AI gateway integration
Create an AI route in the console (AI gateway). Then change your client’s base URL and add the route key.
from openai import OpenAI
client = OpenAI(
base_url="https://vault.yourbank.example/ai/<route-id>/v1", # was the provider's URL
api_key=os.environ["OPENAI_API_KEY"], # your provider key, passed through
default_headers={
"X-RV-Route-Key": os.environ["ONSHORE_ROUTE_KEY"],
# Optional. Without it, values come back masked. With a token whose access rules
# allow plaintext, and a route that restores, they come back in full.
"X-RV-Caller-Token": caller_token,
},
)
reply = client.chat.completions.create(model="gpt-4o-mini", messages=[{"role": "user", "content": message}])| Header | Meaning |
|---|---|
X-RV-Route-Key | Required. The route’s key |
X-RV-Caller-Token | Optional. A bearer token for the person or service the reply is for; decides whether restored values are real or masked |
X-RV-Cloak-Id (response) | The digital twin’s ID, when the route keeps twins |
What happens
- The prompt is cloaked with the route’s cloak setting before it leaves the country.
- The provider answers.
- If the route restores replies, placeholders and stand-in names are swapped back, in full for callers allowed plaintext, otherwise masked.
Replies still stream: values are restored as the stream comes back. Only models listed on the route are allowed.
Without the gateway
Use the Cloak API to cloak and restore text yourself, for any provider or service.
Last updated on