Ship AI features without shipping card data.

Point your OpenAI and Anthropic apps at one base URL. Card numbers, IBANs and customer IDs never reach the provider.

disputes/explain.ts

// one change: baseURL points at your gateway
const client = new OpenAI({ baseURL: "https://nymgate.corp.internal/v1", apiKey: process.env.PG_KEY });

const res = await client.responses.create({
  model: "gpt-5",
  input: `Explain why card ⟦CARD:7F31:0001⟧ for ⟦EMAIL:7F31:0002⟧ was declined at merchant ⟦MERCHANT_ID:7F31:0003⟧.`,
});

Live in an afternoon

No SDK. No code changes beyond the base URL.

01 · Point

Change the base URL

Your existing OpenAI or Anthropic client.

02 · Key

Use a gateway key

Real provider keys stay in gateway secrets.

03 · Stream

Keep streaming

Deltas and tool calls are restored in flight.

04 · Ship

Ship it

Card data never reaches the provider.

For engineers

Questions engineers ask.

Does streaming still work?: Yes. Server-sent events are parsed per protocol and tokens are restored even when split across chunks.
What about tool calls?: Arguments are tokenized on the way out and restored as parsed JSON on the way back.
What about embeddings?: Not supported yet. Those requests are rejected, never sent raw.
How much latency does it add?: About 177 ms at p95 for a 2,000-character prompt in our benchmark.

Give your teams AI your security team can sign off.

Book a pilot call Read the security model