The model answers. It never sees the BVN.
Point your chatbot's OpenAI-compatible client at the gateway. It tokenizes the prompt, sends it on, then swaps the tokens in the reply back to real values before returning it.
Support copilots and chatbots paste customer messages straight into a model hosted abroad, identifiers included.
Change the base URL
Your client keeps working. Only where it points changes.
Tokenize the prompt
Identifiers are replaced before the request leaves Nigeria.
Restore the reply
The customer sees real values. The provider only ever saw tokens.
One line changes
from openai import OpenAI
client = OpenAI(
base_url=f"https://ai.onshore.ng/{ROUTE_ID}/v1", # was api.openai.com
api_key=os.environ["OPENAI_API_KEY"],
default_headers={"X-RV-Route-Key": os.environ["ONSHORE_ROUTE_KEY"]},
)
reply = client.chat.completions.create(
model=MODEL, messages=[{"role": "user", "content": message}],
)OpenAI-compatible
Works with clients and models that speak the same API.
Streaming replies
Tokens are restored as the reply streams back.
Reads free text
Customer messages go through the same detection engine as everything else.
Per-app rules
Choose what each chatbot may send and what it gets back.
A record of each call
See what was caught, without storing the sensitive values in the log.
Your own rules
Custom data types apply here as well.
Start with one leak.
Tell us which flow worries you most. We'll show you what it would take to keep it onshore.