Use the API in your Gateway installation

Every request needs the Gateway key from your customer delivery in the Authorization field. Keep it in your secret manager and never send it in logs or client-side code.

Authorization: Bearer $MASKERA_GATEWAY_KEY

Send text and receive a masked version. When your policy allows restoration, the response also contains a restoration key that maps placeholders to original values. Gateway does not send the original text or restoration key to Maskera.

curl -s https://maskera-gateway.ert-natverk.internal/v1/mask \
  -H "Authorization: Bearer $MASKERA_GATEWAY_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Min granne Lars bor på Kungsholmen, personnummer 19900101-0000."}'

Response:

{
  "text": "Min granne [NAMN_1] bor på [PLATS_1], personnummer [PERSONNUMMER_1].",
  "map": {
    "[NAMN_1]": "Lars",
    "[PLATS_1]": "Kungsholmen",
    "[PERSONNUMMER_1]": "19900101-0000"
  },
  "entities": { "NAMN": 1, "PLATS": 1, "PERSONNUMMER": 1 },
  "meta": { "ner": true, "duration_ms": 24, "chars": 62, "policy": "default" }
}

textrequired

The text to mask. At least 1 character; the default limit is 100,000 characters and is controlled by your policy.

ner

Set this to false if you only want to find personal identity numbers, phone numbers, email addresses and card numbers. It is faster but does not find names or places, and requires your policy to allow rule-based masking.

return_map

Set this to false if you do not need the restoration key. Your policy can also disable the key, and true cannot override the policy.

The request has the same format as OpenAI. Only change the URL in your existing code. Gateway replaces detected personal data before the AI request and restores it in the response. The AI service key is stored in Gateway's configuration, so your application never needs to handle it.

curl -s https://maskera-gateway.ert-natverk.internal/v1/openai/chat/completions \
  -H "Authorization: Bearer $MASKERA_GATEWAY_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [
      {"role": "user", "content": "Sammanfatta: Lars Svensson ringde om fakturan."}
    ]
  }'

In Python, the official OpenAI SDK is unchanged: only base_url changes.

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["MASKERA_GATEWAY_KEY"],
    base_url="https://maskera-gateway.ert-natverk.internal/v1/openai",
)

The error message does not contain your text. The response uses this format: {"error": {"message": "Description of the error"}}. The status code indicates the type of error that occurred.

When there are too many concurrent requests, Gateway responds with 503 and a Retry-After field. Wait that number of seconds and try again.

GET /healthz responds with 200 when the process is running, and GET /readyz when the license is valid and the model is loaded. Use them in your monitoring system.

{ "status": "ok", "version": "1.2.3" }

The customer portal shows approved releases. Complete guidance for updating, verifying and rolling back is included in the customer delivery's operations guide.

The machine-readable OpenAPI file can generate client code and be used in API tools: /gateway/openapi.json.