
  2. Enter the maximum USD amount
Once the cap is reached, queries are rejected until the next billing period.

​
Rate limits

Prevents query abuse by limiting frequency.
SettingDescription
Per user per minuteMaximum queries a single user can send per minute
Per workspace per hourMaximum queries across all users in the workspace per hour

​
Blocked topics

Prevent the AI from responding to specific topics.
  1. Open the Compliance tab
  2. Add topics to the Blocked topics list
The AI will decline to answer queries about blocked topics.

​
Violation tracking

All guardrail violations are recorded and surfaced in:
  • The Compliance tab (violations sub-tab)
  • The workspace dashboard
  • Individual conversation stats
Each violation includes the rule triggered, a description, the conversation, user, and timestamp.