Add guardrails and a budget

Enable the Prompt Guard

Under SURFACES in the dashboard sidebar, select LLM, then open the surface you created. The canvas already carries a Prompt Guard node on the request path, between the Access Point and the LLM node, shown greyed out until it’s configured. Select it, switch on Enabled, then select Edit request guards… to open its fullscreen editor.

In the fullscreen editor, select Add Prompt Guard and set Type of detection to Regular Expression. Select Built-in Pattern twice, then set the first rule’s pattern to Email Addresses and the second rule’s pattern to Credit Card Numbers. Each new rule starts on the next unused pattern, so pick the pattern from each rule’s list. Set When detected to Redact the text that matches the pattern.

Select the save button (Save changes) under Manage.

Send a request containing PII

Replace <YOUR_APPLIANCE_HOST> and <YOUR_SURFACE_ROUTE> with your values from Create your first LLM Surface:

curl -k -X POST "https://<YOUR_APPLIANCE_HOST><YOUR_SURFACE_ROUTE>/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [
      { "role": "user", "content": "Draft a reply to jane.doe@example.com about her order." }
    ]
  }'

Replace gpt-5.5 with whichever model your surface’s LLM node is actually configured with, if different.

Prompt Guard matches the email address before the request reaches OpenAI and replaces it with a fixed token such as <EMAIL>, so the model never sees the raw value. See Guardrails → and PII protection and NER IDs → for how redaction compares to a NER ID pseudonym.

Add a monthly budget

Select the LLM node on the canvas to open its settings panel, then switch on Enforce a usage limit:

Select the save button (Save changes) under Manage. The limit is enforced immediately: once this surface’s month-to-date spend reaches $5, further requests are blocked with a `429` response before they reach OpenAI. The request that crosses $5 is itself allowed through.